<?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 Sandeep Chinchali on Medium]]></title>
        <description><![CDATA[Stories by Sandeep Chinchali on Medium]]></description>
        <link>https://medium.com/@sandeepchinchali?source=rss-19b39523c706------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*OHzw6gT3wdq0IJXvHEPxiQ.png</url>
            <title>Stories by Sandeep Chinchali on Medium</title>
            <link>https://medium.com/@sandeepchinchali?source=rss-19b39523c706------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 18:40:23 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@sandeepchinchali/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[How to communicate scientific experiments]]></title>
            <link>https://sandeepchinchali.medium.com/how-to-communicate-scientific-experiments-ac5d47f20aa5?source=rss-19b39523c706------2</link>
            <guid isPermaLink="false">https://medium.com/p/ac5d47f20aa5</guid>
            <category><![CDATA[scientific-research]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[scientific-method]]></category>
            <category><![CDATA[technical-writing]]></category>
            <dc:creator><![CDATA[Sandeep Chinchali]]></dc:creator>
            <pubDate>Sun, 28 Sep 2025 05:20:05 GMT</pubDate>
            <atom:updated>2025-09-28T05:21:10.188Z</atom:updated>
            <content:encoded><![CDATA[<h3>Overview</h3><p>This guide is for those in industry and academia to plan machine learning (ML) experiments and communicate them effectively. I hope you can save countless hours in meetings and debugging by<em> planning first, coding later </em>and getting actional feedback from peers<em>.</em></p><p>Below, I provide a running example for semantic search of time series, like the products developed by my students’ startup <a href="https://medium.com/@synthefy">Synthefy</a>. If you find this self-explanatory, please skim to the template at the end.</p><h3>Example</h3><h4><strong>Objective:</strong></h4><p>We want to build a platform to search time-series data and synthesize realistic variants using natural language.</p><h4><strong>Experiment 1: Are off-the-shelf timeseries embeddings sufficient for vector search?</strong></h4><h4><strong>Hypothesis:</strong></h4><p>Given a query time-series, we can find its embedding using various off-the-shelf embedding vectors (e.g., Meta Kats, CLAP). Then, we can reliably find the closest time-series in terms of cosine similarity using a standard vector database (e.g., Pinecone).</p><h4><strong>Why it Should Work:</strong></h4><p>See this <a href="https://docs.pinecone.io/docs/time-series">example from Pinecone.</a> Clearly, it works on the Kaggle stocks dataset.</p><h4><strong>Required Background Knowledge:</strong></h4><p>Understanding of a vector database and time series embeddings.</p><h4><strong>Experimental Terminology:</strong></h4><p>First, let us define key terminology.</p><ol><li><strong>Query Time-Series: </strong>This is the time series I want to search for. This can be a sine wave with specific parameters (amplitude, frequency, phase) or a real time-series. I won’t add any noise to it.</li><li><strong>Time Series Database: </strong>This is the set of time-series from which we want to retrieve the most similar example to the query. We store their embeddings using a vector database, such as Pinecone.</li><li><strong>Embedding Process: </strong>This is a code parameter that can toggle between Kats, CLIP, or a custom embedding that we create.</li></ol><p><em>Key Idea: </em>Suppose I have a query time-series. Then, I successively add more noise to it and store these variants in the time series database. The distance between the noise-free query and the variants with the lowest noise <em>should be</em> minimized. This means I can search for the needle (query time-series with zero noise) in the haystack (set of noisy variants in the database).</p><h4><strong>Anticipated Subtleties and Pitfalls:</strong></h4><ol><li>Embedding space is not sufficiently descriptive. To solve this, I will switch between Kats, CLAP, or conclude I need my own embedding.</li><li>We need a diversity of examples, not just sine waves. I may need ARIMA time-series, other parametrized toy examples, and finally real data.</li><li>The noise needs to be carefully plotted to not “over-power” the original signal. Finding this trade-off must be hard.</li></ol><h4><strong>Algorithm:</strong></h4><p><strong>Inputs:</strong> The inputs are a specific embedding process, parameterized set of time-series (e.g., sine), a query time-series, and a database.</p><p><strong>Output:</strong> The output is a ranked list of the closest time series to the query.</p><p><strong>Steps:</strong></p><ol><li>Pick a sine wave of a specific amplitude, phase, frequency.</li><li>Step over noise levels. The noise could affect the raw time-series, phase, or frequency, but the easiest is additive noise.</li><li>For each noise level, create 100 variants for statistical significance.</li><li>Store all the variants in the time series database.</li><li>Compute the $L_2$ norm embedding distance or cosine similarity between the query and time series database.</li><li>Repeat for variations in the query and the way we create variants for statistical confidence <strong><em>(get creative here!)</em></strong>. I do not know the precise answer.</li></ol><h4><strong>Desired Plot:</strong></h4><h4>Key Trend We Want to Confirm:</h4><p>As the noise level increases, the median embedding distance between the noise-free query and all vectors in the database increases.</p><p>For real data, we visually see that the 5–10 “closest” neighbors have a similar trend as the query.</p><p><strong>X-axis: </strong>The noise level.</p><p><strong>Y-axis: </strong>The $L_2$ embedding distance.</p><p><strong>Type of Plot: </strong>Line-plot with standard deviation.</p><p><strong>Legend: State the series and colors in the legend.</strong></p><p><strong>Color Scheme:</strong> State how you use uniform colors (e.g., your algorithm is always orange) and why.</p><p><strong>Actual Plot:</strong></p><p>Use the <a href="https://github.com/UTAustin-SwarmLab/Swarm-Visualization">SwarmViz</a> utilities.</p><h4><strong>Did it work?</strong></h4><p>Tell me in one sentence whether the plot confirmed the hypothesis and state why, ideally with a quantitative number or clear visual trend. If you state the FID score is .05, how should I judge if this number is good?</p><h4><strong>How is this statistically significant?</strong></h4><p>Show the variance across multiple seeds and draw error bars if possible. Also, repeat for different datasets. How am I sure this is not a fluke result that will take us on a wild goose chase for months?</p><h4><strong>Next Idea:</strong></h4><p>In one sentence, tell me what you will logically do next and why?</p><h3>Template</h3><p>Fill in the template concisely (ideally 1–2 complete sentences per section).</p><h4>Objective</h4><h4>Experiment Question</h4><h4>Hypothesis</h4><h4>Why it Should Work</h4><h4>Required Background Knowledge</h4><h4>Experimental Terminology</h4><h4>Anticipated Subtleties and Pitfalls</h4><h4>Algorithm</h4><p><strong>Inputs:</strong></p><p><strong>Output:</strong></p><p><strong>Steps:</strong></p><h4>Desired Plot (Fill this in BEFORE doing an experiment)</h4><p><strong>Key Trend We Want to Confirm:</strong></p><p><strong>X-axis:</strong></p><p><strong>Y-axis:</strong></p><p><strong>Type of Plot:</strong></p><p><strong>Legend:</strong></p><p><strong>Color Scheme:</strong></p><p><strong>Actual Plot:</strong></p><h4>Did it work?</h4><h4>How is this statistically significant?</h4><h4>Next Idea</h4><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ac5d47f20aa5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Tokenomics for Decentralized AI and DePIN]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://sandeepchinchali.medium.com/tokenomics-for-decentralized-ai-and-depin-e9deaa9bc800?source=rss-19b39523c706------2"><img src="https://cdn-images-1.medium.com/max/960/0*wqD5jBAcYn_vrMpp" width="960"></a></p><p class="medium-feed-snippet">Why Decentralized Physical Infrastructure Networks Make a Ton of Sense</p><p class="medium-feed-link"><a href="https://sandeepchinchali.medium.com/tokenomics-for-decentralized-ai-and-depin-e9deaa9bc800?source=rss-19b39523c706------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://sandeepchinchali.medium.com/tokenomics-for-decentralized-ai-and-depin-e9deaa9bc800?source=rss-19b39523c706------2</link>
            <guid isPermaLink="false">https://medium.com/p/e9deaa9bc800</guid>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[decentralized-finance]]></category>
            <category><![CDATA[tokenomics]]></category>
            <category><![CDATA[blockchain]]></category>
            <dc:creator><![CDATA[Sandeep Chinchali]]></dc:creator>
            <pubDate>Sat, 18 Jan 2025 03:52:45 GMT</pubDate>
            <atom:updated>2025-01-18T03:52:45.677Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[AI Meets Crypto: Hype or Hope? A Professor’s View]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://sandeepchinchali.medium.com/ai-meets-crypto-hype-or-hope-a-professors-view-7a0c1cc8c1e6?source=rss-19b39523c706------2"><img src="https://cdn-images-1.medium.com/max/1024/0*IhrmrqeyiZKKtSxb" width="1024"></a></p><p class="medium-feed-snippet">As a professor, I&#x2019;m often skeptical when two trending topics are mashed together and labeled revolutionary. Too frequently, this leads to&#x2026;</p><p class="medium-feed-link"><a href="https://sandeepchinchali.medium.com/ai-meets-crypto-hype-or-hope-a-professors-view-7a0c1cc8c1e6?source=rss-19b39523c706------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://sandeepchinchali.medium.com/ai-meets-crypto-hype-or-hope-a-professors-view-7a0c1cc8c1e6?source=rss-19b39523c706------2</link>
            <guid isPermaLink="false">https://medium.com/p/7a0c1cc8c1e6</guid>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[decentralized-ai]]></category>
            <category><![CDATA[decentralized-finance]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <dc:creator><![CDATA[Sandeep Chinchali]]></dc:creator>
            <pubDate>Wed, 08 Jan 2025 04:10:57 GMT</pubDate>
            <atom:updated>2025-01-08T04:10:57.808Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The 4 C’s of Generative AI: Where It Works and Where It Falters]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-snippet">Generative AI is reshaping industries by automating, assisting, and accelerating tasks that once seemed untouchable by machines. However&#x2026;</p><p class="medium-feed-link"><a href="https://sandeepchinchali.medium.com/the-4-cs-of-generative-ai-where-it-works-and-where-it-falters-e9de4f581972?source=rss-19b39523c706------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://sandeepchinchali.medium.com/the-4-cs-of-generative-ai-where-it-works-and-where-it-falters-e9de4f581972?source=rss-19b39523c706------2</link>
            <guid isPermaLink="false">https://medium.com/p/e9de4f581972</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-copilot]]></category>
            <category><![CDATA[coding]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <dc:creator><![CDATA[Sandeep Chinchali]]></dc:creator>
            <pubDate>Tue, 07 Jan 2025 01:51:29 GMT</pubDate>
            <atom:updated>2025-01-07T01:51:29.751Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Bittensor and Decentralized AI]]></title>
            <link>https://sandeepchinchali.medium.com/decentralized-ai-and-bittensor-an-academics-view-b0ad811d2728?source=rss-19b39523c706------2</link>
            <guid isPermaLink="false">https://medium.com/p/b0ad811d2728</guid>
            <category><![CDATA[bittensor]]></category>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[generative-ai]]></category>
            <category><![CDATA[bittensor-tao]]></category>
            <dc:creator><![CDATA[Sandeep Chinchali]]></dc:creator>
            <pubDate>Mon, 19 Feb 2024 04:52:41 GMT</pubDate>
            <atom:updated>2024-02-20T03:22:05.485Z</atom:updated>
            <content:encoded><![CDATA[<p>We’re starting to see significant interest in blockchain-based decentralized AI systems. In this post, we analyze the dynamics of a widely popular project, Bittensor, and provide guidelines on when/whether it makes financial sense to join the ecosystem.</p><h3><strong>Bittensor in a Brief Nutshell</strong></h3><p>Imagine a world where anyone with a GPU or CPU can compete in a marketplace to train the best AI models for a diverse set of socially-beneficial tasks, such as predicting the stock market, detecting cancer from radiology scans, or training more human-aligned LLMs. Such a world would be possible if the world’s compute owners could be securely and transparently rewarded based on the quality of the AI models they produce. This is precisely the world Bittensor is enabling through the following key agents, all tied together in a blockchain-based incentive system:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K5nEMpJd3wqO8h5zy5u2TA.png" /></figure><p><em>Subnet Owners and Creators:</em></p><p>Subnets are groups of machines working to achieve a common AI task, such as labeling data, predicting the stock market etc. In theory, subnets can be created by anyone in the community, and are vetted based on their popularity and utility of the task they propose. To create and maintain a subnet, the owners are rewarded with a native cryptocurrency named $TAO.</p><p><em>Miners:</em></p><p>Miners are compute nodes that compete to train the best AI models. Miners are rewarded based on the quality of their AI predictions. Ideally, this quality should be assessed in a quick, transparent manner by another set of machines, such as the error of stock market predictions. Miners must pay a small registration fee, in $TAO, to compete in the ecosystem. Poor performing miners are periodically booted from the ecosystem and refunded their $TAO deposit.</p><p><em>Validators:</em></p><p>A distributed set of validators check the accuracy of AI outputs from miners. For example, validators can obtain the current value of a stock and compare it to miners’ predictions, assign an error (loss function), and distribute rewards to miners based on their respective errors. Since validators play a crucial role in assessing rewards and maintaining the integrity of the system, they must stake a significant amount of $TAO as a deposit. This $TAO can be slashed if they provide erroneous validation results. In return, validators are rewarded $TAO for maintaining the integrity of the decentralized AI ecosystem.</p><h3><strong>Should I Join the Ecosystem?</strong></h3><p>Suppose I have some combination of significant AI expertise, spare compute resources, and/or happen to already have a large amount of $TAO. Where should I dedicate these resources?</p><p>The answer is a bit surprising once you look at the data. For example, take a snapshot of time in taostats.io, a popular analytics tool for Bittensor. We see that a few “whale” validators who have staked a huge amount of $TAO get a large number of rewards as emissions, around $40K per day at the time of writing. Indeed, the wallets making the most money are typically all validators. In comparison, the best performing miner is ranked 66 amongst all top-earning wallets and makes only a paltry $940 per day. Of course, this is for a certain day and specific subnet 8, the popular time series forecasting subnet, but similar trends persist for other subnets as well. This leads me to the following, opinionated recommendations for joining the ecosystem:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*4yUIl8JdLEq6SrzV" /><figcaption>Subnet 8 (Timeseries forecasting) for 2/18/2024. The check mark on the left indicates the node is a validator. Notice the sharp drop off with daily rewards on the right, which gracefully degrades with the validator’s stake.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NvJoYKgtO-TnA6BT" /><figcaption>Subnet 8 (Timeseries forecasting) for 2/18/2024. In stark contrast, the highest performing miner is ranked 65 with a mere ~$940 rewards per day.</figcaption></figure><p><em>Option 1: Join as a subnet creator or owner (High Reward, High Risk)</em></p><p>You must have a creative idea for an AI marketplace as well as significant human resources, time, and coding ability to create and test a new subnet. However, the rewards are high if your subnet becomes popular, but there is significant risk<strong> </strong>of wasted effort if the subnet idea is rejected by the masses.</p><p><em>Option 2: Join as a validator (Fork over the major $TAO)</em></p><p>If you have sizable GPU compute, disk space, etc. it makes sense to be a validator. However, you must have sizable $TAO to stake already, since there are limited validator spots. Moreover, they are given to those we stake the most $TAO.</p><p><em>Option 3: Join as a miner (Only need AI chops, minimal rewards today)</em></p><p>If you have fairly good AI expertise and modest compute (e.g., a single “standard” GPU), you can easily become a miner. In fact, you can easily adapt standard models from HuggingFace for the subnets’ task. However, at the time of writing, the rewards seem minimal even for the best performing miners. Of course, if the price of $TAO skyrockets due to a combination of speculative demand from exchanges as well as <em>organic demand for real AI services</em>, today’s modest $TAO emissions could be worthwhile in the long run.</p><h3><strong>A Gentle Critique of Bittensor</strong></h3><p>On a whole, the idea and execution of Bittensor are brilliant. The disparity between miner and validator rewards seems troubling, since the best AI minds are not incentivized to create continually improving models. Finally, the token price seems to be driven mostly by speculation and hype.</p><blockquote><em>But, what if the predictions and outputs from the best AI models were actually sold to external customers?</em></blockquote><p>This is exactly what many subnet owners are working towards, as we describe next.</p><h4><strong>Bittensor Needs Organic Demand for AI Services</strong></h4><p>Imagine you are a regular Web2 customer and just want to buy the “best” predictions for the stock market or produce images from the “best” stable diffusion model for your niche interest of penguins. Then, you can pay for AI predictions in fiat currency or its equivalent in $TAO (if you fancy Web3) to access the best predictions from miners. If this takes off, we will have fiat currency injected into the economy and the price of $TAO can rise since it will be demanded for (a) speculation and trading, but (b) for <em>real-life AI services by the masses</em>.</p><h3><strong>Financial Modeling for Decentralized Infrastructure Systems</strong></h3><p>As a miner, validator, or $TAO investor, it’s crucial to have financial projections for the organic demand for AI services on Bittensor, the cost of GPU compute, and the speculative demand for $TAO. For example, we might ask:</p><blockquote>“As a miner, my cost for an Nvidia GPU is $5/hr on a popular cloud platform. If my AI model produces the top 5% of predictions, and we have $5000/day of demand for stable diffusion inferences, how much $TAO would I expect to earn daily?”</blockquote><blockquote>“Given my daily cost of running a GPU, the cost of training, and my expected model quality, how much demand should I service to become profitable in 2 months?”</blockquote><p><strong>If you’re interested in data-driven and “what-if” answers to such questions, check out our </strong><a href="https://arxiv.org/abs/2210.12881"><strong>tokenomics paper</strong></a><strong> (open-source code coming soon)!</strong> It can model the financial economics of any decentralized physical infrastructure system, such as decentralized AI, GPU compute, 5G wireless, and much much more!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b0ad811d2728" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How To Write a PhD Thesis Statement]]></title>
            <link>https://sandeepchinchali.medium.com/how-to-write-a-phd-thesis-statement-d592e7ebff0f?source=rss-19b39523c706------2</link>
            <guid isPermaLink="false">https://medium.com/p/d592e7ebff0f</guid>
            <category><![CDATA[academia]]></category>
            <category><![CDATA[phd-thesis]]></category>
            <category><![CDATA[academic-writing]]></category>
            <category><![CDATA[phd]]></category>
            <category><![CDATA[stem-education]]></category>
            <dc:creator><![CDATA[Sandeep Chinchali]]></dc:creator>
            <pubDate>Sun, 18 Feb 2024 20:26:16 GMT</pubDate>
            <atom:updated>2024-02-18T21:04:21.042Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>The thesis statement sells your work in one-line</strong></h3><p>As a CS PhD student at Stanford, I walked into a new faculty member’s office and asked him to join my thesis committee. He looked at me and said, “Well, what is your one sentence thesis statement?” Unprepared for this question, I rattled off some long-winded answer. Needless to say, he was lukewarm about joining my committee.</p><p>The experience taught me that one needs a clear and concise thesis statement for the PhD. In this guide, I describe how to craft such a statement with examples from my own PhD thesis.</p><h3><strong>My PhD Thesis Statement</strong></h3><p>My PhD thesis was on cloud and networked robotics. My statement was:</p><blockquote>It is possible to design systems and algorithms for distributed inference, learning, and control between fleets of networked robots and the cloud that systematically trade-off cloud accuracy and systems costs (network bandwidth, storage, cloud computation, and energy).</blockquote><p>Let’s provide some guidelines for writing your statement:</p><ol><li>The thesis statement must be one sentence.</li><li>It needs to be specific.</li><li>It is an assertion, not a vague statement, such as: “This thesis is about cloud robotics”.</li><li>The assertion tells the reader what is possible now (and what was not possible before) thanks to your contribution.</li></ol><h3>You need 3 Papers to Support Your Case</h3><p>A common misconception I see in PhD students is that you need several papers to get a PhD. In fact, in my lab at UT Austin, you only need 3 concrete papers that logically build on each other and demonstrate the claim in your thesis. Here’s an example from my PhD thesis.</p><h3>Contribution 1: Distributed Inference</h3><blockquote>We develop an <a href="https://link.springer.com/article/10.1007/s10514-021-09987-4">algorithm</a> for distributed inference between a robot and the cloud that allows a robot to flexibly trade-off the accuracy of cloud computation and network and compute latency.</blockquote><blockquote>Specifically, we cast distributed inference as a Markov Decision Process (MDP), where a robot chooses whether to invoke a local or cloud model given its real-time network and physical state.</blockquote><p>Again, let’s provide some guidelines:</p><ol><li>The contribution statement is succinct — it’s only two sentences.</li><li>Sentence one describes the capability.</li><li>Sentence two gives a specific hint about the technical insight and solution.</li><li>Given these two sentences, a creative reader should be able to get the key idea to re-create your paper without even reading it.</li></ol><p>I now show how my next two papers <em>logically build upon</em> the first contribution in my thesis. For brevity, I leave out the second sentence on the technical insight and details.</p><h3>Contribution 2: Distributed Learning</h3><blockquote>We develop an<a href="https://link.springer.com/chapter/10.1007/978-3-030-71151-1_27"> algorithm </a>to statistically sample valuable events from terabytes of robot sensory data (video and LiDAR) to continually improve computer vision models in the cloud with minimal network upload, storage, and human labeling costs.</blockquote><h3>Contribution 3: Learning Compressible Data Representations for Networked Machine Learning</h3><blockquote>We develop an <a href="https://roboticsproceedings.org/rss17/p046.pdf">algorithm</a> to learn and transmit salient features of sensory data for machine (as opposed to human) perception that maximizes the accuracy of networked perception tasks while minimizing bandwidth consumption.</blockquote><h3>Why Should Anyone Sign Off on Your PhD Thesis?</h3><p>Let’s analyze how, together, my three main papers build a cohesive, logical thesis.</p><ol><li>The first paper provides an algorithm to offload inference, such as computer vision tasks from a low-power robot to the cloud.</li><li>The second paper builds upon this algorithm to extend it beyond real-time inference to offline distributed learning.</li><li>Then, the third paper provides an algorithm to encode sensory data for the first two tasks of distributed inference and learning.</li></ol><p><em>Collectively, these papers build upon each other and provide primitives for a cloud robotics systems “stack”.</em></p><p>What will your thesis be? Reach out on my academic email or <a href="https://twitter.com/SPChinchali">twitter</a> for more guides.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d592e7ebff0f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>