<?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 Victor I. Afolabi on Medium]]></title>
        <description><![CDATA[Stories by Victor I. Afolabi on Medium]]></description>
        <link>https://medium.com/@victor_iyi?source=rss-3219c452aea6------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*_XNbjcKGHuMDESz9CSWg8w@2x.jpeg</url>
            <title>Stories by Victor I. Afolabi on Medium</title>
            <link>https://medium.com/@victor_iyi?source=rss-3219c452aea6------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 17:48:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@victor_iyi/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[Enhancing Human Intelligence: 2024 Update]]></title>
            <link>https://medium.com/@victor_iyi/enhancing-human-intelligence-2024-update-309ab8d0a18e?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/309ab8d0a18e</guid>
            <category><![CDATA[brain-computer-interface]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[human-intelligence]]></category>
            <category><![CDATA[cognitive-augmentation]]></category>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Fri, 26 Apr 2024 14:00:09 GMT</pubDate>
            <atom:updated>2024-04-26T15:07:05.098Z</atom:updated>
            <content:encoded><![CDATA[<p>The <a href="https://medium.com/good-audience/https-medium-com-victor-iyi-enhancing-human-intelligence-episode-2-f15d941024ec">last episode of Enhancing Human Intelligence</a> was published in 2018. It’s high time to revive the project and re-structure it to keep up with the latest developments in AI and neuroscience. Here I would explain some of the re-structure, current state and future of the project.</p><p>Enhancing Human Intelligence (EHI) has been a passion project of mine for more than 8 years. I dedicated my down time to the research, development and planning of this project. Needless to say I haven’t had much downtime since 2018 <em>(coincidentally when </em><a href="https://arxiv.org/abs/1706.03762"><em>The Transformer architecture</em></a><em> was released that went on to disrupt the AI landscape)</em>. EHI is an ambitious project that I estimated in 2017 that the version 0.1.0 will be available sometime 2025 <em>(next year)</em>. I haven’t had time to work on EHI for many years now due to personal and professional reasons. As we all know, research takes time and money and I didn’t have enough of both to continue the project all on my own. However, I’m reviving the project and this time back to the drawing board because time has passed and so has technology.</p><h3>Quick recap on what EHI is all about</h3><p>Artificial intelligence is improving at a staggering rate, however, human intelligence is having trouble catching up. Even the best researchers and engineers at the forefront of this technology struggles to understand the full depth of this technology, and it grows in competence even faster than ever before, while the human intelligence stales at a very slow rate. EHI is supposed to help humanity not get left behind intellectually. Some other emerging benefits include treating patients with mental health crisis or patients with speech impairment and Alzheimers, just to mention a few amongst others . Its purpose is to use artificial intelligence to augment human intelligence and vice versa.</p><blockquote>You might be asking yourself that: “Isn’t this just going to allow artificial intelligence take more control over our lives?”, and you would be right. It is an interesting question which will be answered in my next publication on Enhancing Human Intelligence.</blockquote><p>In <a href="https://medium.com/good-audience/https-medium-com-victor-iyi-enhancing-human-intelligence-episode-2-f15d941024ec">previous episodes,</a> I talked about the <strong>Dual Learning System</strong> which provides a two way communication system between the brain and digital wearable devices. I also went into some details about the components of <em>Enhancing Human Intelligence</em> which includes retrieving information from the brain, transforming human thoughts and a <em>Sage</em> system <em>(the knowledge graph in particular)</em> and the possibility of transmitting the processed result back to the human brain through sensory substitution.</p><h3>Decoding Brain Signals</h3><p>EHI is designed as a fashionable wearable technology that extracts electrical signals from the brain and interprets human thoughts into natural language, processes the natural language and perform actions and commands that was interpreted by this system. It also goes a step further into sending processed data into the human brain.</p><p>Amazing, huh? Well, skeptics might say: “<em>we don’t yet have the technology that can read human thoughts”.</em> However a recent publication by Jerry Tang, Amanda LeBel, Shailee Jain &amp; Alexander G. Huth in a <a href="https://www.nature.com/articles/s41593-023-01304-9">Nature article published on the 1st of May, 2023</a> will suggest otherwise.</p><p>In simple terms, they introduced a non-invasive decoder that reconstructs continuous language using fMRI brain recordings. This decoder generates intelligible word sequences that recover the meaning of perceived speech, imagined speech and even silent videos, demonstrating that a single decoder can be applied to a range of tasks. Check the link below on insightful illustration on scientific reconstruction of language.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F11eF1QqOYow&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D11eF1QqOYow&amp;image=http%3A%2F%2Fi.ytimg.com%2Fvi%2F11eF1QqOYow%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/380233ecaa0cca872f6f64f3b33a0e86/href">https://medium.com/media/380233ecaa0cca872f6f64f3b33a0e86/href</a></iframe><p>I highly suggest learning more about this line of research. If you’re interested in an overview discussion of the paper, you can <a href="https://youtu.be/UnHjhHS9ACo">watch this video</a> discussing the clinical &amp; ethical implications of this line of research. It dives in deeper into the clinical trials, how the system was trained, how the data was collected and the many impact it has in clinical neuroscience. It isn’t a stretch to suggest that this system will only get better and safer over the coming years.</p><h3>Current state of EHI</h3><p>As I noted earlier, EHI has been in the works for more than 8 years. In the world of science and technology, that might as well be forever ago. <a href="https://generativeai.pub/">New methods have been developed</a>, technology has advanced, <a href="https://www.gsam.com/content/gsam/us/en/institutions/market-insights/gsam-connect/2023/artificial-intelligence-disruption.html">artificial intelligence is disrupting many industry</a>. New treatments have also been discovered, <a href="https://www.sciencedaily.com/news/health_medicine/">scientific and technological breakthroughs</a> have been achieved and with all this; I’ve gone back to the drawing board.</p><p>One of the building blocks of EHI is <a href="https://github.com/victor-iyi/sage">Sage</a> — which was described in previous episode as a “<em>digital brain</em>”. Sage is a component in the second half of the <strong>Dual Learning System (DLS)</strong> where the first half is of course the human brain and the second half includes Sage. To recap, The DLS serves as a medium where human intelligence and artificial intelligence intercepts and learn from one another.</p><p>Development of Sage and it’s sub-system has been rather slowed down or completely abandoned due to reasons mentioned earlier. However, I am resurrecting these systems, restructuring them and even adding new sub-system. One of the major reasons for restructuring the system has a lot to do with the current AI bubble and the race to Artificial General Intelligence (AGI). Some of the previous structures isn’t the most efficient ways to go about building a complex system like this given the latest advances in science and technology. I think it is also important to note that there will <em>always</em> be some updates or restructuring of the system as new innovations are discovered and new research comes along. I should be sharing some of the restructuring work in future publications.</p><p>Here are some of the work that’s currently on going:</p><ol><li><strong>Project SAGE (Research &amp; Development)</strong></li></ol><ul><li><em>Re-structuring of Sage</em>: Initial structure was around Knowledge Graphs and using AI tools such as Graph Neural Networks (GNNs), Graph Embeddings, Semantic parser, etc that acts upon user’s decoded thoughts. However, the new structure is centred around <em>Sage AI</em>, it then involves use of knowledge graphs when needed. For example, not all human thoughts is centred around making connections to real world objects and their relationship with one another. However, if a thought comes along that requires this functionalities, it will be available; giving the users more accurate and factual response.</li><li><em>Sage infrastructure and sub-systems</em>: As mentioned earlier, Sage can be thought of as the software stack powering EHI. Therefore requiring various sub-systems like decoding &amp; interpreting brain signals into natural language or actionable tasks, executing and automating tasks, Dual Learning System, and so much more.</li><li><em>Sage AI</em>: I wanna talk about Sage AI so bad, because it’s one of the most exciting stack of the Sage sub-system in my opinion. However, I will be diving deeper into what it is, how it works and the benefit it brings to the entire system in future publications, hopefully the next publication on “Enhancing Human Intelligence”. 😃</li></ul><p>2. <strong>Project &lt;<em>REDACTED</em>&gt; (Research)</strong></p><ul><li>Research on neurofeedback and real time decoding of human thoughts</li><li>Keeping up with research in neuroscience, particularly non-invasive BCI</li><li>Sensory substitution by David Eagleman</li></ul><h3>Future of EHI</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*--noINbBZL426C1jUMvCzg.jpeg" /><figcaption>Image generated by DALL•E</figcaption></figure><p>I am an incredibly curious person who wants to know everything from quantum physics to material science to psychology and many more. However, it’s practically impossible to be an expert at every single one of these fields. However, I can have multiple AI agents that challenge even the best experts in these fields, and having complete access over these agents at my beck and call by just thinking about these topics is a superpower that I’ve always wanted to posses. I think human intelligence does not need to be limited to 10,000 hours of learning just to get good at a given task. Our cognitive abilities can be supercharged by the symbiosis of artificial intelligence and human ingenuity.</p><p>Artificial intelligence is growing at an alarming rate, hence the more reason why we need EHI to become part of our daily lives so that we’re not left behind in this race for intelligence. Enhancing Human Intelligence paints a picture of a future where human cognitive abilities can be significantly augmented through the use of advanced technologies. As we continue to push the boundaries of what’s possible, I remain committed to exploring these advancements responsibly and ethically.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=309ab8d0a18e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Enhancing Human Intelligence]]></title>
            <link>https://medium.com/good-audience/https-medium-com-victor-iyi-enhancing-human-intelligence-episode-2-f15d941024ec?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/f15d941024ec</guid>
            <category><![CDATA[human-intelligence]]></category>
            <category><![CDATA[brain-computer-interface]]></category>
            <category><![CDATA[knowledge-graph]]></category>
            <category><![CDATA[naturallanguageprocessing]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Wed, 04 Sep 2019 21:45:27 GMT</pubDate>
            <atom:updated>2019-09-27T22:44:21.993Z</atom:updated>
            <content:encoded><![CDATA[<h3>Episode Two</h3><p>Today’s technology has really gone advanced, where we can perform some incredible tasks with augmented reality, mixed reality, virtual reality and even controlling digital objects with our minds. It’s no secret that these technologies are going to become an integral part of our daily life in the nearest future. In the <a href="https://blog.goodaudience.com/enhancing-human-intelligence-601733274aa4">last episode of Enhancing Human Intelligence</a>, I discussed extensively about the brain, and how it is possible to get information <em>from and to </em>the brain without any invasive or cochlear implants<em>(i.e., you don’t need to drill your skull in order to get useful information from the brain)</em>. I highly recommend reading the last episode if you haven’t or read it again to refresh your memory. 😊</p><p>Hello world! I’m Victor and I am <em>an Artificial Intelligence and Machine Learning Developer</em><strong><em>(</em></strong><em>&amp;</em><strong><em> </em></strong><em>the Director of AI</em> <a href="https://us.115garage.com/"><em>@115Garage</em></a><em>). </em>In the <a href="https://blog.goodaudience.com/enhancing-human-intelligence-601733274aa4"><em>last episode of Enhancing Human Intelligence</em></a>, I said I will be going into more details about the setup of the proposed non-invasive device, that aims to enhance human intelligence. But after some thoughts, I figured it’ll be much better to discuss: Why we need this? How is it going to enhance our intelligence? What are the various components to make it work? <em>Assuming we have the working device</em>, how and why should you even use it.</p><blockquote><strong>DISCLAIMER</strong>: This content is exclusively licensed to <strong>Victor I. Afolabi. Intellectual Property (Media Ownership) is covered under the </strong><a href="http://www.wipo.int/edocs/mdocs/mdocs/en/cdip_21/cdip_21_inf_2.pdf"><strong>World Intellectual Property Organization (CDIP/21/INF/2, p.13, s.2.3)</strong></a><strong>.</strong></blockquote><blockquote><strong>PS: It is important to note that a number of details has been left out or not disclosed due to obvious reasons.</strong></blockquote><h3>What is Possible Today?</h3><p>Today, we’ve gone really far in seeking information from our brain, tapping into our motor neuron and even transferring motor control from one human to another. Check out the short video below to see what I mean and get a sense of some of the advancements in the neuroscience revolution.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FrSQNi5sAwuc%3Fstart%3D38%26feature%3Doembed%26start%3D38&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DrSQNi5sAwuc&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FrSQNi5sAwuc%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/7a4cb87fd81c357c3c6ebdfd568ac858/href">https://medium.com/media/7a4cb87fd81c357c3c6ebdfd568ac858/href</a></iframe><p>We’ve also been able to control digital devices with our minds. Invasive methods like <a href="https://www.youtube.com/watch?v=kaThzeghWnM"><em>Deep Brain Stimulation (DBS)</em></a><em> or </em><a href="https://en.m.wikipedia.org/wiki/Electrocorticography"><em>Electrocorticography (ECoG)</em></a><em> which is a type of electrophysiological monitoring that uses electrodes placed directly on the exposed surface of the brain to record electrical activity from the cerebral cortex. In layman’s term, they involve cracking open up your skull, to place electrodes that serves as a receiver which can then send useful signals.</em></p><p><em>However, we’ve been able to read similar signals without needing to drill into the skull. In fact, we can have the device into a very fanciful wearable, with the help of </em><a href="https://en.m.wikipedia.org/wiki/Electromyography"><em>Electromyography(EMG) – which is primarily for evaluating and recording the electrical activity produced by skeletal muscles. Check out the video below to get more context on how this technology works.</em></a></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FO0IjaXuTUz0%3Fstart%3D69%26feature%3Doembed%26start%3D69&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DO0IjaXuTUz0&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FO0IjaXuTUz0%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/067eb0aa6a5592f2d69cdab8c506dc6e/href">https://medium.com/media/067eb0aa6a5592f2d69cdab8c506dc6e/href</a></iframe><h3>Interfacing with devices through silent speech</h3><p>My point is that we’ve been able to retrieve information from the brain without non-invasive approaches or drilling into your brain. <a href="https://www.media.mit.edu/videos/fi-alterego-2018-04-04/">Just imagine we were able to retrieve a human’s inner thoughts</a> — <em>(best to watch the video as well)</em>. We can perform series of operations with this information like, search the web, perform actions, control digital devices, get answers to any questions on the world wide web. What exactly can we do with this information? What are the limitations, how do we communicate the resulting information back to the user?</p><h3>Introducing Sage Knowledge Graph</h3><p>Before stating the stages and components involved in giving human access to unlimited intelligence, it is worth explaining one of the major knots in the system — Sage. Before going into what <em>“Sage”</em> is all about, let’s talk about its name; <a href="https://en.wikipedia.org/">Wikipedia</a> describes<em> “</em><strong><em>Sage”</em></strong><em> has someone who has attained wisdom.</em> Sage in this context is <em>an open-source </em><a href="https://thenextweb.com/podium/2019/06/11/what-is-a-knowledge-graph-and-how-does-one-work/"><em>Knowledge Graph</em></a><em>,</em> but it’s much more than just a <a href="https://whatis.techtarget.com/definition/graph-database"><em>Graph Database</em></a> that deals with linked-data alone. It comprises of extensive functionalities that might be described as a “<em>digital brain”</em>. Describing all the functionalities, components and what makes it revolutionary would not fit into this mini section of its introduction. Therefore, I’ll soon write out a publication that goes into details and explains what sage really is, and why it’s such a revolutionary invention.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1HL2Ox52zeUxh0ok70sHbQ.jpeg" /><figcaption>A revolutionary idea!</figcaption></figure><p>However, I’ll give a bit of context on Sage’s components. It comprises of every other functionality any Knowledge Graph possesses, but not quite. It could be described as an <em>“Artificially Intelligent Knowledge Engine”</em>. It is powered by <em>state-of-the-art</em> Natural Language systems plus some revolutionary ideas that makes it stands out from every other Knowledge Graph. <em>(A bit on its features in the upcoming section).</em></p><h4>Stages &amp; Components of Enhancing Human Intelligence</h4><p>In the previous publication of Enhancing Human Intelligence, I gave a brief overview of the scope on how I intend to make this happen and given various facts about the brain and the proposed device to create a <em>two-way</em> communication between the brain &amp; the digital world. (<em>Please refer to that article if you haven’t already or it might be helpful to go over it again, to refresh your memory)</em>. Various components that goes into development of a device that enhances human intelligence are:</p><ol><li><strong>Retrieving Relevant Information from the Brain: </strong>Information is being received from the brain when prompted <em>(or learns when it’s being invoked),</em> in form of <em>“human thoughts”</em> and converts it into a (predefined discrete or continuous) <em>“representation”</em>.</li><li><strong>Transforming Human Thoughts:</strong> Once the information is retrieved from the brain and converted into an <em>“intermediate representation”</em>. These <em>“representations”</em> are being transformed into a <em>“programmable data structure” </em>–that can serve as a valid data format for any program <em>(or software)</em>. An AI system is trained to learn and understand patterns associated with the <em>data</em>. This might involve filtering out some noisy data that are considered <em>“irrelevant”</em>.</li><li><strong>The Sage Knowledge Graph:</strong> A light-weight embedded version of the sage knowledge graph will be deployed on <em>this device</em>, in which the transformed data is being read by “sage” at this stage. Sage processes these data and converts them into a <em>“sage internal representation”</em> which includes performing <em>state-of-the-art</em> natural language tasks (like semantic parsing, entity recognition, and many more). It also builds and hierarchical structure in order to understand what the original thoughts are about <em>(questions, actions, thoughts, discussions, answers, communications, etc)</em>.</li><li><strong>More Processing with the Sage Engine: </strong>Sage then uses its revolutionary Query API to answer questions about entity or entities of interest or related operation. <em>For instance</em>, if a question is being asked like: <em>“Who’s the last president of the United States?”</em>. Sage understands that the entity of interest here is “<a href="https://www.wikidata.org/wiki/Q76">Barack Obama”</a>, and it returns all related data that exists on the Sage’s graph database which might include <em>(height:</em><strong><em> </em></strong><em>1.85m, nationality</em><strong><em>: </em></strong><em>American, education:</em><strong><em> </em></strong><a href="https://www.google.com/search?q=Harvard+Law+School&amp;stick=H4sIAAAAAAAAAOPgE-LUz9U3MMrNyi1SAjMNTarKi7Wks5Ot9AtS8wtyUoFUUXF-nlVqSmlyYklmft4iViGPxKKyxKIUBZ_EcoXg5Iz8_BwAt4kmokoAAAA&amp;sa=X&amp;ved=2ahUKEwi__KaTirjkAhUfSBUIHRCvAd0QmxMoATA4egQIDhAR"><em>Harvard Law School</em></a><em> (1988–1991), </em>parents<em>:</em><strong><em> </em></strong><a href="https://www.google.com/search?q=Ann+Dunham&amp;stick=H4sIAAAAAAAAAOPgE-LUz9U3MMrNyi1S4gIxjTPSKgrjtSSzk630C1LzC3JSgVRRcX6eVUFiUWpeSfEiVi7HvDwFl9K8jMRcACPDovVBAAAA&amp;sa=X&amp;ved=2ahUKEwi__KaTirjkAhUfSBUIHRCvAd0QmxMoATA5egQIDhAW"><em>Ann Dunham</em></a><em>, </em><a href="https://www.google.com/search?q=Barack+Obama+Sr.&amp;stick=H4sIAAAAAAAAAOPgE-LUz9U3MMrNyi1S4gIxjQuTkyuytCSzk630C1LzC3JSgVRRcX6eVUFiUWpeSfEiVgGnxKLE5GwF_6TE3ESF4CI9AEZna9NHAAAA&amp;sa=X&amp;ved=2ahUKEwi__KaTirjkAhUfSBUIHRCvAd0QmxMoAjA5egQIDhAX"><em>Barack Obama Sr.</em></a><em>, children:</em><strong><em> </em></strong><a href="https://www.google.com/search?q=Malia+Obama&amp;stick=H4sIAAAAAAAAAOPgE-LUz9U3MMrNyi1S4gIx0zPMUjIMtaSyk630C1LzC3JSgVRRcX6eVXJGZk5KUWreIlZu38SczEQF_6TE3EQAKC33DkMAAAA&amp;sa=X&amp;ved=2ahUKEwi__KaTirjkAhUfSBUIHRCvAd0QmxMoATA6egQIDhAb"><em>Malia Ann Obama</em></a><em>, </em><a href="https://www.google.com/search?q=Sasha+Obama&amp;stick=H4sIAAAAAAAAAOPgE-LUz9U3MMrNyi1S4gIz8wrTK8u1pLKTrfQLUvMLclKBVFFxfp5VckZmTkpRat4iVu7gxOKMRAX_pMTcRACcuT_xQwAAAA&amp;sa=X&amp;ved=2ahUKEwi__KaTirjkAhUfSBUIHRCvAd0QmxMoAjA6egQIDhAc"><em>Sasha Obama</em></a><em>, and many more) –in a structured format</em>.</li><li><strong>Decoding Results from Sage:</strong> Information that are considered <em>“less useful”</em> to the current questions or actions are being filtered out in a very efficient manner. The resulting information gotten back from the sage engine is then being converted into another similar <em>“representation”</em> that can be understood and sent back to the brain, as if the human brain already had all these information at their disposal <em>(or already stored in their brain)</em>.</li><li><strong>Dual-Learning System:</strong> Putting it all these together, I think the above stages explains what I meant when I introduced the “<strong><em>Dual-Learning System”</em></strong> in the <a href="https://blog.goodaudience.com/enhancing-human-intelligence-601733274aa4">last episode of Enhancing Human Intelligence</a>.</li></ol><p>All these complicated operations happen literally within milliseconds, and the user can use the device however they see fit.<strong><em> All these operations are happening in real-time.</em></strong></p><blockquote><strong>RECALL: It is important to note that a number of details has been left out or not disclosed due to obvious reasons.</strong></blockquote><blockquote>The training timeline is still under active thoughts, but the goal is to start performing these tasks almost immediately the device is purchased (because of the revolutionary Dual-Learning System &amp; the Sage Knowledge Graph engine).</blockquote><h3>Challenges</h3><ul><li><strong>Time:</strong> The Sage Project is still in its very early stage of development. As with any revolutionary system, it takes a while to move from start to finish. The project started on the white board 9th of January, 2018. The planning and architecture went on for about 21 months. The first prototype written in Cython <em>(a C binding for Python)</em> proved out not to be as efficient and effective as I wanted it to be. So I went back to the drawing board and decided to write the entire system in a <em>safe</em> Systems Programming Language — <a href="https://rust-lang.org/">Rust</a>. The first code base commit was made on 6th of October, 2019. The project estimation is estimated to be near completion by 2025, <em>(6 years)</em>.</li><li><strong>Security:</strong> There has been some security concerns when I talked to my friends about this. Would our brain not be hacked? Can’t false information be sent? What if I don’t want it to access my thoughts? The answer to the last part, it only accesses your thoughts when you want it to, Just like you say “Hey Google”, or “Siri”, before they start listening to you. As for the other questions I feel this post is getting too long and I’ll definitely explain some security concerns in the next episode of Enhancing Human Intelligence.</li><li><strong>Fashionable:</strong> I’ve always said I want the device to be indistinguishable from normal eye glasses or other wearables, <em>(i.e. not looking too techie)</em>. This becomes a challenge as many components of the system has to be cramped into the little handle of both sides of the eye glasses. Although, I’m pretty positive that this won’t be an issue in a few years as the field of Nanotechnology is growing at a staggering rate! In addition, it doesn’t have to be an eye glasses, although I thought it’d be cool to have an eye glass. <em>One</em> — because they are directly mounted to your brain. <em>Two</em> — the lenses can be used as a display which could come in handy for complete automation without involving any mobile device or computers.</li></ul><h4>Conclusion</h4><p>This revolutionary device not only gives you access to any information ever existed on the world wide web within milliseconds, it also monitors your health and serve as your personal digital assistant. You need no initial training for weeks before you start to use the base functionalities, although the functionality gets better with time as you use it more and more.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7gJ6mZd5Ea3O85PAuyLqJA.jpeg" /></figure><p>And I’ve always been asked a question that goes like this: <em>“Victor, you said this is supposed to enhance our intelligence, but what if one day we didn’t have it on? What happens then?”</em>. Well, the simple answer is that the more you see a movie; the more you know what’s being said, the more you read a book; the more you can quote some exact line without even looking. The same principle happens to this, because you’ve been given access to information every now and then, there comes a time when you wouldn’t even need it to know certain things because all this information are eventually going to be stored in your <a href="https://www.verywellmind.com/what-is-the-hippocampus-2795231">hippocampus</a> <em>(the part of your brain responsible for imagination &amp; memory)</em>.</p><p>Even though it might take a little longer to get there, once we’re there, we’d have access to unlimited information we need at an instant. We no longer need to be forced to learn what doesn’t interest us. The moment we have access to the world’s information <em>(without having to reach out or search for it)</em>, the possibilities of application are endless.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f15d941024ec" width="1" height="1" alt=""><hr><p><a href="https://medium.com/good-audience/https-medium-com-victor-iyi-enhancing-human-intelligence-episode-2-f15d941024ec">Enhancing Human Intelligence</a> was originally published in <a href="https://medium.com/good-audience">Good Audience</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deep Learning: Where did it Come From?]]></title>
            <link>https://medium.com/115garage/deep-learning-where-did-it-come-from-f13523f89ed1?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/f13523f89ed1</guid>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Mon, 12 Aug 2019 20:50:02 GMT</pubDate>
            <atom:updated>2019-09-17T21:40:31.611Z</atom:updated>
            <content:encoded><![CDATA[<p>Deep Learning is one of the most popular buzzword thrown around in the tech community today. Everyone talks about Deep Learning, Machine Learning, Artificial Intelligence and how it’s the future of technology. The problem is — many people don’t know what they really are.</p><h3>Artificial Intelligence is a Field of Study</h3><p>I like to think of Artificial Intelligence as Engineering. Well, in Engineering, we have several sub-fields like: Civil engineering, Mechanical engineering, Chemical engineering, Software engineering, Electrical engineering, etc… So also does Artificial Intelligence have its own sub-fields: Computer Vision, Natural Language Processing, Expert Systems, Robotics, Optimization, Machine Learning, Data Science, Fuzzy Logic, etc…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0MCoUsTusG8gVsouTmEFmQ.jpeg" /><figcaption>Artificial Intelligence subfields</figcaption></figure><h3>Machine Learning</h3><p>One of the sub-fields of Artificial Intelligence is Machine Learning. In Machine Learning, instead of writing a program to solve a problem, we write a program to <em>learn to solve a problem from examples</em>. And it’s in this process of learning that the system gets better over time. Machine learning has 3 major categories: Supervised Learning, Unsupervised Learning &amp; Reinforcement Learning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T0zc6hUBd81hiORQbgx7ow.jpeg" /><figcaption>Rule-based cat finder</figcaption></figure><p>Why is it important to learn how to solve a problem from examples? Well I’ll show you an example. Let’s say we are given a task to write a program to recognize a cat in a picture. How do we do that? Well, one thing we can do is write some rules as to what a cat will look like. We can say to a computer, in Mathematical language that: <em>a cat has a round face with two pointy ears, a long tail and a chubby body</em>. Awesome! Well, this seems to work well enough for this cat <em>(on the left)</em>, right?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-o66smbA7Qksiscdiei2qQ.jpeg" /><figcaption>Cats hidden from viewport &amp; with different postures</figcaption></figure><p>Okay, what if we’re now given this cat <em>(on the right)</em> here? Our rules starts to break. But that’s fine, we can always update our rules and have more complex description of a cat. But then we’re given these other pictures of a cat who is hidden from viewport, and all these silly cats. Now what do we do? Notice that our program which we’ve spent forever writing rules for only works for cats and nothing else. If we’re told to recognize a dog, we’ll have to start writing the rules all over again. Yikes!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*swX9bAJaFBedBdgyHM5scg.jpeg" /><figcaption>Machine Learning</figcaption></figure><p>But what if we have a system that takes in a picture of a cat, and gives us back a score that <em>“I’m 92.6% sure that this is a cat”</em>? Wouldn’t that be great?! Well this is exactly the reason for Machine Learning. We give the machine lots of pictures of a cat and that of a dog, and then the machine starts to write those rules that we would have otherwise written ourselves. Now this system can be used for many other purposes. We could just swipe out the picture of a cat with that of a dog, or a car or an antelope.</p><h3>Machine Learning &amp; The Neural Network</h3><p>There are tons of Machine Learning models out there that are used for performing these tasks <em>(i.e learning from examples)</em>: Linear Regression, Support Vector Machines, Decision Trees, Logistic Regression, K-Nearest Neighbors, Naïve Bayes, Random Forest, K-Means, Mean Shift, just to name a few. But out of all these models there’s one that stands out — <strong>Neural Network</strong>. It was built roughly on the principles of how the brain works.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mqcQVuV4R9tCTGQyqWEG5A.jpeg" /><figcaption>Shallow Neural Network</figcaption></figure><p>A shallow neural network has 3 layers: Input layer, a hidden layer &amp; output layer. Each layer is made up of series of neurons <em>(which is a fancy term for Linear Transformation i.e. </em><em>y = Wx + b)</em>. What the input layer does is to take in some input data <em>(e.g. picture of a cat in our case)</em>, perform some transformation and give the result of those transformation to the next layer <em>(hidden layer)</em>. The hidden layer also does it&#39;s own transformation and sends the result to the output layer –which could say I&#39;m 94.7% sure that this is a cat!</p><p>This obviously is a very simplified version of what the neural network is. Anyways, the most interesting thing is what happens when you increase the number of hidden layers in the network.</p><h3>Let’s Go Deeper!</h3><p>This is what gave birth to the sub-field of Machine Learning called <strong>Deep Learning</strong>. Deep Learning is all about having many hidden layers and orchestrated in all sorts of ways that it outperforms all other vanilla Machine Learning models across a wide range of tasks. With more hidden layers, comes the need of more computing resources to train these deep networks &amp; also lots of data for the network to work with.</p><blockquote><strong>Therefore</strong>: when you have a deep neural network architecture + big data + sh#t loads of compute power –This is Deep Learning.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gccuMDV8fXjcvz1RSk4kgQ.png" /><figcaption>Neural Network Zoo</figcaption></figure><p>Today, there have been many deep neural network architectures –some popular ones are AutoEncoders, Generative Adversarial Network, Residual Network, Transformer, Inception, VGG16 and many more…</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f13523f89ed1" width="1" height="1" alt=""><hr><p><a href="https://medium.com/115garage/deep-learning-where-did-it-come-from-f13523f89ed1">Deep Learning: Where did it Come From?</a> was originally published in <a href="https://medium.com/115garage">115Garage</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Enhancing Human Intelligence]]></title>
            <link>https://medium.com/good-audience/enhancing-human-intelligence-601733274aa4?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/601733274aa4</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[neuroscience]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Mon, 03 Sep 2018 23:18:34 GMT</pubDate>
            <atom:updated>2019-10-24T08:06:49.186Z</atom:updated>
            <content:encoded><![CDATA[<p>On November 2017, I started digging into Human Intelligence and I was interested in why some “special human beings” have the ability to <a href="http://www.guinnessworldrecords.com/world-records/most-pi-places-memorised">memorize the value of PI (π) up to the 70,000th decimal place</a> or know how to <a href="https://www.onlinecollegecourses.com/2012/07/29/10-most-impressive-polyglots-in-world-history/">read and write in 58 different languages</a>. I’m fascinated in what the brain has to offer and the secret to super intelligence. I kept wondering: <em>“Everyone has this super-smart brain but not everyone can use it to its full capability, WHY?”</em>.</p><p>Hello, World! I’m Victor and I am <em>an Artificial Intelligence engineer/researcher (&amp; Software Engineer).</em> In this first part of <strong>Enhancing Human Intelligence</strong>, I want to give an overview of why we need this and the role Human Intelligence could contribute to the future, not to talk of the billions of unsolved problems we are facing in the world today. In the <em>next part</em> I’ll give a report on my research progress on making this technology a reality and not just a stuff of Science Fiction <em>(Sci-Fi)</em>.</p><blockquote><strong>DISCLAIMER</strong>: This content is exclusively licensed to <strong>Victor I. Afolabi</strong>. <strong>Intellectual Property (Media Ownership)</strong> is covered under the <a href="http://www.wipo.int/edocs/mdocs/mdocs/en/cdip_21/cdip_21_inf_2.pdf">World Intellectual Property Organization (CDIP/21/INF/2, p.13, s.2.3)</a>.</blockquote><blockquote><strong>PS:</strong> It is important to note that a number of details has been left out due to obvious reasons.</blockquote><h3>Overview</h3><p>We’ve managed to create <a href="https://en.wikipedia.org/wiki/Intelligent_Systems"><em>“Intelligent Systems”</em></a> that excel across a wide range of tasks, and understand us better than we understand each other. This is because machines are logical devices, that can process terabytes of data during an episode of <em>Star Trek: The Next Generation</em>. But as humans, we don’t come equipped with that kind of capability. We select actions that feels right to us. If I give you a very long article to read, you’d probably glance over it, and you’d magically know what the article is about <em>(or at least a good guess)</em>. A machine won’t do the same thing, it’ll <em>“read-in”</em> every single line, convert them to some sort of vector representation, run it through a bunch of Math and come back with a score: something you wouldn’t even consider doing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*go3L3oEzXpYjhY4Ltat0qQ.png" /></figure><p>Enhancing Human Intelligence is all about giving human the ability to process high dimensional data in ways we’d normally do it, but on a much more larger scale. We will be able to comb through the entire <em>History of Wall-Street Transactions</em> in a matter of seconds. It’s an experience that only few human being currently possesses. Our brain is a very fascinating general purpose computing device that is being underutilized because we have only 5 or 6 sensory organs that acts as inputs to our biological brain.</p><p>I am interested in creating <a href="https://en.wikipedia.org/wiki/Brain%E2%80%93computer_interface">Brain-Computer Interface</a> to augment our intelligence by adding much more sensory inputs to the brain. I know this sounds a bit fanciful, but I hope I’ll be able to convince you by the end of this article that it is at least plausible given various facts about the brain and recent advances in technology.</p><h3>Why Brain Matters</h3><p>The human brain is a really fascinating one, there has been thousands of research geared towards understanding the brain and each time we discover something completely mind-blowing. Neuroscientists, for hundreds of years are still actively researching the brain, especially the <strong>Neocortex <em>(the organ of intelligence)</em></strong>. Modern intelligent machines have been built on the principles of the Neocortex and the hierarchical structure of it’s anatomical and functional layouts. Other interesting part of the brain is <strong>The Hippocampus <em>(responsible for imagination and memory)</em></strong>.</p><p>One of the most important scientific problems of all time is <a href="https://numenta.com/blog/2016/06/16/can-neuroscientists-understand-the-brain/"><strong>Figuring out How the Brain Works</strong></a>. This fascinating organ has incredible functionalities — it makes predictions, models objects, generates behaviours, amongst many others. It is what allows a toddler to crawl across the room or learn how to open a door. It is what enables you to learn something by touching it with one hand and recognize it with the other. Your brain is what makes you, you! But how does it work?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v3rv0sxgwPfacSssRCCIlQ.png" /><figcaption>Neocortical column</figcaption></figure><p>Inside a part of your brain called the <strong>Neocortex</strong> are about 200k cortical columns as thick as a strand of spaghetti, 2½ millimetres long, arranged in a hierarchical layer –all doing the same thing. Companies like <a href="https://numenta.com/">Numenta</a> are knee on understanding how this tiny organ works and believe it is a fundamental piece in building machine intelligence.</p><h3>Why Human Intelligence</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9iY3Ct9RLWc26rAfxqJewA.png" /><figcaption>The Skull</figcaption></figure><p>For in this first part, I won’t be going over the more technical aspect of the brain and intelligence but all you need to know for now is that the brain doesn’t see the world as we all see and perceive it. It is locked up in a dark silent vault called <a href="http://teachmeanatomy.info/head/osteology/skull/"><strong>The Skull</strong></a>. All it receives are electrochemical signals which puts together all of, –“<em>these”</em>! Your subjective world. <em>Please refer to this </em><a href="https://www.youtube.com/watch?v=4c1lqFXHvqI"><em>TED Talk by David Eagleman</em></a><em> to learn more about this, which will turn out to be an important foundation that I will build on</em>. The brain is a general purpose biological device that processes incoming electrochemical signals, that are sent by our sensory organs. The brain does a really incredible job at giving pattern to these random signals coursing through our body and putting together a picture of a cat ticking your leg with its fur, or seeing a 3D object in a 2D drawing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/543/1*vJXr7jFHr4-c0wRFsHpQcA.jpeg" /><figcaption><a href="https://www.eagleman.com/research/sensory-substitution">Sensory substitution by David Eagleman</a></figcaption></figure><p>According David Eagleman, <a href="https://www.eagleman.com/research/sensory-substitution">sensory substitution</a> has been proven to be effective for helping the deaf hear the what the world sounds like again and blind people navigate through complex environmental obstacles by attaching electrotactile devices to the human body and sending these signals to the brain to make sense of the sensory inputs.</p><h3>Dual-Learning System</h3><p>My approach is to create a <strong><em>Dual-Learning System</em></strong>, which is an implementation of the <em>Brain-Computer Interface</em>. It is a two-way connection with the biological brain. Useful information are sent to the brain for biological pattern matching <em>(which the brain is surprisingly good at)</em> and a wearable device with high speed, high dimensional data crunching capabilities and built-in machine learning mechanisms.</p><p>I propose a wearable device <em>–a smart glass with specialized pair of handle, made up of EEG-like (electroencephalogram) transmitter that sends information to the brain</em>. More of the setup in the next part of <em>Enhancing Human Intelligence</em>, but to give a quick primer: <em>The brain receives random signal from the artificial device –which also learns to send “useful information” to the brain using an external reward signal</em>. If you’re familiar with <a href="https://arxiv.org/abs/1801.10459"><strong>Actor-Critic Model in Deep Reinforcement Learning</strong></a>, you can think of the device as the <em>“Actor”</em> and the brain as the <em>“Critic”</em>. The device takes a random action <em>(to start with)</em>, and the brain interpret these actions as either <em>“right/wrong”</em>. Based on feedback from the brain, an action is performed, transitioning to the next observable state in which a new action is being selected and the cycle continues.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FjaZ9aHhb_2JpbCW0jp8TA.png" /><figcaption>Learning architectural setup</figcaption></figure><p>For those that aren’t familiar with <a href="https://towardsdatascience.com/reinforcement-learning-w-keras-openai-actor-critic-models-f084612cfd69">Actor-Critic</a>, you may think of this setup as having two intelligent networks, one network <em>(a wearable)</em> receives information from raw unstructured data transmitted across the web or any (general) problem domain, and learns the most efficient way to feed the pre-processed data to the other network <em>(the brain)</em> so that it achieves an overall goal in the problem space.</p><h3>Challenges</h3><p>Just like any great project, there are bound to be challenges and drawbacks. As for the <em>Dual-Learning</em> mechanism, there are two major challenges. Other technical challenges will be listed as I dive deeper into this setup in the next chapters. The two obvious challenges are <em>Forgetfulness</em> &amp; <em>Learning Speed</em>.</p><h4>1. Forgetfulness</h4><p>Even though the biological brain learns to solve a given problem in a short period of time, it is subject to forgetting these learning process. Just like we forget what we were thought during Math class and ripping our hair off during the examination period. A solution to this problem is an <em>eXternal Memory Matrix (xMM)</em> and an <em>Asynchronous Psychological Mechanism (APM)</em>.</p><h4>2. Learning (Training) Speed</h4><p>My vision for this technology is to <em>learn anything in a reasonably short period of time</em>. However the <em>Dual-Learning System</em>, has a very slow theoretical timeframe. Learning something completely from scratch could take up to 2 weeks of training hours <em>–BAD!</em> This is because the brain is still by definition learning like it normally does, just that it now has raw, digital information as part of its learning stack.</p><h3>Conclusion</h3><p>I hope I’ve been able to at least whet your appetite and maybe convince you about the possibility of a harmless <em>Brain-Computer Interface</em> that involves no <strong><em>invasive surgery</em></strong> or <strong><em>cochlear implant</em></strong>. My main industrial application for this technology is the Educational Industry that has become a joke in recent years. Millions of students across the world find it challenging to understand what they learn in school.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eWYm3Dm7s0t8lLWYRW4pow.jpeg" /></figure><blockquote>I constantly imagine a 12 year-old kid having the same knowledge as an 80 year old professor today. I then wonder what an 80 year-old professor of the future would have unlocked?</blockquote><p>I’m currently working on lower level system abstractions for developers across the world to stretch their creativity and build on a more safer, secure future for all of us. Making an higher level <em>API</em> accessible for web developers, mobile developers, embedded systems and machine learning engineers. We’ve built intelligent machines that helps us speed up productivity, I think it’s high time we unlocked our unlimited potential and sped up our own creativity. <strong><em>This will be the spawn of a new evolution of Intelligence!</em></strong></p><p>Check out the <a href="https://blog.goodaudience.com/https-medium-com-victor-iyi-enhancing-human-intelligence-episode-2-f15d941024ec">next episode</a> of <strong>Enhancing Human Intelligence</strong>.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2F7b0181%3Fas_embed%3Dtrue&amp;dntp=1&amp;url=https%3A%2F%2Fupscri.be%2F7b0181%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/748a06e9019942f2677bba917ef55804/href">https://medium.com/media/748a06e9019942f2677bba917ef55804/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=601733274aa4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/good-audience/enhancing-human-intelligence-601733274aa4">Enhancing Human Intelligence</a> was originally published in <a href="https://medium.com/good-audience">Good Audience</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Year From Now You’d Have Wished You Started Today]]></title>
            <link>https://medium.com/@victor_iyi/a-year-from-now-youd-have-wished-you-started-today-9e0186aac27?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/9e0186aac27</guid>
            <category><![CDATA[life-lessons]]></category>
            <category><![CDATA[motivation]]></category>
            <category><![CDATA[work-life-balance]]></category>
            <category><![CDATA[truth-and-life]]></category>
            <category><![CDATA[motivational]]></category>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Sat, 03 Mar 2018 23:04:55 GMT</pubDate>
            <atom:updated>2018-03-03T23:04:55.243Z</atom:updated>
            <content:encoded><![CDATA[<p>When asked what surprised him most about humanity, the Dalai Lama answered, <strong><em>“We sacrifice our health in order to make wealth, then we sacrifice our wealth in order to get back our health”.</em></strong> We’re so anxious about what will happen in the future, that we don’t enjoy the present. The result being, we don’t live in the present or the future. We end up living feeling like we’re never going to die and then die never having truly lived.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*l3kqc4DNt5bca0umre6FpA.jpeg" /></figure><p>Ellen Goodman said, <strong><em>“normal is getting dressed in clothes that you buy for work and driving through traffic in a car that you’re still paying for, in order to get the job that you need to pay for the clothes and car and the home that’s empty all day just so you can afford to live in it”.</em></strong></p><p>We live in a world where day dreams are more common than dream jobs, where money clouds our search for meaning in search for profit. We sacrifice our quest for purpose and in the pursuit of those paycheques we give up our passion. We’re completely rooted to our routines, we’re plagued by indecision and lack of vision.</p><p>They say that the most powerful prison is one, when you don’t even know that you’re locked up. See, the suits are our uniforms, our cufflinks are the handcuffs, our ties are the chains, our boss is like the guard and well, our cubicles are our cells. And now I know why they call it the glass ceiling because it shatters our dreams. The only thing we excel at is Microsoft. Now that’s a Powerpoint.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7ay1XbA6vhsleQNnaQ0j_g.jpeg" /></figure><p>You know what it’s like, we spend our whole day trying to get our work on time, but then spend all of our time thinking about work. We wish our out of office could be permanent. We wish that our inbox would always remain empty. Our LinkedIn resumes are more interesting than our actual jobs. Our profile pictures are the happiest we’ve been in a long time. We live for the weekend but then end up feeling weak. Clockwork and monotony, but honestly could Mondays be more meaningful? Could Tuesdays be less torture? Could Wednesdays be more worth it? Could Thursdays be more transformative? Could Fridays be more fulfilling?</p><p>We countdown the next 50 weeks knowing there’s another 50 years. We countdown for the next escape knowing that there really isn’t one. Because even when we’re away from our desks, our minds are still there with the check lists in our heads. We stuff ourselves into our work clothes, then our buses, then the cars, then the trains and then our offices only to wait 5 days to live again. We don’t have time to eat dinner together so we end up eating out of take-away boxes not realising that the TV box takes away from our relationships. We eat on the go, meet on the go, we even sleep on the go and then wonder why everyone cheats on the go.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6-pGhXpsxLLmJUUPZejibg.jpeg" /></figure><p>We’re told it’s normal to have a 9 to 5. I mean, actually, sorry it’s a 9 to 9. It’s fine if you don’t get enough time with the person you love that took you so long to find. It’s normal to work late nights for the next holiday. Work 11 months in a row for 1 month of leave with pay. Hey, who came up with that? It’s not normal, we have to reclaim our lives because right now we can all barely survive, and because when he comes late home from work she says, <em>“You work so hard for a job that would replace you tomorrow if you dropped dead.”</em> And he says, <em>“I don’t have time for this right now.”</em> And then she says, <em>“Nothing ever changes”</em>.</p><p>We force ourselves out of bed to live the same day again and again and call it a life. Imagine this for a moment — what if Oprah listened to her haters and doubters? What if Steve Jobs settled for a real job? What if The Rock never broke through depression? What if Ellen never overcame the bullying? Imagine a world where everyone lived their passion. We’d be better people, we’d be better partners we’d be better parents.</p><p>See, I’ve never seen a strong person with an easy past. We all have a passion, we have a genius inside us. We have that potential. I know that we’ve all been in that position feeling confused and seriously lacking mission but then I ask myself — at the end of my life, what will I regret? And the answer is clear. The pain of regret far out weights the pain of risk. So next time you’re going through challenges, just remember this: <strong>A winner is just a loser who tried one more time.</strong> All you need is one person to say <em>“yes”</em>. There are decades when nothing happens and there are days where decades happen. <em>A year from now you would’ve wished you started </em><strong><em>today.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gq-3zpy4rrzaDw48KIpAtw.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9e0186aac27" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Views on Modern Relationships]]></title>
            <link>https://medium.com/@victor_iyi/relationship-status-87cfda4d39b2?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/87cfda4d39b2</guid>
            <category><![CDATA[relationships]]></category>
            <category><![CDATA[love]]></category>
            <category><![CDATA[dating]]></category>
            <category><![CDATA[emotions]]></category>
            <category><![CDATA[thoughts-and-feelings]]></category>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Mon, 19 Feb 2018 22:18:23 GMT</pubDate>
            <atom:updated>2018-02-19T23:01:46.159Z</atom:updated>
            <content:encoded><![CDATA[<p>We’re the generation that <em>doesn’t</em> want relationships. We want <em>a second cup of coffee</em> in our Instagram’s of lazy Saturday mornings. We want <em>a second pair of shoes</em> in the artsy pictures we take of our feet. We want that <strong>“Facebook official relationship”</strong> that everyone can <em>like</em> and <em>comment on.</em> We want the <em>post</em> that <em>wins relationship goals.</em> We want <em>that date</em> for <em>Sunday morning brunch,</em> someone to drown our sorrows for the <em>Monday blues.</em> We want that Taco Tuesday partner, <em>someone to text us</em> <em>“Good morning”</em> on a Wednesday. We want a <strong>+1</strong> for all the <em>weddings</em> we keep getting invited to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ncbhJ_6Vp2e56Z3KDBHn5g.jpeg" /></figure><p>How did they do it? How did they find their <strong>happily ever after?</strong> But we’re the generation that <strong>doesn’t want a relationship.</strong> We <em>swipe left</em> in hopes of finding the right one, we try to special order our <em>soul mates</em> like we do on <strong>Postmates.</strong> We read <strong><em>“5 Ways To Know He’s Into You”</em></strong> and <strong><em>“7 Ways She Will Fall For You”</em></strong>, thinking we can up-cycle a person into a relationship like a <em>Pintrest project.</em> We invest more time into our <em>Tinder profiles</em> than we do with <em>our personalities,</em> yet we’re the ones who don’t want a relationship.</p><p>We <em>talk</em> and we <em>text</em>. We <em>Snapchat</em> and we <em>sext.</em> We go get <em>coffee</em> and grab <em>a beer.</em> Anything to avoid an actual date. We send <em>private messages</em> to meet up, small talk for an hour only to return home and then <em>small talk via text.</em></p><p>We forgo any chance of achieving <strong>real connection</strong> by mutually playing games with <strong>no winner.</strong> The only thing we end up winning is <strong>most likely to be alone.</strong> We want the <em>facade</em> of a relationship without the <em>work</em> of a relationship, we want the <em>hand holding</em> without the <em>eye contact.</em> We want the <em>teasing</em> without the <em>serious conversations.</em> We want the pretty promise <strong>without</strong> the actual commitment. We want to celebrate the anniversaries <strong>without</strong> the 365 days that lead up to them. We want the happily ever after <strong>without</strong> the effort in the here and now, we want to have deep conversations <em>but</em> still keep things shallow. We want that <em>“World Series”</em> kind of love <em>without</em> killing to go bat. We want someone to <em>hold our hands,</em> but we don’t want to put the power to hurt us in their hands. We want to be <em>swept off our feet</em> but at the same time remaining safely, independently <em>standing on our own.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*r5pHwCpBtdPJMLnoY245Hw.jpeg" /></figure><p>We want to keep chasing love, but <strong>we don’t actually want to fall into it.</strong> We don’t want relationships, we want <em>friends with benefits,</em> we want <em>Netflix &amp; Chill</em> and <em>nudes on Tinder.</em> We want everything that will give us the <em>“illusion of a relationship”</em> without an actual relationship. We want the rewards with <strong>no risk,</strong> we want the payout with <strong>no cost.</strong> We want to <em>connect enough</em> but <em>not too much,</em> we want to <em>commit a little</em> but <em>not a lot.</em> We take it <strong>slow</strong>, we see where it <strong>goes,</strong> we don’t want to label things, we just go with the <strong>flow.</strong> We keep one foot out the door, we keep <strong>one eye open,</strong> we keep people at arms length, <strong>toying with their emotions</strong> but mostly <strong>toying with our own.</strong></p><p>When things get too close to being real we run, we hide, we leave, we say to ourselves <strong>“there’s more fish in the sea”.</strong> We want the <em>“downloadable person”</em> that’s a <em>perfect fit,</em> just like an app you can <em>update whenever there’s a hitch,</em> compartmentalised into a folder that we can delete <em>when we have no more need for it.</em> We don’t want to <strong>unpack our baggage,</strong> or worse; help someone else <em>unpack theirs.</em> We hide everything behind an <em>Instagram filter,</em> we choose a <em>Netflix show</em> over a <em>real conversation.</em> We feel entitled <em>to love</em> like we feel <em>entitled to full-time jobs</em> out of college. We want a <strong>placeholder</strong> not a <em>person</em>, we want a <strong>warm body</strong> not a <em>partner</em>. We want someone to <em>eat with</em> while we scroll through our <em>newsfeed.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fpp1CQ1PAYPebEP1jHIarw.jpeg" /></figure><p>See, what we need to recognise is that the things we <em>truly want,</em> the things that are <em>deeply meaningful,</em> the things that are <em>genuinely fulfilling,</em> all require <em>patience.</em> They all require <em>work,</em> they all require <em>energy.</em> See, the challenge is we all want to <em>be with someone</em> who makes us happy, when what we need to do is <em>be someone who makes us happy.</em> We sit with out friends discussing the rules but no one even knows <strong>what game we’re trying to play</strong> because the problem with our generation <em>not wanting relationships</em> is that — at the end of the day, <strong>we actually do.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dVt74b8OUI6eU9LPN4yLAA.jpeg" /></figure><p><em>Inspired by </em><a href="https://www.youtube.com/channel/UCbV60AGIHKz2xIGvbk0LLvg/featured"><em>Jay Shetty</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=87cfda4d39b2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google’s Artificial Intelligence Built an AI That Outperforms Any Made by Humans]]></title>
            <link>https://medium.com/@victor_iyi/googles-artificial-intelligence-built-an-ai-that-outperforms-any-made-by-humans-3eb69fffdbae?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/3eb69fffdbae</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[google-brain]]></category>
            <category><![CDATA[futurism]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Sun, 03 Dec 2017 17:27:40 GMT</pubDate>
            <atom:updated>2017-12-03T17:33:42.910Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/578/1*4ZXbPTMUHRGRTutEKC1AqA.jpeg" /></figure><blockquote><strong>What is this all about?</strong></blockquote><blockquote>Google’s AutoML project, designed to make AI build other AIs, has now developed a computer vision system that vastly outperforms state-of-the-art-models. The project could improve how autonomous vehicles and next-generation AI robots “see.”</blockquote><p><a href="https://medium.com/@javafolabi/googles-artificial-intelligence-built-an-ai-that-outperforms-any-made-by-humans-842fb4e5b16b">Part 1</a> | Part 2</p><p>In the <a href="https://medium.com/@javafolabi/googles-artificial-intelligence-built-an-ai-that-outperforms-any-made-by-humans-842fb4e5b16b">last post</a> I talked about an AI that can build other AIs and how it achieved state-of-the-art performance compared to any of it’s human counterparts. In this post we’ll review what this means for the future of AI and technology in general.</p><h3>A View of the Future</h3><p>Machine learning is what gives many AI systems their ability to perform specific tasks. Although the concept behind it is fairly simple — an algorithm learns by being fed a ton of data — the process requires a huge amount of time and effort. By automating the process of creating accurate, efficient AI systems, an AI that can build AI takes on the brunt of that work. Ultimately, that means AutoML could open up the field of machine learning and AI to non-experts.</p><p>As for NASNet specifically, accurate, efficient <a href="https://futurism.com/microsoft-our-computer-vision-algorithms-have-surpassed-humans/">computer vision algorithms</a> are highly sought after due to the number of potential applications. They could be used to create sophisticated, AI-powered robots or to help visually impaired people regain sight, <a href="https://futurism.com/computer-vision-and-mobile-technology-could-help-blind-people-see/">as one researcher suggested</a>. They could also help designers improve self-driving vehicle technologies. The faster an <a href="https://futurism.com/tesla-brings-on-a-leading-computer-vision-expert-to-give-sight-to-ai/">autonomous vehicle can recognize objects</a> in its path, the faster it can react to them, thereby increasing the safety of such vehicles.</p><p>The Google researchers acknowledge that NASNet could prove useful for a wide range of applications and have open-sourced the AI for <a href="https://github.com/tensorflow/models/tree/master/research/slim/">inference on image classification</a> and <a href="https://github.com/tensorflow/models/tree/master/research/object_detection">object detection</a>. “We hope that the larger machine learning community will be able to build on these models to address multitudes of computer vision problems we have not yet imagined,” they wrote in their blog post.</p><p>Though the applications for NASNet and AutoML are plentiful, the creation of an AI that can build AI does <a href="https://www.fastcodesign.com/90152427/googles-new-ai-designs-ai-better-than-humans">raise some concerns</a>. For instance, what’s to prevent the parent from passing down unwanted biases to its child? What if AutoML creates systems so fast that society can’t keep up? It’s not very difficult to see how NASNet could be employed in automated surveillance systems in the near future, perhaps sooner than regulations could be put in place to control such systems.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FjNRJkNjSljA%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DjNRJkNjSljA&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FjNRJkNjSljA%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/07b914fe5fb76360cacc7c0a7709b7b6/href">https://medium.com/media/07b914fe5fb76360cacc7c0a7709b7b6/href</a></iframe><p>Thankfully, world leaders are working fast to ensure such systems don’t lead to any sort of dystopian future.</p><p>Amazon, Facebook, Apple, and several others are all members of the <a href="https://futurism.com/partnership-on-ai-tech-giants-unite-to-develop-synthetic-intelligence-for-humanity/">Partnership on AI to Benefit People and Society</a>, an organization focused on the responsible development of AI. The Institute of Electrical and Electronics Engineers (IEE) has proposed <a href="https://futurism.com/powerful-tech-organization-working-protect-ai/">ethical standards for AI,</a> and DeepMind, a research company owned by Google’s parent company Alphabet, recently announced the creation of group focused on the <a href="https://futurism.com/a-world-leader-in-ai-just-established-an-ethics-committee-for-artificial-intelligence/">moral and ethical implications of AI.</a></p><p>Various governments are also working on regulations to prevent the use of AI for dangerous purposes, such as <a href="https://futurism.com/following-elon-musk-letter-uk-government-plans-to-ban-fully-autonomous-weapons/">autonomous weapons</a>, and so long as humans maintain control of the overall direction of AI development, the benefits of having an AI that can build AI should far outweigh any potential pitfalls.</p><h4>References:</h4><p><a href="https://www.fastcodesign.com/90152427/googles-new-ai-designs-ai-better-than-humans">Co.Design</a>, <a href="https://research.googleblog.com/2017/11/automl-for-large-scale-image.html">Google Research</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3eb69fffdbae" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google’s Artificial Intelligence Built an AI That Outperforms Any Made by Humans]]></title>
            <link>https://medium.com/@victor_iyi/googles-artificial-intelligence-built-an-ai-that-outperforms-any-made-by-humans-842fb4e5b16b?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/842fb4e5b16b</guid>
            <category><![CDATA[futurism]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[reinforcement-learning]]></category>
            <category><![CDATA[google-brain]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Sun, 03 Dec 2017 17:10:09 GMT</pubDate>
            <atom:updated>2017-12-03T17:32:37.880Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/578/1*4ZXbPTMUHRGRTutEKC1AqA.jpeg" /></figure><blockquote><strong>What is this all about?</strong></blockquote><blockquote>Google’s AutoML project, designed to make AI build other AIs, has now developed a computer vision system that vastly outperforms state-of-the-art-models. The project could improve how autonomous vehicles and next-generation AI robots “see.”</blockquote><p>Part 1 | <a href="https://medium.com/@javafolabi/googles-artificial-intelligence-built-an-ai-that-outperforms-any-made-by-humans-3eb69fffdbae">Part 2</a></p><h4>An AI That can Build AIs</h4><p>In May 2017, researchers at <a href="https://research.google.com/teams/brain/">Google Brain</a> announced the creation of <a href="https://research.googleblog.com/2017/05/using-machine-learning-to-explore.html">AutoML</a>, an artificial intelligence (AI) that’s capable of generating its own AIs. More recently, they decided to present AutoML with its biggest challenge to date, and the AI that can build AI created a “child” that outperformed all of its human-made counterparts.</p><p>The Google researchers <a href="https://futurism.com/googles-new-ai-is-better-at-creating-ai-than-the-companys-engineers/">automated the design of machine learning models</a> using an approach called <a href="https://futurism.com/googles-machine-learning-software-has-learned-to-replicate-itself/">reinforcement learning</a>. AutoML acts as a controller neural network that develops a child AI network for a specific task. For this particular child AI, which the researchers called <strong>NASNet</strong>, the task was recognizing objects — people, cars, traffic lights, handbags, backpacks, etc. — in a video in real-time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6SZR7lZIf1QsJmP2Q5qWiw.jpeg" /><figcaption>Image Credit: Google Research</figcaption></figure><p>AutoML would evaluate NASNet’s performance and use that information to improve its child AI, repeating the process thousands of times. When tested on the <a href="http://www.image-net.org/">ImageNet image classification</a> and <a href="http://cocodataset.org/">COCO object detection</a> data sets, which the Google researchers call <em>“two of the most respected large-scale academic data sets in computer vision,”</em> NASNet outperformed all other computer vision systems.</p><p>According to the researchers, NASNet was 82.7% accurate at predicting images on ImageNet’s validation set. This is 1.2% better than any <a href="https://arxiv.org/abs/1709.01507">previously published results</a>, and the system is also 4% more efficient, with a 43.1% mean Average Precision (mAP). Additionally, a less computationally demanding version of NASNet outperformed the best similarly sized models for mobile platforms by 3.1 percent.</p><p><strong><em>See what this means for the future in the </em></strong><a href="https://medium.com/@javafolabi/googles-artificial-intelligence-built-an-ai-that-outperforms-any-made-by-humans-3eb69fffdbae"><strong><em>next part</em></strong></a><strong><em>.</em></strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=842fb4e5b16b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Neural Nets Basics]]></title>
            <link>https://medium.com/@victor_iyi/neural-nets-basics-part-one-93fc3ad3f80?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/93fc3ad3f80</guid>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[numpy]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[backpropagation]]></category>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Sun, 12 Nov 2017 21:13:20 GMT</pubDate>
            <atom:updated>2017-12-03T17:37:17.802Z</atom:updated>
            <content:encoded><![CDATA[<p>You can download the <a href="https://github.com/victor-iyiola/neural-net-basics">full source code here</a></p><p>If you’re like me, learning about neural net could be pretty confusing. As a curious person who just wants to see this so called <em>neural network</em> in action. Materials on the internet about neural networks could be so confusing (too mathematical or too much code). So I decided to make the best of both. Teach you what you’ll need to get started developing neural networks.</p><p>Let’s build a two layer neural network to predict the mappings between binary digits. The only dependency needed is <a href="https://pypi.python.org/pypi/numpy"><strong>numpy</strong></a>. If you’re not familiar with <a href="http://www.numpy.org/"><strong>numpy</strong></a> it’s okay. You’ll still be able to follow up just fine!</p><p>After installing <strong>numpy</strong> by running <em>pip install numpy </em>or downloading it <a href="https://pypi.python.org/pypi/numpy">here</a>. We’re gonna want to import <strong>numpy</strong>. It’s common convention to import <strong>numpy</strong> as <strong>np</strong></p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/695bffbfdfd2ac5e7601970ee1f7ab24/href">https://medium.com/media/695bffbfdfd2ac5e7601970ee1f7ab24/href</a></iframe><p>Let’s define our features and labels, shall we?</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/095e0be22b513294e1e34f21371d911a/href">https://medium.com/media/095e0be22b513294e1e34f21371d911a/href</a></iframe><p><strong>X</strong> is our <em>input/feature</em>, while <strong>y</strong> is our <em>target/label</em>. Features in machine learning are the relevant information about a piece of data. For example: if I want to predict the score of the next soccer match, it’ll be nice if I have player’s statistics and past match history <em>(I’m not a soccer fan)</em>. Those statistics represent our features while the score can be our <em>target/labels</em>, since that’s what we want to predict.</p><p>Now that we understand what features and labels are. It’s time to define some <a href="https://en.wikipedia.org/wiki/Hyperparameter_(machine_learning)"><em>hyperparameters</em></a><a href="https://en.wikipedia.org/wiki/Hyperparameter_(machine_learning)).">.</a> <em>Hyperparameters</em> in machine learning are the tuning knobs for our network. For instance, how many layers do we want? How many training iterations do we want to perform? Usually, one wouldn’t know what the best <em>hyperparameters</em> are right off the bat. You’ll need some tweaking to know the best set of <em>hyperparameters</em> to use. In this case we’re gonna keep it simple. Let’s define our <em>input dimension</em>, <em>hidden dimension</em> &amp; <em>output dimension</em>.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/12a6ac21feee05b1a7a75fd0dcbd9851/href">https://medium.com/media/12a6ac21feee05b1a7a75fd0dcbd9851/href</a></iframe><p>It is time to start building our model. We’ll start by initializing our weights. Weights are the major sauce in Machine learning. When we say we are training a neural network, what we really mean is that we are trying to find appropriate weights that are going to best give accurate solutions to the problem to be solved. If you want an in-depth explanation on weights in machine learning check out <a href="http://datathings.com/blog/post/neuralnet/">this blog post</a></p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/bebab3b0933f516f8d1be91a593a7127/href">https://medium.com/media/bebab3b0933f516f8d1be91a593a7127/href</a></iframe><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*sFZUx7FwEZhz4sJbztQoxA.png" /><figcaption>Sigmoid activation function</figcaption></figure><p>We will choose the <a href="https://en.wikipedia.org/wiki/Sigmoid_function">sigmoid activation function</a> as our <em>“nonlinearity”</em>. While it can be several kinds of functions. <strong>A sigmoid function</strong> maps any value to a value between 0 and 1. We use it to convert numbers to probabilities. It also has several other desirable properties for training neural networks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*OgFk_M_QNNn_1aqEMrhpcA.png" /><figcaption>Sigmoid function: Derivative</figcaption></figure><p>In order to <em>“backpropagate”</em> (which we’ll talk about in a second) we are going to find the derivative of the sigmoid function. This can be represented in Python as just one line of code. #smiles</p><p>If you’re unfamiliar with derivatives, just think about it as the slope of the sigmoid function at a given point (as you can see above, different points have different slopes). For more on derivatives, check out this <a href="https://www.khanacademy.org/math/differential-calculus/taking-derivatives/derivative_intro/v/calculus-derivatives-1">derivatives tutorial from Khan Academy</a>.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/6707464e049e47f7723748175f7ce51a/href">https://medium.com/media/6707464e049e47f7723748175f7ce51a/href</a></iframe><p>It’s training time!</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/a6c71c3cda9460bd6cfb1f6594ab2f22/href">https://medium.com/media/a6c71c3cda9460bd6cfb1f6594ab2f22/href</a></iframe><p>It’s time to start training this Neural net! Here’s the plan: We are going to multiply our input with the first set of weights <em>(line 3)</em>. We’ll then pass the result to the next “layer” to multiply with the next weight <em>(line 4)</em>. At this point we have our prediction. But our prediction would <em>(usually)</em> be wrong. So we’re going to estimate by how much we missed i.e. calculate error between our prediction and the actual output <em>(line 7)</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/396/1*Kb2zrzDY1bPdufHJdVoVHg.png" /><figcaption>Neural network (visualisation)</figcaption></figure><p>In order to know how bad our network performed in layer two we’ll need to do an <em>element wise multiplication </em>of our layer 2’s error with our sigmoid derivative <em>(Good thing we already wrote the function line 8)</em>. Now we know how bad our network performed in layer two.</p><p>Since layer 2 depends on layer 1, we would need to know how bad layer 1 is also. Because if layer 1 get it wrong, layer 2 will also do poorly. Therefore we’ll pass the error we got in layer 2 to layer 1 by matrix multiplying layer 2’s error with the <em>weight transposed</em> that connects both layers <em>(line 11)</em>. After this we can do an <em>element wise multiplication</em> of the error with the sigmoid’s derivative <em>(line 12, just like we did for layer 2)</em>.</p><p>Now that we have our error propagated to the first layer, it’s time to update our weights, such that in the next iteration we won’t make the mistake we made at the current time step.</p><p>This whole process is called <strong>“Backpropagation”.</strong> The<em> backpropagation algorithm</em> is the reason why our neural network works so well. And sadly, it’s the reason why neural networks generally take days or weeks to train as your dataset scales.</p><h4>Conclusion</h4><p>We looked at how to build a very simple neural network without too much Math or too much code (just as promised). This is the first step to building <a href="https://medium.com/@karpathy/software-2-0-a64152b37c35"><em>Software 2.0</em></a> (like <a href="https://medium.com/@karpathy">Andrej Karpathy</a> said). Try to modify the dataset or switch datasets and see what happens. You can also try printing out the weights and each layer’s results.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=93fc3ad3f80" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[No Free Lunch Theorem: What you should watch out for when developing your algorithms.]]></title>
            <link>https://medium.com/115garage/no-free-lunch-theorem-what-you-should-watch-out-for-when-developing-your-algorithms-e9abff4c1d99?source=rss-3219c452aea6------2</link>
            <guid isPermaLink="false">https://medium.com/p/e9abff4c1d99</guid>
            <category><![CDATA[algorithms]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[optimization]]></category>
            <category><![CDATA[search]]></category>
            <dc:creator><![CDATA[Victor I. Afolabi]]></dc:creator>
            <pubDate>Sat, 11 Nov 2017 15:28:23 GMT</pubDate>
            <atom:updated>2017-11-11T15:28:23.728Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*N7uHKUsqKr6I3jhiHJv0rQ.jpeg" /><figcaption>No such thing as free lunch! <a href="https://1.bp.blogspot.com/-ukDz8Eub6Gg/TmdpHEhP7XI/AAAAAAAACYM/TCwc1g8Yiyk/s1600/free-lunch-demotivational-poster-1222110638.jpg">(source)</a></figcaption></figure><p>The <a href="https://en.wikipedia.org/wiki/No_free_lunch_theorem"><strong>“No free lunch”</strong></a> theorem states that there’s no one model that works for every problem. The assumptions of a great model for one problem may not hold for another problem. So it is common in Machine Learning to try multiple models and find one that works best for a particular problem. A model that works well could also be trained by multiple algorithms — for example <a href="https://en.wikipedia.org/wiki/Linear_regression">linear regression</a> could be trained by the <a href="http://www.statisticshowto.com/what-is-a-regression-equation/">normal equations</a> or by <a href="https://machinelearningmastery.com/implement-linear-regression-stochastic-gradient-descent-scratch-python/">gradient descent.</a></p><h3>No Free Lunch (NFL) Meatphor</h3><p>In the “no free lunch” metaphor, each <em>“restaurant”</em> (problem-solving procedure) has a <em>“menu”</em> associating each <em>“lunch plate”</em> (problem) with a <em>“price”</em> (the performance of the procedure in solving the problem). The menus of restaurants are identical except in on regard — the prices are shuffled from one restaurant to the next.</p><p>For an omnivore who is a s likely to order each plate as any other, the average cost of lunch does not depend on the choice of restaurant. But the Vegan who goes to lunch regularly with a carnivore who seeks economy might pay a high average cost for lunch. To methodically reduce the average cost, one must use knowledge of (<em>a) what one will order and (b) what the order will cost at various restaurants.</em></p><p>That is; improvement of performance in problem-solving hinges on using prior information to match procedures to problems.</p><h3>NFL in Search and Optimization</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/623/1*0QP3OeK7BAOWGlUcDG6VSw.png" /><figcaption>General purpose Algorithm (<a href="http://photos1.blogger.com/blogger/1699/2130/1600/Nflt.png">Source</a>)</figcaption></figure><p>Some computational problems are solved by searching for good solutions in a space of candidate solutions. A description of how to repeatedly select candidate solutions for evaluation is called a search algorithm. On a particular problem, different search algorithms may obtain different results, but over all problems, they are indistinguishable. It follows that if an algorithm achieves superior results on some problems, it must pay with inferiority on the other problems. In this sense there is no free lunch in search. Alternatively, following Schaffer, search performance is conserved. Usually search is interpreted as optimization, and this lead to the observation that there is no free lunch in optimisation.</p><h3>Conclusion.</h3><p>Depending on the problem, it is important to assess the trade-offs between <strong>speed</strong>, <strong>accuracy</strong>, and <strong>complexity</strong> of different models and algorithms and find a model that works best for that particular problem.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e9abff4c1d99" width="1" height="1" alt=""><hr><p><a href="https://medium.com/115garage/no-free-lunch-theorem-what-you-should-watch-out-for-when-developing-your-algorithms-e9abff4c1d99">No Free Lunch Theorem: What you should watch out for when developing your algorithms.</a> was originally published in <a href="https://medium.com/115garage">115Garage</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>