<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by M.Rasoul Zahmatkesh on Medium]]></title>
        <description><![CDATA[Stories by M.Rasoul Zahmatkesh on Medium]]></description>
        <link>https://medium.com/@M.RasoulZahmatkesh?source=rss-2829acb0d7c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*uiyb-szum06wigeop3q87g.png</url>
            <title>Stories by M.Rasoul Zahmatkesh on Medium</title>
            <link>https://medium.com/@M.RasoulZahmatkesh?source=rss-2829acb0d7c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 17:57:20 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@M.RasoulZahmatkesh/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Role of Robotics in the Future of Offshore Platforms]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/the-role-of-robotics-in-the-future-of-offshore-platforms-873569e6be9e?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/873569e6be9e</guid>
            <category><![CDATA[robotics]]></category>
            <category><![CDATA[future]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[offshore]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Tue, 18 Aug 2026 16:15:45 GMT</pubDate>
            <atom:updated>2026-08-18T16:15:45.409Z</atom:updated>
            <content:encoded><![CDATA[<p>Offshore platforms are one of the most important industrial structures in the oil and gas extraction from marine resources industry. These platforms are generally located far offshore and are exposed to very severe environmental conditions such as storms, big waves, high pressure, high humidity and corrosion from sea water. In these environments, the risks to the workers were always considerable. Thus, the use of modern technologies such as robotics, artificial intelligence and automated systems has become one of the main solutions to enhance safety and efficiency in the offshore industry.</p><p>In recent years robotics have had a major impact on the design, operation and maintenance of offshore platforms. Until now, a large part of the activities related to inspection, maintenance and equipment control were carried out manually by human workers. The operations were expensive and dangerous, with the possibility of falls from height, explosions, exposure to hazardous substances and accidents when working underwater. Now robots can do many of these things more accurately and with less risk.</p><p>One of the key applications of robotics on offshore platforms is for inspection and monitoring of equipment. Marine structures are subject to seawater and severe weather conditions, which lead to corrosion and structural failure. The Remotely Operated Vehicle (ROV) can be used for inspection of the under water parts of platforms , foundations , cables and pipelines . These robots have advanced cameras, a range of sensors and measurement devices, which allow engineers to get accurate information on the state of structures. This technology is used to find potential problems before they become big problems.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uvHODQWVQmyo-fixW95HhA.png" /><figcaption>Offshore in Future</figcaption></figure><p>Drones will also play an important role in the future for offshore platforms. These aerial robots can check the high parts of the platforms, towers, outdoor equipment and the hard-to-reach areas. Drones with high-quality cameras and image-processing technologies can detect cracks, leaks and surface damage. This cuts down the need for workers to work at dangerous heights and makes the inspection process faster and more efficient. <br>Another important use of robotics is for repair and maintenance work. Pumps, compressors, turbines, pipelines and other equipment used on offshore platforms need regular servicing. Some maintenance tasks can be performed by industrial robots and they can work in environments that are difficult or dangerous for humans to be in. More advanced robots will be able to perform complex operations like underwater welding, pipeline repair and component replacement in the future.</p><p>The combination of robotics and artificial intelligence is being used in offshore industries to develop a new generation of intelligent systems. Modern robots are not only remotely controlled, but also analyze information about their environment and make the right decisions with the help of artificial intelligence algorithms. An intelligent system can process data received from a number of sensors and calculate the probability of equipment failure. This method is called predictive maintenance and it lowers the cost of repair and eliminates the risk of a sudden stop in production. <br>Robotics can dramatically increase the efficiency of offshore platforms, improving safety and cutting costs. Manual operations are mostly time-consuming and dependent on weather conditions. But many robots can operate in environments that are difficult for humans. In addition, the constant data collection by robots allows more precise decision making and better resource management.</p><p>In the future, the “smart offshore platform” will become more and more popular. The platforms will be the interaction of robots, smart sensors, automated control systems and advanced communication networks. Such platforms will be able to automate a lot of their processes and thus reduce the need for continuous human presence in hazardous environments. The nature of work will also change from physical labor to system control, data analysis and technology management for workers. <br>While there are many benefits of using robotics offshore, several challenges also exist. Robotic equipment is expensive to purchase and install. The design of systems to operate in the harsh marine environment requires advanced technical knowledge. Cybersecurity is another important issue in automated systems. The more automated systems depend on digital technologies, the greater the danger of cyberattacks. It is therefore important to develop safe, reliable and robust robotic technologies so that they are used extensively.</p><p>Equally important is training the workforce to operate new technologies. Engineers and offshore workers require the skills to operate, manage and maintain robotic systems. The future of this industry will depend on collaboration between humans and machines; robots will do dangerous and repetitive work, whereas humans will concentrate on managing, analyzing and making decisions. <br>Outside the oil and gas industry, robotics is also becoming more important, in particular in the renewable marine energy area. Offshore wind farms and other clean energy installations operate in conditions similar to offshore platforms and require regular inspection and maintenance. Robots deployed in such areas can reduce the cost of operation and speed up the development of offshore renewable energy.</p><p>In conclusion, robotics will be one of the main drivers for the future of offshore platforms. This technology is expected to have a great impact to transform the marine industry through the improvement of safety, the reduction of costs, the increase of operational accuracy and the undertaking of complex activities in hazardous environments. Robots will not replace humans completely, but the human/technological collaboration will bring in a new generation of more efficient, safer and more sustainable offshore platforms. The future of the offshore industry will undoubtedly be closely linked to the development of robotics and artificial intelligence.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=873569e6be9e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Offshore Platforms Work: From Waves and Corrosion to AI]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/how-offshore-platforms-work-from-waves-and-corrosion-to-ai-2ab7f18a53ba?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/2ab7f18a53ba</guid>
            <category><![CDATA[monitoring]]></category>
            <category><![CDATA[offshore]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[energy]]></category>
            <category><![CDATA[oil-and-gas]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Sat, 15 Aug 2026 15:31:22 GMT</pubDate>
            <atom:updated>2026-08-15T15:31:22.473Z</atom:updated>
            <content:encoded><![CDATA[<p>The first thing you usually notice when you see a picture of an offshore platform is its size. The most complex part of an offshore platform is not necessarily visible. The structure is based on structural analysis, hydrodynamics, geotechnical, process, mechanical, electrical, control and safety engineering. All these systems have to work together to ensure the platform is able to operate safely for many years in one of the world’s most challenging engineering environments.</p><p><strong>The Ocean Is Always Loading </strong><br>A building on land might stay relatively stable for most of its life. <br>Offshore platform is another. Waves are continuously loading the structure. There are other forces at work in ocean currents. Wind acts on the upper part of the platform, while currents and waves act on the underwater members. These environmental loads may act at the same time. <br>They estimate hydrodynamic forces using the <strong>Morison Equation</strong>, a common engineering approach for slender offshore members. <br>Mainly there are two components in this equation: drag and inertia. <br>But real offshore analysis is much more complex. <br>Engineers also need to consider the wave height, wave period, wave direction, current velocity, wind speed and the dynamic response of the structure in addition to the design of the structure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OIcVa-LqzJIX7a42Xj1qVA.png" /><figcaption>Offshore</figcaption></figure><p><strong>real waves aren’t nice sine waves. <br></strong>Waves in the real ocean are not like simple mathematical sine waves. <br>In the natural sea states the waves have different frequencies, amplitudes and directions. For this reason engineers often use a <strong>Wave Spectrum</strong> to describe the energy distribution of the sea state. A wave spectrum describes the distribution of wave energy at different frequencies. <br>This is especially true for floating platforms, since a vessel may respond strongly to waves of certain frequencies.</p><p><strong>When Structure Starts Resonating <br></strong>All structures have natural frequencies. Resonance may occur if the frequency of the environmental loading approaches one of the natural frequencies of the structure. When resonance occurs the motion of the structure can be greatly amplified. This can be a serious problem in offshore structures. Large movements can increase the loads on risers, mooring lines, joints, pipelines and other components. <br>Therefore engineers study the mass, the stiffness, the damping and the natural frequencies of the system in an effort to control the dynamic behavior.</p><p><strong>VIV (Vibrations due to water flow):<br></strong>Another important phenomenon is Vortex-Induced Vibration or VIV. <br>When water flows past a cylindrical member vortices may form on alternate sides of the structure. They generate forces that change. <br>The frequency of these forces under certain conditions may be near the natural frequency of the member resulting in large vibration. <br>VIV is especially relevant for Risers and subsea pipelines, as continuous vibration can reduce their fatigue life.</p><p><strong>The soil is a part of the structural system <br></strong>The steel structure is only a part of the load-bearing system of a Jacket platform. Loads from the structure are transmitted to the seabed via the piles and the behavior of the soil has a direct influence on the overall performance of the platform. Therefore, before the construction detailed Geotechnical Investigations are carried out. Engineers have to know different layers of soil, their strength, density, stiffness and behavior under vertical and lateral loads. In simple terms, the design of a Jacket cannot be done properly without knowing what is underneath the seabed under it. <br><strong>Corrosion:</strong> One of Steel’s Greatest Enemies Corrosion is one of the permanent problems in offshore engineering. Seawater is an aggressive environment for steel. <br>A particularly critical area is the Splash Zone, where the structure is repeatedly exposed to seawater, spray and alternating wet and dry conditions. There are several ways of controlling corrosion including protective coatings, Cathodic Protection and Sacrificial Anodes. <br>But no protection system can be expected to work perfectly for ever. <br>That’s why regular inspection and maintenance is essential throughout the life of an offshore structure.</p><p><strong>Fatigue:</strong> <strong>The Failure That Can Take Years</strong> One of the dangerous aspects of fatigue is that it can creep up and go unnoticed for some time. An offshore platform undergoes thousands or millions of loading cycles during its lifetime. Each wave causes another cycle of loading. After many cycles, small cracks can be observed, especially at locations with high stress concentration. Welded tubular joints are one of the most important fatigue sensitive areas. S-N curves are commonly used by engineers to estimate fatigue life. The aim is to ensure that critical members are able to sustain the expected number of load cycles over the design life of the platform.</p><p><strong>How do we inspect underwater structures? <br></strong>How do engineers do an inspection of a structure hundreds of meters <strong>below the surface of the sea? </strong><br>One big answer is the ROV, or Remotely Operated Vehicle. An ROV is a robot controlled from the surface and operates under water. It can be fitted with cameras, sonar, sensors, lights and mechanical arms. ROVs inspect Jackets, pipelines, subsea equipment, mooring systems, risers and other underwater components. In the future, combining these inspections with Artificial Intelligence can make them even more powerful. Computer Vision models can be used to process images collected by ROVs to automatically detect corrosion, cracks, deformation, and other types of damage.</p><p><strong>Digital Twin: A digital replica of the platform </strong><br>Digital Twin is one of the most promising technologies in offshore engineering. A Digital Twin is a digital replication of a physical structure or system. The digital model can be linked to sensor data in real time or at periodic intervals. Take for example a compressor with sensors measuring temperature, pressure, vibration and power use. If the data is slowly drifting away from the normal operating patterns then the system may detect a higher probability of failure . The same logic applies to offshore structures. Sensors can be used to monitor parameters such as vibration, strain, corrosion, deformation, etc. The data collected can then be compared to the digital model to identify abnormal behavior.</p><p><strong>Artificial Intelligence Is Coming to Offshore Engineering </strong><br>Artificial Intelligence can change many things in offshore operations. <br>Predictive Maintenance is one of the most important applications. Traditional maintenance involves equipment being maintained on a fixed schedule. Predictive maintenance is a different story. Rather than wait for equipment to fail or service equipment only based on time, AI models analyze actual operational data and try to predict when a failure is likely to occur. This can help to prevent unexpected shutdowns and cut down on unnecessary maintenance. Artificial intelligence can also be used for image inspection, corrosion detection, crack detection, sensor data analysis, production optimization and anomaly detection.</p><p><strong>Safety over Production</strong> <br>An offshore oil and gas platform can contain large volumes of hydrocarbons. So fire and explosion are the main safety concerns. <br>Fire and Gas Detection Systems are used for detection of flammable, toxic gasses, smoke and fire. An Emergency Shutdown System can shut down critical processes if dangerous conditions are detected. Adjacent equipment may be protected and controlled by firewater and deluge systems. Emergency escape routes Muster Areas, Lifeboats and other evacuation appliances are also designed to protect personnel in the event of a major incident.</p><p><strong>What if a ship runs into the platform? <br></strong>Another important accidental load situation is a ship collision. Even when a large ship is moving slowly, it can have a lot of kinetic energy. Therefore, offshore platforms may be designed to account for potential vessel impacts. The risk assessment can include factors such as vessel size, vessel speed, collision angle, traffic routes and platform location. Some critical areas may also be reinforced to reduce the effects of a possible collision.</p><p><strong>The Future: Intelligent Offshore Platforms </strong><br>Offshore engineering will become more automated in the future. <br>Dangerous or repetitive tasks can be reduced by Digital Twins, Artificial Intelligence, underwater robots and continuous structural monitoring. <br>Floating Offshore Wind is also becoming a significant part of the energy industry. Floating wind turbines can be deployed in deep water, using technologies similar to those developed for offshore oil and gas. All of the Semi-Submersible, Spar and TLP concepts are adaptable to floating wind. <br>Also, new technologies are being developed for offshore hydrogen production using offshore wind electricity to produce hydrogen thru electrolysis. This implies that the future of offshore engineering should not be seen in terms of oil and gas alone. <br>Offshore infrastructure could be incorporated into a much bigger energy system that includes oil and gas, offshore wind, hydrogen, subsea technology, robotics and artificial intelligence. <br>At the end of the day, offshore engineering is not just building a big structure in the ocean. It is about managing forces, risk, energy, data, safety and environment at the same time. <br>It is that combination of challenges that makes offshore engineering one of the most advanced and interesting fields of modern engineering.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2ab7f18a53ba" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Offshore Platforms: A Complete Introduction to Their Types and Applications]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/offshore-platforms-a-complete-introduction-to-their-types-and-applications-301d83d9d9b9?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/301d83d9d9b9</guid>
            <category><![CDATA[oil-and-gas]]></category>
            <category><![CDATA[oil-and-gas-industry]]></category>
            <category><![CDATA[energy]]></category>
            <category><![CDATA[offshore]]></category>
            <category><![CDATA[chemical-engineering]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Wed, 12 Aug 2026 14:52:35 GMT</pubDate>
            <atom:updated>2026-08-12T14:52:35.254Z</atom:updated>
            <content:encoded><![CDATA[<p>Offshore platforms are some of the most impressive achievements of modern engineering. They are built and operated in an environment very different from the land. They are constantly exposed to waves, currents, wind, corrosion, storms and hard working conditions. <br>An offshore oil platform is often thot of as a huge steel structure out in the middle of the sea. But an offshore platform is a different story. <br>It could be a huge steel structure, an oil and gas processing plant, a drilling unit, a power plant, a control center and also a place for working and living for hundreds of people. <br>Offshore projects are complex and require engineers from many different disciplines such as structural engineering, mechanical engineering, petroleum engineering, drilling, electrical engineering, process engineering, control systems, geotechnical engineering, and marine engineering. <br><strong>Why offshore platforms? <br></strong>A large proportion of the world’s oil and gas resources is found under the seabed. <br>As nearshore and land-based resources were developed, the oil and gas industry moved further offshore. Early offshore drilling was mostly done in relatively shallow water where simple fixed structures could be used. <br>Advances in technology allowed engineers to develop new systems that could operate farther from the coast and in much deeper water. <br>This resulted in many different types of offshore platforms, from large fixed steel structures to floating production systems operating in hundreds or even thousands of meters of water. <br>One of the most important criteria when selecting a platform is water depth.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fMoPOsMc7b0wM58OmP-fcQ.png" /><figcaption>Offshore Platform</figcaption></figure><p>As the water gets deeper it becomes harder and more expensive to build a fixed structure all the way down to the sea floor. At greater depths, a floating platform is usually a more practical solution.</p><p><strong>Fixed Platforms:</strong> When the structure is attached to the seabed Fixed platform is one of the oldest and most known types of offshore structures. <br>They have a main structure that is permanently attached to the sea floor. Environmental forces such as waves, wind and currents are transferred thru the structure to the foundation and then to the seabed. <br>A very common example is the Jacket platform. <br>A Jacket is generally a big steel frame made of tubular members. These members are joined together to form a strong and steady structure. <br>When the Jacket has been delivered to the offshore field, large steel piles are driven into the sea floor. These piles connect the structure to the soil and transfer vertical and horizontal loads into the ground. <br>The upper part of the platform is known as the Topsides. <br>Topsides may contain drilling equipment, oil and gas processing equipment, cranes, electrical systems, control rooms, accommodation facilities and many other systems. <br>The major advantage of Jacket platforms is that the technology is mature. They are especially well suited to fields with a long production life. <br>However, as the water depth increases Jackets become much larger and heavier. Transportation and installation are more difficult and costly.</p><p><strong>Gravity-Based Structures:</strong> They Depend On Their Own Weight Another type of fixed offshore structure is the Gravity-Based Structure, or GBS. <br>Instead of relying mainly on piles, a GBS uses its enormous weight to remain stable on the seabed. <br>These are often made of reinforced or prestressed concrete. They have a large mass , resisting horizontal forces and bouyancy . <br>One major advantage of GBS platforms is their high structural capacity. <br>However, their enormous size and weight create significant challenges during construction, transportation, and installation.</p><p><strong>Compliant Towers:</strong> Allowing Controlled Movement at first, it may seem that an offshore structure should be as rigid as possible. <br>But this is not always the best solution. <br>A Compliant Tower uses a different design philosophy. Instead of completely resisting environmental forces with a very rigid structure, it allows a controlled amount of movement. <br>The idea is to let the structure deform within a safe range and absorb part of the energy generated by waves and other environmental loads. <br>Compliant Towers can be used in deeper water than traditional fixed platforms.</p><p><strong>When Water Gets Deeper:</strong> Floating Platforms, as water depth increases, constructing a fixed platform all the way to the seabed becomes extremely challenging. <br>Instead of building a structure that reaches the bottom, engineers can place the production system on a floating structure. <br>This is where Floating Offshore Platforms become important. <br>Floating platforms remain on the water because of buoyancy. Their position is controlled using systems such as Mooring Lines or Tethers. <br>One important type of floating platform is the Semi-Submersible. <br>A Semi-Submersible has submerged floating sections and vertical columns that support the upper deck. <br>Because much of its buoyancy is located below the water surface, the platform can have relatively good stability and reduced wave motion. <br>This makes Semi-Submersibles useful for deepwater drilling and production.</p><p><strong>TLP:</strong> A Platform Held by Tension the Tension Leg Platform, or TLP, is one of the most interesting offshore designs. A TLP has positive buoyancy, hich means it naturally wants to move upward. This upward movement is controlled by vertical tendons, or Tethers, connected to the seabed. <br>The Tethers are kept under tension, which significantly limits the vertical movement of the platform. This characteristic makes TLPs useful for applications where controlled platform motion is important.</p><p><strong>Spar Platforms:</strong> Deep Stability in the Water <br>A Spar Platform is very different in shape from many other floating platforms. Its main body is usually a large vertical cylindrical structure with much of it extending below the water surface. <br>“<strong>It provides a stable system because a lot of the structure is underwater.</strong>” <br>Spar platforms can be used for drilling and production and are particularly useful for deepwater operations.</p><p><strong>FPSO: </strong>A Floating Oil and Gas Facility One of the most important offshore systems is the FPSO, which stands for Floating Production, Storage and Offloading. Unlike many traditional platforms, an FPSO does much more than produce oil and gas. It can receive fluids from subsea wells, process the produced oil and gas, store crude oil, and transfer the oil to another vessel. This is especially helpful in offshore fields far from the coast. <br>Instead of installing a very long pipeline to transport the crude oil to land, it can be stored on the FPSO and transferred to a Shuttle Tanker. <br>This can make FPSO based developments attractive for remote offshore fields. Jack-Up Rigs Platforms for Offshore Drilling The Jack-Up Rig is used mainly for drilling operations. It has a floating hull and a number of long legs. Once the rig is in position, the legs are lowered to the seabed. Then the hull is lifted from the water. Once erected the rig behaves very much like a fixed structure. Thus the Jack-Up rigs are particularly useful for shallow and moderate water depths.</p><p><strong>Drillships:</strong> Deep Water Drilling aDrillship is just a ship with a full drilling system. One of its most unique features is the Moon Pool, an opening in the middle of the ship thru which drilling equipment is lowered toward the seafloor. Drillships can work in very deep water. Many drillships currently are Dynamic Positioning (DP) vessels, meaning they can hold their position. Sensors determine the position of the vessel and a computer-controlled system controls thrusters to maintain the ship near the desired location. The Future of Offshore Platforms The future of offshore engineering is not just oil and gas. Much of the technology used in offshore oil and gas is now being applied to Offshore Wind. Floating wind turbines can be used in deep water locations that are unsuitable for fixed-bottom foundations. Offshore wind turbines can be supported by floating structures like Semi-Submersibles, Spars and TLPs. At the same time, technologies such as Artificial Intelligence, Digital Twins, underwater robotics and autonomous inspection systems are changing the way offshore facilities are operated and maintained. The industry is shifting from simple fixed structures in shallow water to floating platforms, subsea systems, renewable energy, automation and smart offshore infrastructure. <br>There is no single type of platform that is best for all projects. The right choice depends on factors including water depth, seabed conditions, environmental conditions, distance from shore, production capacity, field life, installation requirements, safety and overall project economics. That is what makes offshore engineering so exciting, as every project requires a different combination of technologies and engineering solutions.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=301d83d9d9b9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Weights in Mathematics and Artificial Intelligence: One Word, Two Different Concepts]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/weights-in-mathematics-and-artificial-intelligence-one-word-two-different-concepts-559ef6a75b87?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/559ef6a75b87</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[weight]]></category>
            <category><![CDATA[mathematics]]></category>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Wed, 05 Aug 2026 07:30:51 GMT</pubDate>
            <atom:updated>2026-08-05T07:30:51.927Z</atom:updated>
            <content:encoded><![CDATA[<p>When we first learn about artificial intelligence and neural networks, one of the most common terms we encounter is <strong>“weight.”</strong> Interestingly, the same word also has an important and long-established meaning in mathematics.</p><p>Although both fields use the same term, the idea behind it is different. In mathematics, a weight usually represents <strong>the importance or influence of a value in a calculation</strong>. In artificial intelligence, a weight is a <strong>learnable parameter</strong> that allows a model to discover patterns from data. In this article, we will first explore the concept of weight in mathematics, then examine weights in neural networks, and finally compare these two concepts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UFfQXPPsMGIMN6ZLU3itLw.png" /><figcaption>Weights in Mathematics and Artificial Intelligence</figcaption></figure><p><strong>What Is a Weight in Mathematics?</strong></p><p>In mathematics, a weight is usually a numerical value that determines how much influence a specific quantity has on the final result. Simply put:A weight is a number that shows how important a value is in a calculation.</p><p><strong>Example 1: Weighted Average</strong></p><p>Imagine a student has three exams:</p><ul><li>Exam 1: Score 90 with 20% importance</li><li>Exam 2: Score 80 with 30% importance</li><li>Exam 3: Score 70 with 50% importance</li></ul><p>If we calculate a simple average:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/246/1*Udc9nwVmDx9ZlFkim6VvMg.png" /></figure><p>However, not all exams have the same importance. Therefore, we use weights:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/430/1*R6_4LB8GRYjg7hrIuh1qTw.png" /></figure><p>The result is:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/234/1*2BhEm42X_Xrv5VrxAW0VBQ.png" /></figure><p>Here, the weights determine how much each score affects the final result.</p><p><strong>Example 2: Weight in Mathematical Equations</strong></p><p>Consider this simple equation:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/105/1*y3yCyo-T5NAwFNB7wNvFqg.png" /></figure><p>The number <strong>5</strong> acts as a weight.</p><p>If:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/87/1*XwPFtnR0a5AQ1yhMcARHgw.png" /></figure><p>Then:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/189/1*kNt3osZk7lDj9xDZwF4jhw.png" /></figure><p>The value 5 determines how strongly x affects the output.</p><p>If we change the weight:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/122/1*nPf3bqBmWrzWmf3I0cRU2w.png" /></figure><p>Now the influence of x has doubled. Therefore, in mathematics, a weight is usually a <strong>predefined value</strong> used to control the importance of variables.</p><p><strong>What Is a Weight in Artificial Intelligence?</strong></p><p>In artificial intelligence, especially in neural networks, weights play a much deeper role. A neural network consists of many artificial neurons. Each neuron receives inputs, combines them using different weights, and produces an output. A simplified neuron formula is:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/711/1*2k4smnaQnofYOB_z9XUZZw.png" /></figure><p>Where:</p><ul><li>(x) represents input values.</li><li>(w) represents weights.</li><li>(b) represents bias.</li></ul><p>Weights determine how much influence each input has on the final decision.</p><p><strong>A Simple Example of Weights in a Neural Network</strong></p><p>Imagine we want to build an AI model that detects whether an email is spam. The model receives three features:</p><ul><li>Presence of advertising keywords</li><li>Number of links</li><li>Message length</li></ul><p>Inputs:</p><p>Advertising keywords = 1<br>Number of links = 5<br>Message length = 300</p><p>The neural network assigns weights:</p><p>Advertising keywords → Weight 0.8<br>Links → Weight 0.5<br>Message length → Weight 0.1</p><p>The calculation becomes:</p><p>Result:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/275/1*EKB5tdQDPPmtCAYvAVdzmQ.png" /></figure><p>Based on this value, the model decides whether the email is likely spam.</p><p>But an important question remains:</p><p><strong>Where do these weights come from?</strong></p><p><strong>How Are Weights Learned in Artificial Intelligence?</strong></p><p>Unlike traditional mathematics, weights in neural networks are usually not manually chosen by humans. The model discovers the best weights by analyzing large amounts of data. The general process is:</p><ol><li>The model starts with random weights.</li><li>It receives input data.</li><li>It produces an output.</li><li>The output is compared with the correct answer.</li><li>The error is calculated.</li><li>The weights are slightly adjusted.</li><li>This process is repeated thousands or millions of times.</li></ol><p>This learning process is performed using algorithms such as <strong>Gradient Descent</strong> and <strong>Backpropagation</strong>. In simple terms:</p><p>Training a neural network means finding the best set of weights for solving a specific problem.</p><p><strong>The Main Difference Between Mathematical Weights and AI Weights</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/883/1*vIaVwX1YUbTD15oVU7nUxw.png" /><figcaption><strong>Mathematical Weights and AI Weights</strong></figcaption></figure><p><strong>A Simple Analogy</strong></p><p>Imagine a teacher calculating a student’s final grade. In a mathematical system, the teacher decides beforehand:</p><ul><li>Final exam: 50% weight</li><li>Project: 30% weight</li><li>Attendance: 20% weight</li></ul><p>These weights remain fixed. Now imagine an AI system analyzing thousands of students and discovering which factors actually influence success. After analyzing the data, it learns that:</p><ul><li>Practice has a strong effect.</li><li>Attendance has a smaller effect.</li><li>Projects have a medium effect.</li></ul><p>In this case, the system has learned the weights by itself.</p><p><strong>Is AI Weight the Same as Mathematical Weight?</strong></p><p>The short answer is:</p><p><strong>Yes and no. </strong>From a mathematical perspective, neural network weights are still numbers used as coefficients in equations. A weight in AI is ultimately just a numerical value. However, the major difference is how these values are created and used. In mathematics:</p><p><strong><em>We choose weights to perform a calculation.</em></strong></p><p>In artificial intelligence:</p><p><strong><em>The model learns weights to make predictions and decisions.</em></strong></p><p>Therefore, AI weights are mathematical concepts that have evolved into something more powerful: numbers that are not only used for calculations but are also learned from experience and data.</p><p><strong>Conclusion</strong></p><p>Weight is one of the concepts shared between mathematics and artificial intelligence, but its role is different in each field. In mathematics, weight is a tool for determining the importance of a value in a calculation. In artificial intelligence, weight is the foundation of learning, allowing models to discover complex relationships between data. A simple way to remember the difference is:</p><p><strong>Mathematical weights tell us what is important. AI weights allow machines to learn what is important. </strong>And this small difference is one of the reasons modern artificial intelligence systems are so powerful.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=559ef6a75b87" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Is Mathematics Really Objective? The Hidden Bias Behind Numbers]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/is-mathematics-really-objective-the-hidden-bias-behind-numbers-9fd74f953067?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/9fd74f953067</guid>
            <category><![CDATA[convolutional-neural-net]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[convolutional-network]]></category>
            <category><![CDATA[bias]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Fri, 31 Jul 2026 06:28:35 GMT</pubDate>
            <atom:updated>2026-07-31T06:36:21.928Z</atom:updated>
            <content:encoded><![CDATA[<p>Mathematics has long been regarded as the most objective language humanity has ever created. Unlike literature or politics, mathematical statements are expected to remain true regardless of culture, geography, or personal beliefs. The equation (2 + 2 = 4) does not change depending on who solves it. Because of this apparent neutrality, mathematics is often viewed as completely free from bias.</p><p>However, modern research in artificial intelligence, statistics, education, and philosophy of mathematics suggests that this assumption deserves closer examination. While mathematical principles themselves may be objective, the ways in which mathematics is developed, taught, applied, and interpreted can introduce different forms of bias. These biases influence scientific research, machine learning models, educational systems, and even public policy.</p><p>Understanding bias in mathematics does not mean questioning the validity of mathematical truths. Instead, it requires recognizing that mathematics is practiced by humans, and human decisions inevitably shape mathematical applications.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TWRIr2sVXkYUGouoYaKfyA.png" /><figcaption>Bias in maths</figcaption></figure><h3>What Does Bias Mean in Mathematics?</h3><p>Bias in mathematics does not imply that numbers themselves are prejudiced. Rather, it refers to systematic distortions that arise during mathematical modeling, data collection, statistical analysis, or algorithmic decision-making.</p><p>Every mathematical model is a simplification of reality. To make complex systems manageable, mathematicians and engineers decide which variables to include, which assumptions to make, and which outcomes to optimize. These decisions are rarely value-free.</p><p>For example, when designing a predictive model, selecting one performance metric over another may significantly affect the final outcome. Optimizing for overall accuracy may unintentionally reduce fairness for minority groups. Thus, bias often originates before any calculation begins.</p><h3>Sources of Mathematical Bias</h3><p>Several factors contribute to bias in mathematical systems.</p><p>The first source is <strong>data bias</strong>. Mathematical models rely on data, and if the data contain historical inequalities or underrepresented populations, the resulting models will reproduce those patterns. In machine learning, this problem is widely recognized because algorithms learn statistical relationships directly from available data.</p><p>The second source is <strong>model bias</strong>. Every mathematical model simplifies reality by making assumptions. Linear regression assumes linear relationships, while many probabilistic models assume independence between variables. If these assumptions do not reflect the real world, systematic errors emerge.</p><p>Another important factor is <strong>measurement bias</strong>. Quantities that appear measurable are often proxies for more complex concepts. Educational achievement, economic success, or intelligence cannot always be captured by a single numerical indicator. When inappropriate measurements are chosen, mathematical conclusions become misleading despite being computationally correct.</p><p>Finally, <strong>human bias</strong> affects mathematical practice itself. Researchers decide which problems deserve attention, which datasets to collect, and which objectives should be optimized. These decisions inevitably reflect human priorities.</p><h3>Bias in Artificial Intelligence</h3><p>Perhaps the most visible example of mathematical bias today appears in artificial intelligence.</p><p>Modern AI systems are fundamentally mathematical models trained through optimization techniques. Neural networks minimize loss functions, statistical models estimate probabilities, and reinforcement learning maximizes expected rewards. None of these mathematical procedures are inherently biased. The challenge arises because the optimization objective may not align with broader societal values.</p><p>Consider facial recognition systems. If the training dataset contains significantly more images of certain demographic groups, the mathematical optimization process naturally becomes more accurate for those groups while performing worse for others. The algorithm is simply minimizing error according to the available data, yet the practical outcome can be unfair.</p><p>Similarly, recommendation systems optimize user engagement. Although mathematically efficient, these systems may unintentionally reinforce existing preferences, create information bubbles, or amplify misinformation. Again, the mathematics is functioning correctly; the bias emerges from the objective being optimized. This illustrates an important distinction: mathematical optimization is objective, but choosing what to optimize is a human decision.</p><h3>Educational Bias in Mathematics</h3><p>Bias also appears in mathematics education. Many educational systems present mathematics as a collection of rigid formulas instead of a creative problem-solving discipline. Students who think differently may be unfairly judged because standardized assessments often reward specific solution methods rather than genuine mathematical reasoning. Cultural differences further influence mathematical education. Word problems frequently assume particular social experiences, language structures, or economic backgrounds. As a result, some students face additional cognitive barriers unrelated to mathematical ability.</p><p>Moreover, the historical development of mathematics is often presented through a narrow perspective that emphasizes only a limited group of contributors. In reality, civilizations across Asia, the Middle East, Africa, and the Americas have all made significant mathematical contributions. Recognizing this diversity creates a more accurate understanding of mathematics as a global human achievement.</p><h3>Can Mathematics Ever Be Completely Neutral?</h3><p>This question has generated extensive philosophical debate. Pure mathematics investigates abstract structures independent of practical applications. In this context, mathematical proofs remain objective because their validity depends solely on logical consistency.</p><p>Applied mathematics, however, operates within human contexts. Mathematical models describe economies, healthcare systems, transportation networks, and social behavior. These applications require assumptions, priorities, and interpretations that inevitably introduce subjective elements. Therefore, neutrality exists at different levels. Mathematical logic itself remains objective, while mathematical practice involves human judgment.</p><p>Recognizing this distinction helps avoid two common misconceptions. The first is believing that mathematics is entirely free from bias simply because it uses numbers. The second is assuming that mathematical results are meaningless because humans influence them. Neither extreme accurately reflects reality.</p><h3>Reducing Bias in Mathematical Systems</h3><p>Although bias cannot be eliminated entirely, it can be significantly reduced through careful design. Researchers should prioritize diverse and representative datasets whenever possible. Statistical validation should evaluate not only overall performance but also subgroup performance across different populations. Transparency is equally important. Mathematical assumptions should be explicitly documented so that users understand the limitations of a model. Explainable AI methods are becoming increasingly valuable because they reveal how mathematical decisions are produced rather than treating algorithms as black boxes.</p><p>Interdisciplinary collaboration also plays a crucial role. Mathematicians, computer scientists, ethicists, and domain experts each contribute perspectives that help identify biases invisible from a single discipline.</p><p>Finally, continuous evaluation is essential. Mathematical systems should not be considered complete once deployed. Instead, they should be monitored and updated as new data and societal conditions evolve.</p><h3>Conclusion</h3><p>Mathematics remains one of humanity’s most reliable tools for understanding the world. Its logical foundations are remarkably robust and objective. Nevertheless, mathematical applications are designed, interpreted, and implemented by humans, making them susceptible to various forms of bias.</p><p>Recognizing bias in mathematics is not an attack on mathematical truth. Instead, it is an invitation to apply mathematics more responsibly. As artificial intelligence becomes increasingly integrated into everyday life, understanding the relationship between mathematical objectivity and human subjectivity becomes more important than ever.</p><p>The future of trustworthy AI depends not only on better algorithms but also on a deeper awareness of the assumptions embedded within mathematical models. By acknowledging these assumptions, researchers can build systems that are not only more accurate but also more transparent, equitable, and socially beneficial.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9fd74f953067" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Bias in Artificial Intelligence: Why Every AI Model Has It]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/bias-in-artificial-intelligence-why-every-ai-model-has-it-5193295b2f5d?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/5193295b2f5d</guid>
            <category><![CDATA[bias-in-ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[bias]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Thu, 30 Jul 2026 07:22:34 GMT</pubDate>
            <atom:updated>2026-07-30T07:22:34.389Z</atom:updated>
            <content:encoded><![CDATA[<p>If you had asked AI researchers a decade ago what the biggest challenge in artificial intelligence was, you would probably have heard answers such as limited data, insufficient computing power, or hardware constraints. Today, despite models with hundreds of billions of parameters trained on millions of GPUs, one of the oldest challenges remains unresolved: bias.</p><p>In recent years, the term “<strong>bias</strong>” has become central to discussions about AI ethics, large language models, and automated decision-making systems. Yet it is often misunderstood and reduced to a synonym for discrimination. In reality, bias is far more fundamental than a social issue. It is deeply rooted in the mathematical foundations of machine learning itself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*98GBMZ1BVEQWejhk1EytVw.png" /><figcaption>Bias in AI</figcaption></figure><p>To understand bias, we first need to recognize that no AI model perceives the world exactly as it is. Every model observes only a limited representation of reality through its training data and then attempts to generalize from those observations to unseen examples. This process of generalization forces the model to make assumptions about how the world works. In machine learning, these assumptions are known as “<strong>inductive bias”</strong>. Without them, learning would be impossible because a model would have no basis for extending its knowledge beyond the data it has already seen.</p><p>For this reason, contrary to popular belief, bias was never originally considered a flaw. Long before the rise of deep learning, the concept emerged in statistics and learning theory. Researchers realized that every learning algorithm succeeds because it imposes a particular structure on the data. Linear regression assumes that relationships between variables are linear. Decision trees assume that complex problems can be solved through recursive partitions of the feature space. Convolutional neural networks are built on the assumption that nearby pixels carry meaningful local information. These assumptions are all forms of bias, yet they are precisely what make these algorithms effective.</p><p>As artificial intelligence became more widely adopted, the meaning of bias expanded beyond mathematics. Today, the term also refers to “<strong>algorithmic bias</strong>”, where a system consistently produces different outcomes for particular individuals, demographic groups, or cultures. A hiring algorithm that repeatedly favors male applicants over equally qualified female candidates, or a facial recognition system that performs significantly worse on certain ethnic groups, illustrates this form of bias. Unlike inductive bias, algorithmic bias usually does not originate from the learning algorithm itself. Instead, it is inherited from the data used to train the model.</p><p>Much of the data used to train modern AI systems reflects human behavior. History, culture, economics, and past decisions are all embedded within these datasets. If an organization historically hired people from one demographic more frequently than another, a machine learning model trained on those records will likely learn that pattern as a statistical regularity. From the model’s perspective, there is no distinction between a genuine causal relationship and a historical correlation. It simply reproduces the patterns that appear most consistently in its training data.</p><p>Even when datasets are carefully collected, bias can still emerge. One major source is sampling bias. Many of today’s large-scale datasets are not representative of the global population. English dominates online content, while developed countries contribute a disproportionately large share of publicly available data. As a result, large language models inevitably develop a richer understanding of some languages, cultures, and regions than others. This imbalance is not necessarily intentional, but it has measurable consequences for model performance.</p><p>Another important source of bias lies in the labels assigned during supervised learning. Nearly every supervised machine learning model depends on annotations created by humans. Yet humans frequently disagree. Determining whether a piece of text is offensive, whether an image contains inappropriate content, or whether a customer represents a high credit risk often depends on subjective judgment rather than objective truth. Consequently, models learn human interpretations of reality rather than reality itself.</p><p>Perhaps the most surprising aspect of bias is that it persists even when the data is perfectly balanced and free of obvious errors. The reason lies in optimization. During training, a model is not searching for truth; it is searching for parameters that minimize a loss function. If it discovers that a seemingly irrelevant feature consistently reduces prediction error, it will exploit that feature regardless of whether it represents a genuine causal relationship. This phenomenon, commonly known as “<strong>shortcut learning</strong>”, has become one of the defining characteristics of modern deep learning systems. Computer vision models, for example, have been shown to classify objects based on background textures, lighting conditions, or image quality instead of the objects themselves because these shortcuts happen to minimize training loss more efficiently.</p><p>Large language models are no exception. They are trained on enormous portions of the internet, and the internet is far from an objective representation of reality. It contains misinformation, political polarization, cultural stereotypes, duplicated content, language imbalance, and countless historical biases. Techniques such as Reinforcement Learning from Human Feedback (RLHF), safety alignment, and constitutional training can reduce harmful behaviors, but they cannot eliminate bias entirely. Instead, they introduce a new set of optimization objectives designed to guide the model toward preferred behaviors.</p><p>One of the most common misconceptions about artificial intelligence is the belief that a completely unbiased AI system will eventually be built. From the perspective of statistical learning theory, this is impossible. Every learning algorithm requires assumptions in order to generalize beyond observed examples. Removing all bias would also remove the model’s ability to learn. The real challenge, therefore, is not eliminating bias but understanding what kind of bias a model contains and whether that bias is appropriate for its intended application.</p><p>This shift in perspective explains why modern AI research has moved away from the impossible goal of removing bias altogether. Instead, researchers focus on measuring, interpreting, and controlling it. Fields such as “<strong>Causal AI</strong>”, “<strong>Mechanistic Interpretability”</strong>, and “<strong>AI Alignment</strong>” seek to answer deeper questions about how models arrive at their decisions, which internal representations they develop, and whether those representations reflect genuine causal mechanisms or merely statistical correlations.</p><p>Ultimately, bias should not be viewed simply as a defect in artificial intelligence. It is an inevitable consequence of learning from finite data while attempting to model an extraordinarily complex world through simplified mathematical assumptions. What makes an AI system trustworthy is not the absence of bias, but the ability to identify its sources, measure its effects, and manage it responsibly. As artificial intelligence becomes increasingly integrated into healthcare, finance, education, scientific research, and public decision-making, understanding bias will become just as important as designing larger models or achieving higher benchmark scores. The future of AI will not be defined by systems that claim to be unbiased, but by systems whose biases are transparent, measurable, and aligned with human values.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5193295b2f5d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Dimensions: The Mathematics of Space, Information, and Reality]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/understanding-dimensions-the-mathematics-of-space-information-and-reality-6270b0888d3a?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/6270b0888d3a</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[mathematics]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[linear-algebra]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Sun, 26 Jul 2026 10:16:40 GMT</pubDate>
            <atom:updated>2026-07-26T11:33:44.591Z</atom:updated>
            <content:encoded><![CDATA[<h3>A deeper mathematical view of dimensions beyond geometry</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*A-iFUxOBkJeE1mGBEdS-PA.png" /><figcaption>Tensors and Dimentions</figcaption></figure><p>When mathematicians talk about dimensions, they are not simply describing the number of directions in physical space. Dimension is a fundamental measure of <strong>structure, freedom, and information</strong>.</p><p>From the perspective of modern mathematics, dimension answers a deeper question:</p><p><em>How many independent parameters are required to describe the state of a system?</em></p><p>This question appears everywhere:</p><ul><li>In geometry: describing shapes and spaces</li><li>In linear algebra: measuring vector spaces</li><li>In topology: studying intrinsic properties of objects</li><li>In physics: modeling the universe</li><li>In artificial intelligence: representing high-dimensional data</li></ul><p>Dimension is not merely a property of an object. It is a mathematical description of the complexity of its possible states.</p><h3>1. Dimension as an Intrinsic Property</h3><p>A common misunderstanding is that dimension is determined by how an object looks.</p><p>A sheet of paper appears two-dimensional because it exists inside three-dimensional space. However, its intrinsic structure is different. Imagine a small creature living on the surface of the paper.</p><p>For this creature:</p><ul><li>Forward/backward movement requires one coordinate.</li><li>Left/right movement requires another coordinate.</li></ul><p>No third coordinate is needed. Therefore, from an intrinsic perspective:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/187/1*mmQlWpYrWo6r2awl4oRPTQ.png" /></figure><p>The paper is a two-dimensional space embedded inside a three-dimensional universe.</p><p>This distinction between:</p><ul><li><strong>extrinsic dimension</strong></li><li><strong>intrinsic dimension</strong></li></ul><p>is one of the most important ideas in modern geometry.</p><h3>2. Dimension in Vector Spaces: The Algebraic Definition</h3><p>In linear algebra, dimension is defined through the concept of a basis.</p><p>A vector space VVV has dimension nnn if every vector in that space can be represented uniquely as a combination of nnn independent basis vectors.</p><p>For example:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/518/1*6AUTdF97GBdp4nGi6h4AwQ.png" /></figure><p>where:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/267/1*aQIrFooYvP23dHTHT1iuNQ.png" /></figure><p>is a basis.</p><p>The dimension is:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/167/1*5rG7hO3P7uqOcMSyOPO3NA.png" /></figure><p>This definition is powerful because it does not depend on visualization.</p><p>A space with:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/231/1*qim8kvgCStAagml9NE31ww.png" /></figure><p>is just as mathematically valid as a three-dimensional cube. The limitation is not mathematics. The limitation is human visualization.</p><h3>3. Dimension and Degrees of Freedom</h3><p>A more general interpretation of dimension is the number of independent variables controlling a system. Consider a robotic arm. Its position depends on:</p><ul><li>joint angles,</li><li>rotations,</li><li>movements.</li></ul><p>If a robot has seven independent joints:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/239/1*BBZPdb7JQyGXDyJZRvHOFQ.png" /></figure><p>then its configuration space has:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/164/1*FgeO4yhNWv7MiNKwUtwrEw.png" /></figure><p>The robot does not live in ordinary physical space. It lives in a mathematical space representing all possible states. This idea appears in:</p><ul><li>robotics,</li><li>physics,</li><li>optimization,</li><li>machine learning.</li></ul><h3>4. Manifolds: When Dimensions Become Local</h3><p>Many real-world spaces are not simple Euclidean spaces. Mathematicians use <strong>manifolds</strong> to describe complex spaces that locally behave like ordinary dimensions. A sphere is a classic example. A sphere exists in:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/59/1*7sQd_ckSpZELIZ2zcjYaOg.png" /></figure><p>but its surface requires only two coordinates.</p><p>Therefore:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/177/1*Wz82JMZ7DQ6tWOyn81vZrw.png" /></figure><p>Locally, a small region of the sphere behaves like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/59/1*27YpdCAzLYIbzcsvP2gYpg.png" /></figure><p>but globally it has a curved structure. This idea forms the foundation of:</p><ul><li>general relativity,</li><li>differential geometry,</li><li>modern physics.</li></ul><h3>5. Dimension in Topology: Counting Structure Instead of Coordinates</h3><p>Topology studies properties that remain unchanged under continuous transformations.</p><p>A coffee cup and a donut are topologically equivalent because both contain one hole.</p><p>Their exact shape does not matter.</p><p>Topology asks:</p><p><em>What properties remain when geometry changes?</em></p><p>Different mathematical definitions of dimension exist in topology, such as:</p><ul><li>Lebesgue covering dimension</li><li>topological dimension</li><li>Hausdorff dimension</li></ul><p>These definitions allow mathematicians to study objects that cannot be described using traditional geometry.</p><h3>6. Fractional Dimensions and Fractal Geometry</h3><p>One of the most surprising discoveries in mathematics is that dimensions can be non-integer.</p><p>Classical geometry assumes:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/225/1*2oE9GUqumdIEpZ8qOxUhBg.png" /></figure><p>But fractal geometry extends this idea.</p><p>A fractal may have a dimension such as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/91/1*bQcuAC9HAg160d1ro0dWjg.png" /></figure><p>because it exists somewhere between a line and a surface.</p><p>The Sierpiński triangle has Hausdorff dimension:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/182/1*ZzDJFukZpYvYzkPLXM4m1Q.png" /></figure><p>which gives approximately:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/148/1*a7abMK4DO0K5uFOmDaeGyA.png" /></figure><p>This reveals that dimension is not simply counting directions.</p><p>It measures how complexity scales with size.</p><h3>7. High-Dimensional Spaces and Machine Learning</h3><p>Modern artificial intelligence operates almost entirely in high-dimensional spaces.</p><p>A dataset is represented as:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/141/1*BksNeW6qXQH56dI7DmbbSA.png" /></figure><p>where:</p><ul><li>n = number of samples</li><li>d = number of features</li></ul><p>For example:</p><p>A human face image may contain millions of pixel values.</p><p>Mathematically:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/175/1*a3U_Ten_wLGoYOELhLhqxw.png" /></figure><p>The neural network does not see a face. It sees a point in a high-dimensional feature space. Learning means finding meaningful geometric structures inside this space.</p><h3>8. The Hidden Geometry of Neural Networks</h3><p>Deep learning models transform data through multiple dimensional spaces.</p><p>A neural layer performs:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/154/1*KCBp94F-OhWAFXKgn1cCZg.png" /></figure><p>where:</p><ul><li>x is an input vector</li><li>W is a transformation matrix</li><li>b is a bias vector</li></ul><p>Each layer creates a new representation space. Early layers may capture simple dimensions:</p><ul><li>edges</li><li>colors</li><li>patterns</li></ul><p>Later layers represent abstract dimensions:</p><ul><li>objects</li><li>concepts</li><li>semantic meaning</li></ul><p>The network gradually learns a new geometry of information.</p><h3>9. The Curse of Dimensionality</h3><p>High-dimensional spaces create unexpected mathematical behavior. In low dimensions, nearby points are easy to understand. But as dimension increases:</p><ul><li>volume concentrates in unexpected regions.</li><li>distances become less informative.</li><li>data becomes sparse.</li></ul><p>For example, a million points can densely fill a two-dimensional space, but in a space with thousands of dimensions, the same number of points may be extremely sparse. This is why modern AI requires:</p><ul><li>dimensionality reduction,</li><li>representation learning,</li><li>embedding techniques.</li></ul><h3>10. Dimension as a Measure of Information</h3><p>The deepest interpretation of dimension is not geometric. It is informational. A dimension represents an independent variable that carries information. A high-dimensional object contains many possible states. A low-dimensional representation compresses those states while preserving important structure. This principle explains why techniques such as:</p><ul><li>Principal Component Analysis (PCA)</li><li>Autoencoders</li><li>Neural embeddings</li></ul><p>are so important. They attempt to discover the true dimensions hidden inside data.</p><h3>Conclusion: Dimension Is the Language of Structure</h3><p>Dimension began as a simple geometric concept describing lines, surfaces, and volumes. Today, it has become one of the most powerful mathematical ideas for understanding reality.</p><p>It describes:</p><ul><li>the freedom of physical systems,</li><li>the structure of abstract spaces,</li><li>the complexity of data,</li><li>the geometry of intelligence.</li></ul><p><strong>The modern view of dimension is not:</strong></p><p><em>“How many directions can we see?”</em></p><p><strong>but</strong>:</p><p><em>“How many independent variables define the reality we are studying?”</em></p><p>From the curvature of spacetime to the hidden spaces inside neural networks, dimension remains one of the deepest mathematical tools for understanding the universe.</p><p>From the curvature of spacetime to the hidden spaces inside neural networks, dimension remains one of the deepest mathematical tools for understanding the universe.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6270b0888d3a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Tensors: The Mathematical Language of AI]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/understanding-tensors-the-mathematical-language-of-ai-010122241d1c?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/010122241d1c</guid>
            <category><![CDATA[tensor]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[tensorflow]]></category>
            <category><![CDATA[dimention]]></category>
            <category><![CDATA[computer-vision]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Thu, 23 Jul 2026 07:45:11 GMT</pubDate>
            <atom:updated>2026-07-23T07:45:11.638Z</atom:updated>
            <content:encoded><![CDATA[<p>Artificial Intelligence (AI) has become part of our everyday lives. From recommendation systems and voice assistants to self-driving cars and medical image analysis, AI is changing how we interact with technology. Behind many of these breakthroughs is a mathematical concept that every machine learning practitioner eventually encounters: the <strong>tensor</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DaJYgoLxtaj1H_xjvDCwrg.png" /></figure><p>If you’re just getting started with deep learning, you’ve probably seen the word <em>tensor</em> in frameworks like TensorFlow or PyTorch. At first, it might seem like another intimidating mathematical term. In reality, it’s much simpler than it sounds. A tensor is simply a way of organizing numerical data so computers can process it efficiently. Once you understand tensors, many concepts in machine learning become much easier to follow.</p><h3>So, What Is a Tensor?</h3><p>A tensor is a mathematical structure used to represent data with one or more dimensions. You can think of it as a container that holds numbers in an organized way.</p><p>The easiest way to understand tensors is to start with familiar objects.</p><p>A single number is called a <strong>scalar</strong>, which is a 0-dimensional tensor.</p><pre>5</pre><p>A list of numbers is called a <strong>vector</strong>, or a 1-dimensional tensor.</p><pre>[2, 5, 8]</pre><p>When numbers are arranged in rows and columns, they form a <strong>matrix</strong>, which is a 2-dimensional tensor.</p><pre>[<br>  [1, 2, 3],<br>  [4, 5, 6]<br>]</pre><p>If you stack several matrices together, you get a <strong>3-dimensional tensor</strong>. You can imagine this as a collection of pages in a book, where each page represents one matrix.</p><p>Rather than treating scalars, vectors, and matrices as separate concepts, mathematics views them as different forms of tensors with different numbers of dimensions.</p><h3>Why Do We Need Tensors?</h3><p>Almost every type of information used in AI can be represented as numbers. Images, videos, audio recordings, text, financial data, and sensor measurements all need to be converted into a numerical format before a machine learning model can process them.</p><p>Tensors provide a universal way to store that information.</p><p>Instead of creating different data structures for different types of data, deep learning libraries represent everything as tensors. This makes it possible to use the same mathematical operations regardless of whether the model is processing an image, a sentence, or a time series.</p><p>For example:</p><ul><li>A photograph is stored as a tensor.</li><li>A sentence is converted into a tensor of numerical values.</li><li>Audio signals become tensors.</li><li>Stock market data is represented as tensors over time.</li></ul><p>This consistency is one of the reasons modern deep learning frameworks are so powerful.</p><h3>Understanding Tensor Dimensions</h3><p>Every tensor has a <strong>rank</strong>, which simply refers to the number of dimensions it has.</p><p>A scalar has no dimensions.</p><pre>42</pre><p>Shape:</p><pre>()</pre><p>A vector has one dimension.</p><pre>[4, 8, 15, 16]</pre><p>Shape:</p><pre>(4)</pre><p>A matrix has two dimensions.</p><pre>[<br>  [1, 2, 3],<br>  [4, 5, 6]<br>]</pre><p>Shape:</p><pre>(2, 3)</pre><p>A common example of a 3-dimensional tensor is a color image. Since most digital images use three color channels — red, green, and blue — their shape is typically written as:</p><pre>(Height, Width, Channels)</pre><p>For an image with a resolution of <strong>224 × 224</strong> pixels, the tensor shape would be:</p><pre>(224, 224, 3)</pre><p>As models become more complex, tensors with four or even five dimensions become common, especially when processing multiple samples at once.</p><h3>Tensors in Deep Learning</h3><p>Deep learning models rarely process one example at a time. Instead, they usually work with <strong>batches</strong> of data because modern GPUs can perform many calculations in parallel.</p><p>Imagine training an image classification model using a batch of 32 RGB images. Each image has a size of 224 × 224 pixels and contains three color channels.</p><p>The resulting tensor would have the following shape:</p><pre>(32, 224, 224, 3)</pre><p>Here:</p><ul><li><strong>32</strong> represents the number of images.</li><li><strong>224</strong> is the image height.</li><li><strong>224</strong> is the image width.</li><li><strong>3</strong> represents the RGB color channels.</li></ul><p>This batch-based representation allows deep learning frameworks to train models much faster than processing each image individually.</p><h3>Common Tensor Operations</h3><p>Neural networks work by repeatedly applying mathematical operations to tensors. During training, millions — or even billions — of these operations are performed every second.</p><p>Some of the most frequently used operations include:</p><ul><li>Addition</li><li>Subtraction</li><li>Element-wise multiplication</li><li>Matrix multiplication</li><li>Reshaping</li><li>Transposing</li><li>Concatenation</li><li>Slicing</li><li>Broadcasting</li></ul><p>Every layer in a neural network receives one or more tensors, transforms them mathematically, and produces new tensors that are passed to the next layer.</p><h3>Why Is TensorFlow Called TensorFlow?</h3><p>The name <strong>TensorFlow</strong> reflects exactly how the framework works.</p><p>The word <strong>Tensor</strong> refers to the multidimensional data structures used throughout the model, while <strong>Flow</strong> describes how those tensors move through a computational graph as different mathematical operations are applied.</p><p>Although TensorFlow introduced many developers to the concept, tensors themselves have been used in mathematics and physics for decades. Today, they form the foundation of virtually every major deep learning framework, including PyTorch, TensorFlow, JAX, and MXNet.</p><h3>Creating a Tensor in Python</h3><p>In PyTorch, creating a tensor is straightforward.</p><pre>import torch</pre><pre>x = torch.tensor([<br>    [1, 2],<br>    [3, 4]<br>])</pre><pre>print(x)<br>print(x.shape)</pre><p>Output:</p><pre>tensor([[1, 2],<br>        [3, 4]])</pre><pre>torch.Size([2, 2])</pre><p>TensorFlow offers a very similar interface.</p><pre>import tensorflow as tf</pre><pre>x = tf.constant([<br>    [1, 2],<br>    [3, 4]<br>])</pre><pre>print(x)</pre><p>Although the syntax differs slightly, both libraries use tensors as their primary data structure.</p><h3>Where Are Tensors Used?</h3><p>Once you begin looking for them, you’ll realize tensors appear in nearly every AI application.</p><p>In computer vision, images and videos are represented as tensors. In natural language processing, words and sentences are converted into numerical tensors before being processed by language models. Speech recognition systems transform audio signals into tensors, while medical imaging applications analyze MRI and CT scans stored as high-dimensional tensors.</p><p>Even financial forecasting systems use tensors to represent historical market data, and autonomous vehicles rely on tensor-based sensor information from cameras, LiDAR, and radar to understand their surroundings in real time.</p><h3>Final Thoughts</h3><p>Tensors are one of the most important concepts in artificial intelligence because they provide a consistent way to represent data of almost any kind. Whether you’re working with text, images, audio, or time-series data, the information is ultimately converted into tensors before it reaches a machine learning model.</p><p>The good news is that you don’t need an advanced mathematics background to start using tensors. If you’re comfortable with numbers, lists, and matrices, you’re already familiar with the basic ideas. As you continue learning deep learning, tensors will quickly become a natural part of your workflow. Understanding tensors is one of the first milestones on the journey toward building and understanding modern AI systems.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=010122241d1c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Mathematics to Intelligence: How Fourier Transform Powers Modern AI (p.4)]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/from-mathematics-to-intelligence-how-fourier-transform-powers-modern-ai-p-4-6f1794ffc7b8?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/6f1794ffc7b8</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[mathematics]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[generative-ai-solution]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Tue, 14 Jul 2026 14:52:23 GMT</pubDate>
            <atom:updated>2026-07-14T14:52:23.373Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dc0FqGVMTDockyOx6s53kQ.png" /></figure><p><strong>The Last Part!</strong></p><p>So far, we’ve followed an incredible mathematical journey. Euler showed that complex numbers, rotations, and trigonometric functions are all connected by a single elegant idea. Fourier took that idea even further, proving that even the most complicated signals can be represented as combinations of simple sine and cosine waves.</p><p><strong>But why does any of this matter for artificial intelligence?</strong></p><p>The connection is more direct than it might seem.At its core, AI is a system for discovering patterns. Whether it’s recognizing faces, understanding speech, or predicting financial trends, every machine learning model tries to separate meaningful information from massive amounts of raw data. The problem is that raw data is rarely presented in the most useful form. Images contain noise, audio recordings include unwanted sounds, and sensor data often hides important structures beneath layers of complexity. Looking at data through the lens of frequency can reveal patterns that are difficult — or sometimes impossible — to notice in its original form. That simple shift in perspective has become one of the most valuable ideas in modern AI.</p><h3>Computer Vision: Looking Beyond Pixels</h3><p>Humans recognize objects almost instantly. We naturally identify faces, edges, shadows, and movement without thinking about the millions of visual details involved. Computers don’t have that advantage. For a machine, an image is nothing more than a large grid of numbers. A high-resolution photograph may contain millions of pixel values, but those numbers don’t indicate where an object begins, where it ends, or what it represents. Early computer vision researchers discovered that analyzing every pixel directly wasn’t the most efficient strategy. Instead, they began studying the image in the frequency domain. Low-frequency components describe smooth regions and overall shapes. Medium frequencies capture textures and repeating patterns. High frequencies highlight sharp edges, corners, and fine details. Separating these components makes many tasks easier. Noise can be reduced, edges become more distinct, and important structures stand out more clearly. Modern convolutional neural networks learn many of these features automatically, but frequency analysis is still widely used in image preprocessing, medical imaging, satellite imagery, and scientific visualization. Rather than replacing classical mathematics, AI continues to build on it.</p><h3>Medical Imaging: Seeing Beyond Raw Data</h3><p>Few technologies demonstrate the importance of Fourier mathematics as clearly as magnetic resonance imaging (MRI). Interestingly, an MRI scanner doesn’t capture a photograph of the body. Instead, it collects measurements in what’s known as <strong>k-space</strong>, a frequency-domain representation of the signal. At first, this data looks nothing like a medical image. It appears as an abstract collection of mathematical values. Only after applying the Inverse Fourier Transform does that information become the detailed anatomical image doctors rely on for diagnosis. In other words, long before an AI model begins identifying tumors or segmenting organs, Fourier mathematics has already reconstructed the image itself. Today, many MRI reconstruction systems combine deep learning with Fourier-based algorithms, producing faster scans while maintaining excellent image quality. This is a perfect example of classical mathematics and artificial intelligence working together rather than competing.</p><h3>Speech Recognition: Understanding Human Voices</h3><p>Imagine having a conversation in a busy airport. Despite announcements, background conversations, and engine noise, your brain can usually focus on a single voice. Teaching computers to do the same is far more difficult. Raw audio consists of thousands of numerical samples every second. While neural networks can process those samples directly, doing so is often inefficient. Instead, most speech recognition systems first convert audio into a <strong>spectrogram</strong> — a visual representation showing how frequencies change over time. This representation makes vowels, consonants, harmonics, pauses, and other speech characteristics much easier to distinguish. Whether you’re talking to a virtual assistant, using voice typing, or watching automatically generated subtitles, frequency-domain analysis is almost certainly part of the processing pipeline.</p><h3>Noise Reduction: Finding the Signal</h3><p>Real-world data is rarely perfect. Microphones capture background sounds. Cameras introduce sensor noise. Medical scanners experience interference, and satellite images are affected by atmospheric conditions. Before machine learning models can extract useful information, much of this unwanted variation needs to be removed. Frequency analysis offers one of the most effective ways to accomplish that. Since noise often appears within predictable frequency ranges, filtering algorithms can suppress those frequencies while preserving the information that actually matters. A well-designed preprocessing pipeline can sometimes improve model performance more than simply making the neural network larger. In many situations, better mathematics beats more parameters.</p><h3>Neural Networks Learn Filters</h3><p>One of the most interesting discoveries in deep learning is that convolutional neural networks naturally learn filters that closely resemble those developed in classical signal processing decades earlier. The earliest layers often detect simple patterns such as horizontal edges, vertical lines, corners, and gradients. As information moves deeper through the network, these simple features combine into increasingly complex representations. Edges become textures, textures become shapes, and shapes eventually become recognizable objects. This hierarchy closely reflects the philosophy behind Fourier analysis: complex structures emerge from combinations of simpler components. The mathematics may be implemented differently, but the underlying intuition remains remarkably similar.</p><h3>Diffusion Models and Frequency Information</h3><p>Recent advances in image generation have been driven by diffusion models capable of producing remarkably realistic images. Although these systems rely heavily on deep neural networks, researchers have found that frequency information still plays an important role. High-frequency components capture fine details such as hair, fabric, grass, and skin texture, while low-frequency components define overall composition, lighting, and large-scale structure. Many recent approaches incorporate frequency-aware techniques to preserve sharp details while reducing visual artifacts. Even in one of AI’s newest technologies, ideas developed centuries ago continue to influence state-of-the-art research.</p><h3>Beyond Images and Audio</h3><p>The influence of Fourier analysis extends far beyond computer vision and speech recognition. Scientists use Fourier operators to solve partial differential equations. Climate researchers analyze atmospheric cycles. Astronomers study periodic signals from distant stars. Financial analysts investigate recurring market behavior, while biologists examine repeating patterns within genetic sequences. Even robotic control systems rely on frequency-based methods to achieve stable and precise movement. Across all of these fields, the same principle applies: understanding frequency often reveals structure that is difficult to see otherwise. Artificial intelligence has simply adopted that same mathematical perspective.</p><h3>AI Builds on Classical Mathematics</h3><p>A common misconception is that deep learning somehow replaced traditional mathematics. In reality, the opposite is true. Modern AI is built upon centuries of mathematical discoveries. Linear algebra provides the language of vectors and matrices. Calculus makes optimization possible. Probability and statistics help models reason under uncertainty. Information theory measures information itself, while Fourier analysis reveals hidden frequency structures within data. Neural networks bring these ideas together into a unified computational framework. Artificial intelligence is not a departure from mathematics — it is one of its most recent achievements. Every major breakthrough in AI rests on mathematical foundations established long before the invention of GPUs, massive datasets, or even computers themselves. Among those foundations, the contributions of Euler and Fourier remain some of the most enduring.</p><h3>Conclusion</h3><p>When people think about artificial intelligence, they often picture massive datasets, powerful GPUs, and sophisticated neural networks. What many don’t realize is that these technologies are built on mathematical ideas developed centuries ago. Euler uncovered elegant relationships between complex numbers, rotations, and trigonometric functions, while Fourier showed that even the most complex signals can be understood by breaking them into simple frequency components. Today, those same ideas quietly power technologies we use every day — from medical imaging and speech recognition to computer vision and generative AI.</p><p>The story of AI is not just a story of faster computers or larger models; it’s also a story of mathematics evolving over time. Concepts from linear algebra, calculus, probability, optimization, and Fourier analysis continue to shape the algorithms behind modern machine learning. Understanding these mathematical foundations gives us a deeper appreciation of how AI works and why it has become so powerful. As artificial intelligence continues to advance, one thing is becoming increasingly clear: many of tomorrow’s innovations will still be built on the timeless mathematical discoveries made by pioneers like Euler and Fourier.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6f1794ffc7b8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Fourier’s Revolutionary Idea: Every Signal Has a Hidden Language (p.3)]]></title>
            <link>https://medium.com/@M.RasoulZahmatkesh/fouriers-revolutionary-idea-every-signal-has-a-hidden-language-p-3-fca37bced73f?source=rss-2829acb0d7c------2</link>
            <guid isPermaLink="false">https://medium.com/p/fca37bced73f</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[fourier-transform]]></category>
            <category><![CDATA[euler]]></category>
            <category><![CDATA[signal-processing]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[M.Rasoul Zahmatkesh]]></dc:creator>
            <pubDate>Mon, 13 Jul 2026 14:38:40 GMT</pubDate>
            <atom:updated>2026-07-13T14:38:40.941Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*toyTD1VtahyPNV7ChpFpbQ.png" /></figure><p>When Joseph Fourier introduced his groundbreaking theory in the early nineteenth century, many of his fellow mathematicians were skeptical. The idea seemed almost too ambitious to believe.</p><p>Fourier proposed something extraordinary: <strong>any sufficiently well-behaved signal — no matter how irregular or complex it appears — can be expressed as the sum of simple sine and cosine waves.</strong></p><p>At the time, this sounded almost impossible. Imagine standing on a hill overlooking the skyline of a large city. Every building has a different height, shape, and architectural style. There seems to be no obvious pattern connecting them all. Now imagine someone telling you that this entire skyline could be recreated simply by adding together enough perfectly smooth waves.</p><p>It feels completely counterintuitive. Yet mathematically, that is exactly what Fourier proved. His insight forever changed how scientists and engineers think about data. Instead of asking,</p><p><em>“What does this signal look like over time?”</em></p><p>he encouraged an entirely different perspective:</p><p><strong>“What frequencies are hidden inside this signal?”</strong></p><p>That subtle shift in thinking became one of the most influential conceptual breakthroughs in modern science.</p><h3>Time Domain vs. Frequency Domain</h3><p>Humans naturally experience the world in the <strong>time domain</strong>. When we listen to music, we hear one note after another. When we watch a movie, we see a sequence of changing frames. When examining an audio waveform, we observe how the signal rises and falls over time. Computers can certainly analyze data this way. But many problems become much easier when the exact same information is viewed from another perspective: the <strong>frequency domain</strong>. Think about a piano. Pressing a single key produces one musical frequency. Playing an entire chord produces several frequencies simultaneously. To us, the chord sounds like one harmonious piece of music. To a computer performing frequency analysis, however, that same chord immediately separates into its individual notes. The music itself hasn’t changed. Only the way it is represented has. That is precisely what the Fourier Transform does. Instead of describing a signal according to how it changes over time, it describes the signal according to <strong>how much of each frequency it contains</strong>. Often, this new representation reveals patterns that remain completely hidden in the original signal.</p><h3>Why Frequency Matters</h3><p>Consider a digital photograph. To us, it’s a collection of objects, colors, shadows, and textures. To a computer, however, it’s nothing more than a large grid of numbers. Some neighboring pixels change gradually. Others change abruptly. These variations correspond to different <strong>spatial frequencies</strong>. Large smooth regions contain mostly low-frequency information. Sharp edges, fine textures, and intricate details contain high-frequency information. The Fourier Transform naturally separates these components. Once they’re separated, engineers can manipulate each frequency range independently.</p><p><strong>Want to blur an image?</strong></p><p>Reduce the high-frequency components.</p><p><strong>Want to sharpen it?</strong></p><p>Enhance those high frequencies.</p><p><strong>Want to compress the image?</strong></p><p>Remove frequencies that contribute little to perceived visual quality. Every JPEG image you’ve ever viewed relies on this fundamental principle. Modern image compression is possible because frequency analysis helps computers distinguish meaningful visual information from redundant data. And this idea extends far beyond photography.</p><h3>Understanding Sound Through Frequencies</h3><p>Speech offers another excellent example. Imagine recording someone saying the word:</p><p><em>“Artificial.”</em></p><p>The resulting waveform appears incredibly complex. Thousands of tiny fluctuations occur every second. Simply looking at the waveform makes it difficult to distinguish vowels, consonants, pitch, or pronunciation. But after applying the Fourier Transform, the picture changes completely. Distinct frequency bands suddenly emerge. Some frequencies correspond to resonances within the vocal tract. Others represent vocal pitch. Others capture consonant sounds. Speech recognition systems rely heavily on this transformation. Rather than feeding raw audio directly into a neural network, many modern systems first convert sound into frequency-based representations such as <strong>spectrograms</strong> or <strong>Mel-Frequency Cepstral Coefficients (MFCCs). </strong>These representations preserve the information humans naturally perceive while filtering out unnecessary complexity. As a result, neural networks learn speech patterns much more efficiently. Without Fourier analysis, today’s voice assistants would struggle to reach their current level of accuracy.</p><h3>The Mathematical Bridge Between Euler and Fourier</h3><p>At first glance, Euler’s Formula and the Fourier Transform appear to be two separate mathematical achievements. In reality, they are deeply connected. Fourier originally described signals as combinations of sine and cosine functions. Later mathematicians realized that Euler’s Formula provided a much more elegant way to express exactly the same idea. Instead of writing separate sine and cosine terms, both can be combined into a single complex exponential:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/294/1*RHpYZn9a7toK50UhyFCO1A.png" /></figure><p>This seemingly small substitution transformed Fourier analysis. Calculations became cleaner. Mathematical proofs became shorter. Algorithms became significantly more efficient. Eventually, complex exponentials became the natural language of the modern Fourier Transform. Instead of viewing signals simply as collections of oscillating sine waves, mathematicians began describing them as rotating vectors in the complex plane. Those rotating vectors are exactly what Euler’s Formula describes. In a sense, Euler provided the mathematical vocabulary. Fourier taught the language how to tell stories. Together, they created a mathematical framework capable of describing nearly every signal encountered in science and engineering.</p><h3>Why AI Still Depends on Fourier</h3><p>It’s easy to assume that deep learning has made classical signal processing obsolete. In reality, the opposite is true. Modern AI systems often become more powerful because they incorporate ideas rooted in Fourier analysis. Researchers continue to use frequency-domain techniques to preprocess data, reduce noise, compress information, accelerate computation, and improve the robustness of machine learning models. Rather than replacing Fourier methods, neural networks frequently build upon them. Images entering convolutional neural networks are often enhanced using frequency-based preprocessing. MRI systems reconstruct scans using Fourier mathematics long before a neural network begins interpreting them. Speech recognition models analyze spectrograms instead of raw audio waveforms. Even today’s most advanced generative AI models increasingly exploit frequency-domain representations to improve both efficiency and image quality. The more deeply you study artificial intelligence, the more obvious this relationship becomes. AI is far more than statistics. It is more than optimization. And it is certainly more than neural networks alone. It is the culmination of centuries of mathematical innovation, with Euler and Fourier standing among its most influential architects.</p><p>At this point, one thing becomes clear. Images, sounds, signals — even the most complex datasets — all contain hidden frequency structures. The Fourier Transform gives us the tools to uncover those structures. But discovering hidden frequencies is only the beginning. The next question is even more fascinating:</p><p><strong>How do modern AI systems use this mathematical foundation to recognize faces, understand speech, generate realistic images, and process human language?</strong></p><p>That is where mathematics leaves the classroom and becomes the engine behind real-world artificial intelligence.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fca37bced73f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>