<?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 Ramakrishnan Sivakumar on Medium]]></title>
        <description><![CDATA[Stories by Ramakrishnan Sivakumar on Medium]]></description>
        <link>https://medium.com/@ramkrishna2910?source=rss-d15f97e87633------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Ht9yE8kxYJF3Dt4d5GDT4g.jpeg</url>
            <title>Stories by Ramakrishnan Sivakumar on Medium</title>
            <link>https://medium.com/@ramkrishna2910?source=rss-d15f97e87633------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 16:27:51 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ramkrishna2910/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[Why do companies Open Source Software?]]></title>
            <link>https://ramkrishna2910.medium.com/why-do-companies-open-source-software-fc1b2c508206?source=rss-d15f97e87633------2</link>
            <guid isPermaLink="false">https://medium.com/p/fc1b2c508206</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[llama-2]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Ramakrishnan Sivakumar]]></dc:creator>
            <pubDate>Mon, 29 Jul 2024 01:06:15 GMT</pubDate>
            <atom:updated>2024-07-29T01:06:15.865Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wZlvkVjiPIoQrFUGcYllQw.jpeg" /></figure><p>Open sourcing software has been around for decades, but it is becoming an increasingly important part of the current technological landscape. Despite its growing prevalence, there remains much debate about why companies choose to give away the software they have built for free. In this blog, I aim to explore some of the key reasons behind this trend, shedding light on the benefits and motivations driving companies to embrace open source.</p><p><strong>Accelerating Development Momentum</strong></p><p>Open sourcing software significantly accelerates development momentum by inviting contributions from the community. It showcases the technical expertise of the team behind the project, enhancing their reputation and credibility. Additionally, open source projects attract top talent, fostering a collaborative environment that leads to better product development. For example, <a href="https://github.com/huggingface/transformers">HuggingFace’s Transformers</a> repository, used in around 200K projects, has 2.6K contributors despite the company employing only about 200 people. This vast network of contributors accelerates innovation and growth far beyond what the internal team alone could achieve.</p><p><strong>Embracing Transparency and Building Trust</strong></p><p>Open sourcing software can serve as a powerful tool to show users exactly what they are using, which is particularly important in scenarios where privacy is a significant concern. For example, Apple prioritizes privacy in its products and has adopted open sourcing for its <a href="https://security.apple.com/blog/private-cloud-compute/">private cloud compute</a>. By open sourcing the runtime in production on their private cloud, Apple ensures that users have access and visibility into the code being deployed. This transparency helps build trust with users by providing them with assurance over how their data is handled.</p><p><strong>Demonstrating Proof of Maturity</strong></p><p>New projects or startups often use open source to demonstrate their level of maturity. By open sourcing parts of their product or the entire project, teams can showcase the robustness and reliability of their offerings. This transparency is particularly valuable for external entities such as investors, potential customers, or developers who wish to independently evaluate the product.</p><p>Additionally, when a company open sources its product, it reflects the confidence they have in their own work. By making their code available for anyone to use without tailoring or controlling the user experience, they signal trust in the quality and stability of their product.</p><p><strong>Enhancing Brand Recognition</strong></p><p>Open sourcing can significantly enhance brand recognition and shape the narrative around a project or company. For instance, Meta is regarded as a leader in open-source AI not solely because they have deployed AI in products used by billions or have access to 600k GPUs, but because they have consistently open sourced high-quality models. This commitment to sharing their advancements has positioned Meta at the forefront of the open-source AI community.</p><p><strong>Educational Benefits</strong></p><p>Open source projects are invaluable educational resources for developers of all skill levels. In a closed-source environment, users miss out on understanding the inner workings of complex and sophisticated software. For instance, PyTorch is widely used in numerous products, and its open-source nature allows users to explore and learn from its detailed implementations, gaining insights into cutting-edge software development.</p><p>Moreover, open source projects attract more attention to the field, raising the overall knowledge base of users and developers. <a href="https://github.com/karpathy/LLM101n">Andrej Karpathy’s open-sourced coursework</a> is popular for this very reason, providing a deep learning resource that enhances the skills and understanding of the community.</p><p><strong>Building an Ecosystem</strong></p><p>Open sourcing software helps build an ecosystem around your products. For instance, Meta’s LLaMA, despite costing millions of dollars to train, has its weights published for free. Meta values the creation of a robust ecosystem over the training costs. LLaMA fosters a vibrant ecosystem of tools, applications, and research, driving the adoption of AI technologies. This strategy positions Meta’s models as industry standards, indirectly benefiting the company through increased usage and integration of their technologies, and establishing Meta as a leader in the AI community.</p><p><strong>Gathering Valuable Feedback</strong></p><p>Open sourcing software enables a diverse user base to test the product in various scenarios and provide feedback through issues. This feedback often significantly influences the project’s roadmap, offering developers insights into what doesn’t work and what the community aims to achieve with the software. Such real-world feedback can be far more valuable than internal testing, guiding improvements and ensuring the product meets the needs of its users.</p><p><strong>Some of the Greatest Open Source Projects</strong></p><p>Open source projects have revolutionized the tech industry, driving innovation and collaboration. Here are some of the most influential open source projects and their remarkable contributions:</p><p><strong>Linux</strong></p><p><a href="https://github.com/torvalds/linux">Linux</a> is an open-source operating system kernel that has become the foundation for various operating systems, known as distributions, like Ubuntu, Fedora, and Debian. Its stability, security, and flexibility make it ideal for a wide range of applications, from web servers and supercomputers to Android devices. Linux powers many systems from the server infrastructures of tech giants such as Google, Facebook, and Amazon to SpaceX rockets, showcasing its significance in the tech world.</p><p><strong>Apache HTTP Server</strong></p><p>The <a href="https://github.com/apache/httpd">Apache HTTP Server</a>, commonly known as Apache, is a powerful and flexible web server software. It plays a crucial role in serving web pages and applications across the internet. Apache’s robustness and extensive module support have made it the go-to choice for hosting websites like WordPress and powering platforms such as Wikipedia. Its ability to handle large volumes of traffic and support for various programming languages highlight its versatility.</p><p><strong>Git</strong></p><p><a href="https://github.com/git/git">Git</a> is a distributed version control system that has transformed collaborative software development. Git allows multiple developers to work on a project simultaneously without overwriting each other’s changes. It is essential for managing source code repositories and tracking changes in software projects. Git powers major platforms like GitHub, GitLab, and Bitbucket, and is used by countless projects, including the development of the Linux kernel itself.</p><p><strong>PyTorch</strong></p><p><a href="https://github.com/pytorch/pytorch">PyTorch</a> is a cutting-edge machine learning library developed by Facebook’s AI Research lab (FAIR). It provides an intuitive framework for building and training neural networks, making it a favorite among researchers and developers. PyTorch is widely used in deep learning applications such as computer vision, natural language processing, and reinforcement learning. It has powered innovations like Uber’s Pyro for probabilistic programming and Tesla’s autonomous driving systems, cementing its place as a cornerstone of modern AI research.</p><p><strong>Conclusion</strong></p><p>The world around us is increasingly powered by open source software, from the operating systems running our devices to the web servers delivering our favorite websites and the machine learning frameworks driving AI innovations. This collaborative approach to software development has proven to be a powerful tool for accelerating progress, fostering community, and driving technological advancements. For companies and developers, leveraging open source is not just an option but a necessity. It unlocks potential for greater innovation, efficiency, and adaptability, ensuring that they stay at the forefront of their fields in an ever-evolving tech landscape. Embracing open source is a testament to the power of shared knowledge and collective progress.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fc1b2c508206" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[GraphCast: A breakthrough in Weather Forecasting]]></title>
            <link>https://ramkrishna2910.medium.com/graphcast-a-breakthrough-in-weather-forecasting-d70fae9ac365?source=rss-d15f97e87633------2</link>
            <guid isPermaLink="false">https://medium.com/p/d70fae9ac365</guid>
            <category><![CDATA[weather-forecasts]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[graph-neural-networks]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Ramakrishnan Sivakumar]]></dc:creator>
            <pubDate>Mon, 27 Nov 2023 05:23:32 GMT</pubDate>
            <atom:updated>2023-11-27T05:23:32.208Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lLXc0acFL3V_oksWjTAMFg.png" /></figure><h3>Introduction</h3><p>Weather forecasting is crucial for public safety, agriculture, transportation, and energy management, helping to mitigate risks and optimize decisions. It also plays a key role in understanding and responding to environmental changes and climate patterns.</p><p>The field of weather forecasting is witnessing a revolutionary shift with the advent of <a href="https://arxiv.org/pdf/2212.12794.pdf"><strong>GraphCast</strong></a>, a graph neural network model designed for meteorological predictions. This shift from traditional numerical weather models to a machine learning-based approach is not just a technological upgrade but a complete reimagining of how we understand and predict atmospheric phenomena.</p><h3>Traditional Weather Forecasting</h3><p>A Computational Behemoth Traditional weather forecasting methods dissect the atmosphere into a grid of small cells, applying physics-based calculations to each. This process, while effective, demands immense computational resources. Major forecasting systems like <a href="https://www.ecmwf.int/en/computing/our-facilities/supercomputer-facility">ECMWF </a>and <a href="https://www.ncei.noaa.gov/products/weather-climate-models/global-forecast">GFS</a> use supercomputers with capabilities measured in petaflops, costing hundreds of millions of dollars. Despite their power, these systems face limitations in speed and flexibility, particularly when dealing with the vast and dynamic nature of atmospheric data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/286/0*NwfzF7snQUWD3N3i.jpg" /><figcaption>Figure 1: 3D dissection of the atmosphere for weather modeling [2]</figcaption></figure><h3>Intro to GraphCast</h3><p>GraphCast introduces a novel approach by leveraging the capabilities of machine learning, specifically graph neural networks (GNNs), to predict weather patterns. Here are the technical aspects that set GraphCast apart:</p><ol><li>Graph Neural Network Architecture: Unlike traditional grid-based models, GraphCast uses a graph-based representation of the atmosphere. This allows for more flexible and dynamic modeling of atmospheric processes, as the graph structure can adapt to represent complex interactions more naturally than rigid grid cells.</li><li>Learning from Historical Data: The model is trained on 40 years of rich historical weather data, allowing it to learn intricate patterns and dependencies in atmospheric dynamics. This deep learning approach enables GraphCast to make more accurate predictions, especially over longer forecast periods.</li><li>Handling Biases and Variances: GraphCast is adept at managing the biases and variances inherent in weather data. It adjusts its predictions based on the most recent and relevant information, a crucial feature in a field where data is constantly evolving.</li><li>Ensemble Forecasting Capability: One of the most promising aspects of GraphCast is its potential for ensemble forecasting. By generating multiple forecasts based on varied initial conditions, it can provide a more comprehensive range of possible weather outcomes, increasing the overall reliability of its predictions.</li></ol><p><strong>Model Architecture:</strong></p><p>GraphCast, a novel machine learning approach for global medium-range weather forecasting, combines a neural network architecture based on Graph Neural Networks (GNNs) with an autoregressive model for predicting weather states. It stands out for its ability to generate accurate 10-day forecasts in less than a minute on a single Google Cloud TPU v4 device. This model is particularly useful for predicting phenomena like tropical cyclone tracks, atmospheric rivers, and extreme temperatures.</p><p>The model inputs the two most recent states of Earth’s weather and predicts the weather six hours ahead. A single weather state is represented on a 0.25° latitude/longitude grid, with each grid point encapsulating various surface and atmospheric variables. GraphCast’s encoder maps these variables to learned node attributes on an internal “multi-mesh” representation, using a single GNN layer. This multi-mesh, derived from an iteratively refined icosahedron, is spatially homogeneous and has a high spatial resolution across the globe.</p><p>The processor of GraphCast uses 16 unshared GNN layers to perform efficient learned message-passing on the multi-mesh. The decoder then maps these learned features back to the latitude-longitude grid, predicting the output as a residual update to the most recent input state.</p><p>GraphCast’s training involved 39 years of historical data from ECMWF’s <a href="https://www.ecmwf.int/en/forecasts/dataset/ecmwf-reanalysis-v5">ERA5 reanalysis</a> archive. The training objective was to minimize the mean squared error (MSE), weighted by vertical level, between the predicted state and the corresponding ERA5 state over multiple autoregressive steps. This process incrementally increased the prediction window from six hours to three days during training.</p><p>The model architecture of GraphCast is comprehensive, consisting of 36.7 million parameters. It is efficient in its processing, requiring only about four weeks of training on 32 Cloud TPU v4 devices</p><p>GraphCast is a sophisticated machine learning model that leverages Graph Neural Networks (GNNs) in an “encode-process-decode” framework. This architecture is highly effective for simulating complex physical dynamics, such as fluid and material behaviors. GNNs are advantageous because they allow for arbitrary patterns of spatial interactions through learned message-passing, which is more flexible and scalable compared to the local or strided interactions in Convolutional Neural Networks (CNNs) or the all-to-all interactions in Transformers.</p><p>GraphCast introduces a unique “multi-mesh” representation (depicted in Figure 2) that enables efficient long-range interactions. This is achieved by dividing a regular icosahedron iteratively to create a hierarchy of meshes. This multi-scale approach, with coarse and fine edges, facilitates both local and long-distance interactions effectively.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/968/1*LrIDLacplCFUTFhEq1RCuQ.png" /><figcaption>Figure 2: Simultaneous multi-mesh message passing [1]</figcaption></figure><p>In GraphCast, the encoder first maps input data from a latitude-longitude grid to learned features on the multi-mesh. Then, a 16-layer deep GNN in the processor executes learned message-passing on the multi-mesh. Finally, the decoder maps the multi-mesh representation back to the latitude-longitude grid for output prediction. This architecture is not limited to regular grids and can be applied to various mesh-like state discretizations.</p><p>GraphCast demonstrates significant computational efficiency. On a single Cloud TPU v4 device, it can produce a high-resolution, 10-day weather forecast in under 60 seconds, which is notably faster than traditional systems like the ECMWF’s IFS system.</p><p>Overall, GraphCast’s GNN-based architecture, multi-mesh representation, and efficient processing make it a highly scalable and effective tool for simulating and predicting complex physical systems.</p><p><strong>Implications and Future Prospects</strong></p><p>The technical prowess of GraphCast has significant implications for the field of weather forecasting:</p><ul><li><strong>Reduced Computational Costs:</strong> By moving away from traditional supercomputing requirements, GraphCast can make weather forecasting more cost-effective and accessible.</li><li><strong>Enhanced Forecast Accuracy:</strong> The model’s ability to learn from vast datasets and adapt to new information can lead to more precise and reliable forecasts.</li><li><strong>Real-time Predictions:</strong> The increased efficiency of GraphCast allows for quicker generation of forecasts, essential for timely weather updates and emergency responses.</li><li><strong>Broader Research Applications:</strong> The model’s flexibility and scalability make it a valuable tool for broader climatological research, including studies on climate change and atmospheric science.</li></ul><h3>Conclusion</h3><p>GraphCast represents a significant leap forward in weather forecasting technology. By harnessing the power of graph neural networks, it offers a more efficient, accurate, and flexible approach to predicting weather patterns. As this technology continues to evolve, it promises to reshape not only how we forecast weather but also how we understand and respond to the atmospheric processes that govern our planet.</p><p><strong>References:</strong></p><p>[1] <a href="https://arxiv.org/pdf/2212.12794.pdf">GraphCast: Learning skillful medium-range global weather forecasting</a></p><p>[2] <a href="https://www.encyclopedie-environnement.org/en/air-en/weather-forecasting-models/">Weather forecasting models</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d70fae9ac365" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[BTS: The Marketing Masters in Disguise]]></title>
            <link>https://ramkrishna2910.medium.com/bts-the-marketing-masters-in-disguise-54dea60492e4?source=rss-d15f97e87633------2</link>
            <guid isPermaLink="false">https://medium.com/p/54dea60492e4</guid>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[music]]></category>
            <category><![CDATA[marketing-strategies]]></category>
            <category><![CDATA[bts]]></category>
            <category><![CDATA[community]]></category>
            <dc:creator><![CDATA[Ramakrishnan Sivakumar]]></dc:creator>
            <pubDate>Tue, 11 Jul 2023 21:39:07 GMT</pubDate>
            <atom:updated>2023-07-11T21:39:07.392Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vkSDH_eX9V7fdaiA5ubJxw.png" /></figure><p>During a trip to India last year, I was taken aback by my nieces’ immense excitement about attending an online BTS concert. What intrigued me further was the realization that they had saved up three months’ worth of their allowances to afford this experience. This surprising revelation sparked my curiosity, prompting me to delve deeper into the phenomenon that is <strong>BTS</strong>.</p><p>Picture this: BTS, a South Korean boy band that has become a global sensation, lighting up stages with their electrifying performances and distinctive musical style. They’re more than just gifted artists; they’re a testament to the power of strategic marketing fused with incredible talent. The world has been utterly captivated by their meteoric ascent, which has been nothing short of a symphony, perfectly orchestrated and brilliantly executed.</p><p>Now, let’s talk numbers. Since their debut in 2013, BTS has been cranking out hit after hit, generating a jaw-dropping ~$4 billion a year of economic impact in the process. And when the world paused due to the COVID-19 pandemic? They didn’t skip a beat. Their online concerts broke new ground, raking in an impressive $18 million in ticket sales. Adding to their collection of accolades, their single ‘Butter’ melted records, becoming the most-viewed video on YouTube in 24 hours and the most-streamed single on Spotify.</p><p>Impressive, right? But there’s more to their success than just eye-popping numbers and chart-topping hits. Dive a little deeper, and you’ll find the master key to their global domination — a marketing strategy so effective it’s a masterclass in itself. So buckle up, because we’re about to take a thrilling ride into the world of BTS, where music meets marketing magic.</p><h3>The Product</h3><p>The cornerstone of BTS’s marketing prowess is their product — their music. Their songs are not mere compilations of tunes and beats but narratives crafted with deeply meaningful lyrics. These narratives are then represented through highly engaging and visually appealing music videos. Their videos, with their distinctive color palette, diversified outfits, and cinematic quality, reflect the apex of production value. It’s this marriage of rich narratives and visual aesthetics that fuels their appeal.</p><h3>The Narrative</h3><p>BTS transcends the norm of pop music by interweaving compelling narratives within their songs, albums, and even eras. Each member of the band carries a unique narrative, forming an interwoven storyline that extends beyond the realm of music and into the domain of personal branding. They engage their audience in a Marvel-like fashion, leaving certain elements open to interpretation, sparking conversations, and creating a sense of communal interpretation.</p><h3>The Promotion</h3><p>BTS employs a promotional strategy that combines both traditional and contemporary elements. They tap into the power of anticipation with teasers, photos, and trailers. They smartly use product placements, especially in international markets, to boost their visibility and expand their reach. The band adopts varying promotional strategies for different age groups, and they’ve also showcased their versatility by collaborating with renowned US artists and featuring on American talk shows.</p><p>Moreover, in Korea, they promote their personality more than their music, appearing on variety shows and thereby adding a human element to their brand.</p><h3>The Price</h3><p>BTS has effectively leveraged the ‘foot in the door’ method to draw fans into their sphere of influence. They provide a slew of free or low-cost content across their massive social media presence on platforms like Twitter, Instagram, and Facebook. This free content, which is often sponsored by collaborations with high-profile brands like Louis Vuitton, Baskin Robbins, and McDonald’s, acts as a powerful engagement tool. The band also capitalizes on merchandise, launching products like keychains, t-shirts, and unique meet-and-greet cards with albums.</p><h3>The Process</h3><p>The BTS experience extends beyond the release of albums. They’ve turned the process of content consumption into an interactive journey. They encourage fans to become a part of their creative ecosystem, inviting them to join dance challenges based on their new albums, thereby driving engagement, spreading their music, and strengthening their fan community.</p><h3>The People</h3><p>Under the guidance of Hybe, which boasts a distributed leadership model across the largest K-pop markets including the USA, Japan, and Korea, BTS has constructed a robust global brand. With an army of 150–200 people working solely on content production, the BTS marketing machine is a behemoth that is incredibly challenging to replicate.</p><h3>The Psychological Factors</h3><p>Beyond their music and marketing strategies, BTS’s psychological connection with their fans is a key ingredient in their recipe for success. The band members share personal stories, vulnerabilities around mental health, and songwriting processes that allow fans to form deep psychological bonds with them. They position themselves not just as stars to be admired from a distance, but as companions offering solace, acting as a form of ‘life support’ to many fans.</p><p>BTS’s marketing strategy, a masterclass in building and sustaining a global brand, extends beyond the conventional. With their foray into virtual reality, their innovative approach continues to push boundaries, shaping the future of music marketing in unprecedented ways. They’ve cracked the code of global marketing strategy, proving that when music meets marketing.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=54dea60492e4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Power of Product Ecosystems]]></title>
            <link>https://medium.com/design-bootcamp/the-power-of-product-ecosystems-3f6afd41ca51?source=rss-d15f97e87633------2</link>
            <guid isPermaLink="false">https://medium.com/p/3f6afd41ca51</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[entertainment]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[product]]></category>
            <dc:creator><![CDATA[Ramakrishnan Sivakumar]]></dc:creator>
            <pubDate>Sun, 12 Feb 2023 03:55:51 GMT</pubDate>
            <atom:updated>2023-02-22T05:32:52.358Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W58EYhLjdeUEbyjkaPAaMA.png" /></figure><p>Nowadays, companies are fond of bundling together products and services to form an interconnected network of diverse applications. This approach has been shown to enhance user experience, but it also increases the reluctance to change multiple services and devices at once, effectively locking in customers. Creating such an ecosystem is no small feat and many companies have attempted and failed. It requires a great deal of persistence and hard work. This concept of making products work seamlessly with one another is not limited to the tech industry. In this article, we’ll examine how this approach is used across many industries and what its effects are.</p><p><strong>The Apple Ecosystem (Direct to Consumer (DTC)):</strong></p><p>The “Apple ecosystem” refers to the range of hardware and software products made by Apple Inc., all of which are designed to work together seamlessly. This includes the iPhone, iPad, Mac, Apple Watch, and Apple TV, as well as the operating systems that run on those devices (iOS, iPadOS, macOS, and tvOS) and the various services offered by Apple (such as the App Store, Apple Music, and iCloud). The term “ecosystem” is used to describe this interconnected network of Apple products and services because they all support and enhance one another, creating a unique and integrated experience for users. But on the other hand, Apple also closes the door on other companies attempting to integrate into this ecosystem by making standards and protocols proprietary.</p><p>The term “High Walled Garden” refers to a closed platform or ecosystem where a company exercises stringent control over the content and services accessible to users. It is frequently utilized in the context of Apple’s ecosystem, which is known for its stringent control over the products and services available to users. This control enables Apple to maintain a consistent and superior user experience across its devices, but it also restricts users in terms of their available options for products and services. Some individuals see this as a desirable aspect of the Apple ecosystem, while others consider it a constraint.</p><p>There are several advantages to using Apple’s ecosystem of products and services. One of the biggest advantages is the seamless integration and compatibility between all of Apple’s products. This means that if you have multiple Apple devices, they will work together seamlessly and allow you to access your content and information on any device. For example, you can start reading an eBook on your iPhone and then pick up where you left off on your iPad. Apple is known for making products that are intuitive, user-friendly, and designed with the user experience in mind. Additionally, the App Store offers a vast selection of apps and games, many of which are exclusive to the iOS platform.</p><p>Overall, the advantages of the Apple ecosystem include seamless integration and compatibility, a wide range of high-quality products and services, and strong security and privacy features.</p><p>There are a few potential disadvantages to the Apple ecosystem. One is that, because Apple tightly controls the products and services that are available to users on its platforms, users may have limited choices and may not be able to access certain content or services that are available on other platforms. Additionally, because Apple’s products are designed to work together seamlessly, users may feel locked into the Apple ecosystem and may find it difficult or expensive to switch to a different platform or ecosystem.</p><p>However, what’s even more intriguing is what this business model empowers the company to accomplish. Once a company has successfully established a product ecosystem with a substantial user base, any new feature it releases will immediately reach millions of people giving them unparalleled distribution and pricing control. This is something that would take another company decades to achieve. Furthermore, the company in control of the ecosystem doesn’t necessarily have to be the first to market with a new product or feature. Instead, it can observe and learn from the mistakes of early adopters, and then perfect the feature for release. It can also incorporate successful ideas from elsewhere, integrating them with its products to provide an enhanced user experience and outcompete its rivals. We’ve seen this happen time and time again with Apple.</p><p>For instance, in the past year,</p><p>Apple introduced <a href="https://developer.apple.com/apple-pay/whats-new/">Pay Later</a>, which directly competes with companies like Afterpay, Klarna, and Affirm, and caused a dip in their stocks after the announcement.</p><p><a href="https://support.apple.com/en-us/HT213244">Continuity camera</a> — Use your iPhone as your webcam, competes with Logitech and other camera vendors</p><p><a href="https://support.apple.com/guide/watch/view-heart-rate-zones-apd30fa26bb4/watchos">Apple Watch fitness zones</a> — Competes with gyms like Orange theory fitness.</p><p>The list goes on with Apple integrating apps such as Medication managers, Password managers into the well-oiled machinery and effectively crushing the competition.</p><p><strong>Disney — Marvel Ecosystem (Media and Entertainment)</strong></p><p>The Marvel Cinematic Universe (MCU) is a massive franchise of superhero films produced by Marvel Studios. It all started with the release of “Iron Man” in 2008, and since then the franchise has expanded to include over 20 films, television shows, and various other media. The films within the MCU are interconnected, sharing a continuity and featuring characters and storylines that cross over from one film to the next. This interconnectivity creates a shared universe, where events in one film can have significant implications for the characters in another. For example, the film “Captain America: Civil War” features characters from the “Avengers” films, and the events of “Thor: Ragnarok” have major impacts on the characters in “Avengers: Infinity War.”</p><p>This interconnected ecosystem of movies and TV shows has had a similar impact as Apple’s product ecosystem, where any new release immediately gains millions of viewers due to the references and connections to the larger universe. The success of the MCU has led to the launch of Disney+, a streaming service that has become one of the largest in the world with over ~160 million viewers. The Marvel Cinematic Universe is known for its complex and interwoven storytelling, which has helped make it one of the most successful film franchises of all time.</p><p><strong>The Microsoft Office Ecosystem (Business to Business (B2B) Technology)</strong></p><p>Microsoft has long been a dominant player in the productivity software market, and its ecosystem of products has helped it maintain a firm hold on the enterprise and office space. The company’s suite of productivity tools, including Office, Exchange, SharePoint, and OneDrive, are used by millions of people around the world, making it a go-to choice for businesses and organizations of all sizes.</p><p>Recently, Microsoft has made a bold move to strengthen its position in the market by offering Microsoft Teams, its popular collaboration and communication platform, as part of its Office Suite. The move aimed to compete with Slack, a popular collaboration tool that has gained significant traction in recent years. By bundling Teams with Office, Microsoft is making it easier for businesses and organizations to access a comprehensive suite of productivity tools, all in one place. With this bundle, Microsoft is hoping to lock in more customers and expand its market share in the productivity software market.</p><p>As a product aficionado, I can appreciate a seamless integration of products, but the downside of this kind of ecosystem is that it can limit creativity and squash new companies and ideas, especially when they have to go up against well-established entities with seemingly limitless resources and the added benefit of a powerful distribution network.</p><p>What do you think? Is it more important to have products work in harmony or to have a level playing field for all players? Or is there a balanced approach that can be taken?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3f6afd41ca51" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/the-power-of-product-ecosystems-3f6afd41ca51">The Power of Product Ecosystems</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[OpenAI’s ChatGPT introduces itself]]></title>
            <link>https://ramkrishna2910.medium.com/openais-chatgpt-introduces-itself-f612e92d9be5?source=rss-d15f97e87633------2</link>
            <guid isPermaLink="false">https://medium.com/p/f612e92d9be5</guid>
            <category><![CDATA[naturallanguageprocessing]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Ramakrishnan Sivakumar]]></dc:creator>
            <pubDate>Sun, 04 Dec 2022 22:51:04 GMT</pubDate>
            <atom:updated>2022-12-04T22:51:04.034Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*n5GgKt7Xr4CjojBs.jpg" /></figure><p>Image generation models such as Imagegen, and DALL.E-2 have become increasingly popular in the recent past, due to their ability to create highly realistic and detailed images that can be used in various applications.</p><p>One of the main reasons for the hype around image generation models is their potential for use in fields such as computer graphics, where they can be used to create photo-realistic images for movies, video games, and other visual media.</p><p>Breakthroughs with similar impact, if not more, have also been made in the field of natural language processing. ChatGPT is a recent announcement from OpenAI that is a large language model optimized for conversational dialogue. So instead of Googling to learn more about the model, I had a conversation with ChatGPT itself:</p><p><strong>Hi, this is Krishna. I am an ML software engineer.</strong></p><blockquote>Hi Krishna, it’s nice to meet you. I’m Assistant, a large language model trained by OpenAI. I’m here to help you with any questions you might have. Is there anything specific you’d like to talk about?</blockquote><p><strong>What is ChatGPT?</strong></p><blockquote>ChatGPT is a natural language generation model developed by OpenAI. It’s a variant of their popular GPT-3 model, which was trained on a large dataset of human conversation. ChatGPT is designed to be able to generate human-like responses to text input in a variety of different styles and formats. It can be used for a variety of applications, including chatbots, virtual assistants, and language translation.</blockquote><p><strong>What technology is ChatGPT based on?</strong></p><blockquote>ChatGPT is built on the same technology as GPT-3, which is a type of artificial neural network called a transformer. This is a deep learning model that’s able to process input text and generate output text by learning to predict the next word in a sequence. In the case of ChatGPT, the model has been trained on a large dataset of human conversation, so it’s able to generate responses that sound more natural and human-like than previous models.</blockquote><p><strong>Can you explain how GPT-3 works?</strong></p><blockquote>Sure, I’d be happy to explain how GPT-3 works. GPT-3 is a state-of-the-art natural language processing model that uses deep learning algorithms to generate human-like text. It is trained on a massive amount of data, which allows it to generate coherent and plausible responses to a wide range of inputs.</blockquote><blockquote>To generate text, GPT-3 takes in a prompt, which could be a sentence or a paragraph, and uses its deep learning algorithms to generate a response that is relevant to the input. The response is generated using a process called transformers, which allows the model to process the input and generate text that is similar to human writing.</blockquote><p><strong>How is GPT-3 different from GPT-2?</strong></p><blockquote>GPT-3 (Generative Pretrained Transformer 3) is the third generation of the GPT (Generative Pretrained Transformer) language model developed by OpenAI. It is an improved version of GPT-2, with a larger and more powerful model architecture that allows it to generate more accurate and relevant responses to a wide range of inputs.</blockquote><blockquote>Some of the key differences between GPT-3 and GPT-2 include:</blockquote><blockquote>GPT-3 is significantly larger and more powerful than GPT-2. It has 175 billion parameters, which is over 10 times more than GPT-2, and can generate more accurate and relevant responses to a wider range of inputs.</blockquote><blockquote>GPT-3 uses a more advanced model architecture than GPT-2, which allows it to better capture the nuances and complexities of natural language. This makes it better at generating human-like text that is coherent and relevant to the input.</blockquote><blockquote>GPT-3 can be fine-tuned for specific tasks or domains, which allows it to generate even more accurate and relevant responses. This makes it a versatile tool that can be used for a variety of natural language processing tasks, such as language translation, question answering, and text generation.</blockquote><blockquote>Overall, GPT-3 represents a significant advance over GPT-2 in terms of model size, architecture, and capabilities. It is a powerful tool for natural language processing tasks and has proven to be a valuable asset for many different applications.</blockquote><p><strong>What is the future of natural language processing?</strong></p><blockquote>The future of natural language processing is likely to involve continued advances in deep learning algorithms and the development of even more powerful language models. These advances will enable natural language processing systems to better understand and generate human-like text, which will open up new possibilities for a wide range of applications.</blockquote><blockquote>One of the key areas of focus for future natural language processing research will be the development of more advanced model architectures that can better capture the nuances and complexities of human language. This will allow natural language processing systems to generate more accurate and relevant responses to a wider range of inputs, and will enable them to be used for more complex tasks.</blockquote><blockquote>Another important area of research will be the development of natural language processing systems that are able to learn and adapt over time. These systems will be able to continually improve their performance by learning from new data and experiences, which will allow them to better serve the needs of users.</blockquote><blockquote>Overall, the future of natural language processing is likely to involve significant advances in deep learning algorithms and the development of more powerful and adaptive language models. These advances will enable natural language processing systems to better understand and generate human-like text, which will open up new possibilities for a wide range of applications.</blockquote><p>It is fascinating to see how far generative models have come over the last few years. Given ChatGPT is able to handle questions of many formats, how does it stand up to our traditional way of answering questions?</p><p>I asked the same question to both Google and ChatGPT.</p><blockquote><strong>“How to find the index of an array where the product is greater than a target value in python?”</strong></blockquote><p>On the left you see Google listing the top search results for the query, where I am likely to find the right answer in the first few links. Whereas ChatGPT the right, displays the answer that I am looking for with an example and explanation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sZi6gagwuYXT_S7wHlYDqg.png" /></figure><p><strong>Is this the future of search?</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f612e92d9be5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What, Why and How ONNX-script?]]></title>
            <link>https://ramkrishna2910.medium.com/what-why-and-how-onnx-script-74dd21ab396f?source=rss-d15f97e87633------2</link>
            <guid isPermaLink="false">https://medium.com/p/74dd21ab396f</guid>
            <category><![CDATA[inference]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[pytorch]]></category>
            <category><![CDATA[ai-framework]]></category>
            <category><![CDATA[onnx]]></category>
            <dc:creator><![CDATA[Ramakrishnan Sivakumar]]></dc:creator>
            <pubDate>Mon, 10 Oct 2022 02:56:50 GMT</pubDate>
            <atom:updated>2022-10-10T22:34:52.439Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>What, Why and How ONNX Script?</strong></h3><p><strong>ONNX (Open Neural Network eXchange) </strong>is an open-source standard that defines a common set of operators and a file format that enables AI developers to use models with various frameworks, tools, runtimes, and compilers. For a deeper dive into ONNX check this <a href="https://medium.com/@ramkrishna2910/the-promise-of-onnx-d43c371f913d">post</a>!</p><p>In this article, I am excited to share some thoughts on @Microsoft’s recent announcement on ONNX Script and what it enables developers to do.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QoIilg4tdJGvocGLU7Hblw.png" /></figure><p><a href="https://github.com/microsoft/onnx-script">ONNX Script</a> enables developers to express ONNX functions and models in Python. This simple yet powerful expression of models/ functions allows for <a href="https://towardsdatascience.com/eager-execution-vs-graph-execution-which-is-better-38162ea4dbf6">eager mode evaluation</a> thus significantly improving debug ability.</p><p>ONNX Script provides developers with the following capabilities:</p><ul><li>A converter to translate a Python ONNX Script function into an ONNX graph.</li><li>A converter that translates ONNX models and functions into ONNX Script. This capability can be used to fully round-trip.</li><li>A runtime(ONNXRuntime) that allows eager mode execution of the model.</li></ul><p>For those familiar with the torch ecosystem, TorchScript provides a way to create serializable and optimizable models from PyTorch code that can be saved from a Python process and loaded in a process where there is no Python dependency. This makes it possible to train models in PyTorch using familiar tools in Python and then export the model via TorchScript to a production environment where Python programs may be disadvantageous for performance and multi-threading reasons.</p><p>ONNX Script essentially provides the Python-based front end to the existing ONNX infrastructure and also the tools to translate the script to serialized ONNX graphs that can be deployed without dependencies.</p><p><strong>Why should you pay attention to ONNX Script?</strong></p><p>ONNX models were traditionally obtained by exporting models built in different source frameworks such as PyTorch and TensorFlow using conversion tools provided by respective frameworks. This has allowed developers to stick to their existing development flow in PyTorch/ TensorFlow and then convert the models to ONNX for deployment to take advantage of the flexibility ONNX provides.</p><p>But the downside to this approach is the issues faced during model conversion from the source framework to ONNX. This is due to a mismatch in operator coverage surface area between the source frameworks(PyTorch/ TensorFlow) and ONNX.</p><p>To solve this issue, instead of brute forcing a 1:1 operator match with the source frameworks, ONNX approaches this problem by breaking down unsupported complex operations into their fundamental components and expressing them in the decomposed form of simpler existing operators through ONNX functions.</p><p><strong>ONNX Script </strong>provides a new way to build, understand and debug complex ONNX functions/models in a way that developers are already used to, which will reduce the inertia in adoption.</p><p>This enables Domain-Specific Hardware accelerators and compiler developers like <a href="https://groq.com/">Groq</a> to focus on a tight set of operations and represent the broader ecosystem of models using this optimized set of fundamental operations.</p><p><strong>How can you use ONNX-script?</strong></p><p>The following are a few ways you can start using ONNX script.</p><ol><li><strong>Represent complex operations using existing ONNX primitives</strong></li></ol><p>The following example shows how a Gaussian Error Linear Units (GELU) activation can be represented using ONNX Script.</p><p>The GELU nonlinearity weights inputs by their percentile, rather than gates inputs by their sign as in <a href="https://paperswithcode.com/method/relu">ReLUs</a>. GELU is mathematically expressed as follows:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7gDZBl-Gb6NWkB41J7tQ9g.png" /><figcaption>GELU — Gaussian Error Linear Units</figcaption></figure><p>Using the above mathematical expression GELU can be represented in ONNX Script as follows</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/8b06ef71aa83dd080f209062324ad8c6/href">https://medium.com/media/8b06ef71aa83dd080f209062324ad8c6/href</a></iframe><p>Explore more examples <a href="https://github.com/microsoft/onnx-script/tree/main/docs/tutorial/examples">here</a>!</p><p>2. <strong>Author ONNX models that can be exported for training/inference.</strong></p><p>ONNX Script enables developers to write entire models in Python and export them into an ONNX model using the following:</p><pre>from onnxscript import script<br>from onnxscript.onnx_types import FLOAT<br>import onnx</pre><pre>@script()<br>def model(input: FLOAT[...]):<br>    ....<br>    ....<br>    ....</pre><pre>model_proto = model.to_model_proto()<br>onnx.save(model_proto, &quot;path/model.onnx&quot;)</pre><p>Note: Currently a function with attributes cannot be exported as a model i.e. the function cannot take arbitrary variables that can be used inside the function.</p><p><strong>3. Potentially replace operations in source frameworks that block conversion to ONNX</strong></p><p>For example, currently, PyTorch models that use torch.bucketize like <a href="https://pytorch-geometric.readthedocs.io/en/latest/_modules/torch_geometric/nn/conv/supergat_conv.html#SuperGATConv">SuperGATConv</a> from PyTorch Geometric are blocked from being converted to ONNX due to the lack of an equivalent operator on the ONNX. ONNX Script can be used to potentially resolve this as follows:</p><pre># pytorch-model.py</pre><pre>import torch<br>import torch.nn as nn<br>from onnxscript import script<br>from onnxscript.onnx_types import FLOAT</pre><pre>class model(nn.Module):<br>    def __init__(self):<br>        super(model, self).__init__()</pre><pre>    def forward(self):<br>        ....<br>        ....<br>        torch.bucketize(...)<br>        ....</pre><pre>@script()<br>def onnx_bucketize(input: FLOAT[...]):<br>    ....<br>    ....<br>    ....</pre><pre># Replace torch.bucketize(...) with onnx_bucketize(...)<br>torch_model = model(...)<br>torch.onnx.export(torch_model, inputs, &quot;model.onnx&quot;)</pre><p>ONNX Script has the potential to significantly improve the developer experience in the ONNX ecosystem. Though this is still the early days of ONNX Script, I am personally excited to see bridges this builds/ strengthens in this fragmented world of AI frameworks!!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=74dd21ab396f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Terrible design — Buick Envision]]></title>
            <link>https://ramkrishna2910.medium.com/terrible-design-buick-envision-25841d430efa?source=rss-d15f97e87633------2</link>
            <guid isPermaLink="false">https://medium.com/p/25841d430efa</guid>
            <category><![CDATA[design-thinking]]></category>
            <category><![CDATA[automobile]]></category>
            <category><![CDATA[cars]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[design]]></category>
            <dc:creator><![CDATA[Ramakrishnan Sivakumar]]></dc:creator>
            <pubDate>Wed, 03 Aug 2022 20:20:55 GMT</pubDate>
            <atom:updated>2023-01-18T16:48:50.854Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Buick Envision — Aesthetic &gt;&gt; Usability 🤷‍♂️</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*YacFILEpkNYZnf6CeNYY9Q.jpeg" /><figcaption>Buick Envision</figcaption></figure><p>As an engineer who gets excited by the thoughtful design choices of products, I get equally disappointed by terrible design choices. This is just me sharing my thoughts on my random observations to spark a conversation and possibly try and understand the thought process behind design decisions.</p><p>In today’s edition of terrible design, we will look at the Buick Envision. A Compact SUV with 228HP.</p><p>I had an opportunity to rent this car with Hertz during my trip to Denver and the following are a couple of design decisions in the interior of the vehicle that caught my attention.</p><p><strong>Air conditioner controls</strong></p><p>Traditionally, vehicle air conditioners were controlled using physical knobs that you can turn clockwise/ anti-clockwise or distinct physical buttons that can be pressed to increase or decrease temperature. Having such physical knobs or buttons enabled quickly locating them when driving without taking your eyes off the road and being reliably able to turn or press a certain number of times to get the desired temperature.</p><p>With the evolution of interior design in vehicles, manufacturers have started taking a more minimal approach and thus reducing the number of physical buttons in the car. This has moved these physical controls into a touchscreen that can house multiple controls under various settings pages just like a smartphone can. The jury is still out on if this is a good design choice and that is a discussion for another time.</p><p>Now we come to Buick Envision’s interior design, it’s a classic design like any other late 2010 vehicle with a 10-inch touch screen and knobs for volume control and a series of buttons to change the mode of climate control. This in no way is a minimal design. But that design language seems to change just for a subset of controls, namely the temperature controls and seat heaters. These buttons are placed awkwardly towards the lowest portion of the dash which makes it harder to reach and are also entirely touch-based controls with no physical guidance. This makes it impossible to change the temperature without taking your eyes off the road. And to top it off these are weak semi-capacitive touch controls that would not register a touch until hit perfectly. Over the 3 days, I drove this car I had to fiddle with this at least a few times to change the temperature by 1F.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NusA7ze_hAK3_-SB_AZt8w.jpeg" /></figure><p><em>Why though?? Is this “modernization” at the cost of usability🤷‍♂️?</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=25841d430efa" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Promise of ONNX]]></title>
            <link>https://ramkrishna2910.medium.com/the-promise-of-onnx-d43c371f913d?source=rss-d15f97e87633------2</link>
            <guid isPermaLink="false">https://medium.com/p/d43c371f913d</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[onnx]]></category>
            <category><![CDATA[pytorch]]></category>
            <category><![CDATA[ml-so-good]]></category>
            <dc:creator><![CDATA[Ramakrishnan Sivakumar]]></dc:creator>
            <pubDate>Sat, 30 Jul 2022 22:43:29 GMT</pubDate>
            <atom:updated>2023-02-22T07:44:46.182Z</atom:updated>
            <content:encoded><![CDATA[<p>A Machine Learning framework is a tool that enables a developer to create models to achieve a given task such as image classification or natural language processing. Some of the most familiar examples of ML frameworks are PyTorch and TensorFlow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/584/1*FpSVm3lUHxuu0zjc8zqGhA.png" /></figure><p><strong>ONNX (Open Neural Network eXchange) </strong>is an open-source standard that defines a common set of operators and a file format that enables AI developers to use models with a variety of frameworks, tools, runtimes, and compilers.</p><blockquote><strong>In this post I will talk about:</strong></blockquote><blockquote>Why was ONNX created?</blockquote><blockquote>What is ONNX?</blockquote><blockquote>The Promise of ONNX</blockquote><blockquote>Fulfilment of the Promise</blockquote><h3><strong>Why was ONNX created?</strong></h3><p>The advent of Machine learning/ Deep learning caused an exponential increase in the number of frameworks/ libraries that were introduced to handle the execution of deep learning/ machine learning models.</p><p>This increase was primarily driven by,</p><p>(a) Hardware vendors who developed their own frameworks to achieve vertical integration and make it easier to develop models for their hardware</p><p>and</p><p>(b) Software vendors who developed their own frameworks to achieve the best performance for specific use cases</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J44RfZaVpdIst8zia_J83g.png" /></figure><p>This explosion in the number of available frameworks to develop deep learning models led to the fragmentation of the ecosystem and made working across frameworks or hardware ecosystems difficult. A choice had to be made on where the models would be developed and trained and moving to a different framework or a different hardware ecosystem later in the data science pipeline was near impossible. This was due to the compatibility of frameworks with compilers for different hardware but we will save the compiler story for another time.</p><p>ONNX was born to remove this inertia in interoperability between frameworks and between different hardware ecosystems.</p><p>Throughout the rest of this article, we will explore if the solution to this problem of too many frameworks is to introduce another framework into the mix?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/938/1*h4EWQXEzuoTQOhCk1bchHQ.png" /></figure><h3>What is ONNX?</h3><p>ONNX is an open specification that consists of definitions for the following:</p><h4><strong>File format for a computational graph model</strong></h4><p><strong>Model:</strong> Is an Acyclic computation data flow graph that holds version information and metadata.</p><p><strong>Graph: </strong>Contains the name of the data flow graph in the model, a list of computation nodes that form the graph, inputs and outputs to the graph.</p><p><strong>Computation Node: </strong>Each computation node in the graph that takes zero or more inputs of <a href="https://github.com/onnx/onnx/blob/main/docs/IR.md#standard-data-types">defined types</a>, one or more outputs of defined types and the type of operation the specific node performs on its inputs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/434/1*BJTLzXm6m8FrhvjKBJBJyQ.png" /><figcaption>Figure 2: An example ONNX graph</figcaption></figure><h4>Standard data types</h4><p>ONNX as a standard supports the following list of datatypes:<br><strong>Tensor type:</strong></p><ul><li>Int8, Int16, Int32, Int64</li><li>Quantized Int</li><li>Uint8, uint16, uint32, uint64</li><li>Float16, float, double</li><li>Bool</li><li>String</li><li>Complex64, complex128</li></ul><p><strong>Non-tensor types:</strong></p><ul><li>Sequence</li><li>Map</li><li>Operators (Built-in/ Custom)</li></ul><h4>Operators</h4><p>Each computation node in the ONNX Graph performs a specific operation and produces one or more outputs. The ONNX standard defines the official list of <a href="https://github.com/onnx/onnx/blob/main/docs/Operators.md">operators</a> that are supported by ONNX graphs. This list of operators is constantly evolving and is kept up to date through ONNX Opsets.</p><p>Each incremental ONNX Opset adds support for new operators and improves the existing list of operators.</p><p>This specification is meant to be an open standard that everyone can agree upon and hence enable translation from any framework to ONNX.</p><h3>The Promise of ONNX</h3><p>The core design philosophies of ONNX are:</p><ul><li>Interoperability</li><li>Portability</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4GN3Dq6UKL0b0aBOgs0Bog.png" /><figcaption>Figure 3: The Promise of ONNX</figcaption></figure><p>ONNX intends to be this intermediary between different frameworks by providing an efficient interface to translate models from one framework to another. <em>Figure 4 </em>below lists the tools that convert models to ONNX format from some of the most common frameworks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*lLBgb7LoUAazJ7hJ6vw2aQ.png" /><figcaption>Figure 4: Onnx Conversion support</figcaption></figure><h3>The Fulfilment of the Promise</h3><p>In reality, the conversion of models from any framework to ONNX has been a challenge. One of the primary hurdles is the difference in surface area of these frameworks.</p><p>The challenge for the conversion libraries is keeping up with the shifting landscapes of both ONNX and their source libraries for every release (~4–6 times a year)</p><p>Conversion of models from one framework to another boils down to being able to represent the underlying math operations from the original model. Figure 5 shows the number of unique operators defined in each framework. ONNX supports ~13% of the operators supported by PyTorch. This significantly limits the number of models that can be actually converted to ONNX.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*U3mjXr7S56_6_o3mCmdkOw.png" /><figcaption>Figure 5: Number of operators per framework</figcaption></figure><p>It is important to note that 1000+ operators on frameworks like PyTorch also include a degree of duplication in the functionality in specific ops and also begs the question, should frameworks like ONNX support the global superset of all operations?</p><p>To solve this issue of operator coverage ONNX has adopted a different strategy rather than chasing the never-ending list of operators on all these frameworks.</p><p><strong>ONNX Functions </strong>help express complex operations using primitive/ existing operators in ONNX. Every upcoming ONNX release is going to be adding new functions and it is easier than ever to <a href="https://github.com/onnx/onnx/blob/main/docs/AddNewOp.md#step1_new_operator_or_function">add your own functions as needed</a>. This helps limit the number of operators that need to be maintained and also benefits compiler developers by reducing the scope and effort involved to optimize a smaller number of operators for their hardware.</p><blockquote>Operators are the basic building blocks used to define ONNX models. With a rich set of operators, ONNX can describe most DNN and ML models from various frameworks. <strong>Functions enable expressing complex operators in terms of more primitive operators</strong>. The ONNX specification includes a core set of operators that enable many models. It is a non-goal to add all possible operators, however more operators are added as needed to cover evolving needs.</blockquote><p>I cannot overstate the resemblance of the above with the <strong>CISC Vs RISC</strong> wars of the previous era. We not only have the front row seats to see how this plays out but also the opportunity to truly influence the direction of this impactful industry that is already shaping the future of technology by contributing to open source projects and pushing the boundaries of the AI software.</p><p>What a time to be alive!!!</p><h4><a href="https://mlearning.substack.com/about"><strong>BECOME a WRITER at MLearning.ai</strong></a></h4><p><a href="https://medium.com/mlearning-ai/mlearning-ai-submission-suggestions-b51e2b130bfb">Mlearning.ai Submission Suggestions</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d43c371f913d" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>