<?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 Shuvro Biswas on Medium]]></title>
        <description><![CDATA[Stories by Shuvro Biswas on Medium]]></description>
        <link>https://medium.com/@shuvrobiswas?source=rss-4762af85f9e3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*5417u6gQdWXEHK6RJgjCbA.jpeg</url>
            <title>Stories by Shuvro Biswas on Medium</title>
            <link>https://medium.com/@shuvrobiswas?source=rss-4762af85f9e3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 06:02:59 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@shuvrobiswas/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[Can you eat & live healthy off of Amazon.com only?]]></title>
            <link>https://medium.com/@shuvrobiswas/can-you-eat-live-healthy-off-of-amazon-com-only-a36bcffbd07b?source=rss-4762af85f9e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/a36bcffbd07b</guid>
            <category><![CDATA[experiment]]></category>
            <category><![CDATA[health]]></category>
            <category><![CDATA[healthy-eating]]></category>
            <dc:creator><![CDATA[Shuvro Biswas]]></dc:creator>
            <pubDate>Mon, 13 Jul 2020 12:32:36 GMT</pubDate>
            <atom:updated>2020-07-13T13:17:19.065Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>So this will be an informal/short one — couldn’t fit it on a tweet, so here goes. Please excuse any typos (feel free to point them out) — time is always of the essence. Not sponsored.</blockquote><h3>TLDR</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mDrZkWAxIRP82gei8dMXHw.png" /></figure><p>Ran an <strong>experiment </strong>recently to live off of <strong>food items available through amazon.com only*</strong>, this ended after 2 days — main finding: <strong>with limited variety, not possible</strong> — the main issue is not calories, rather it’s that you literally have <strong>0 appetite</strong> for the same type of food after some point, and don’t want to have it even if hungry.</p><p>So then you end up <strong>hungry → can’t sleep → lower productivity/energy</strong> the next day. So what was supposed to be a 5 day experiment, ended 2 days in.</p><p>* <strong>Amazon fresh</strong> was not included in this experiment, since this is not available in all areas of America (i.e. this experiment should <strong>generalize </strong>to anywhere in <strong>America </strong>— I moved from <strong>nyc</strong> to <strong>Pennsylvania </strong>recently, you can imagine the rest..).</p><h3>More details for those interested</h3><p>I mostly had <strong>owyn protein shakes</strong> (pictured above), which is 180 calories per bottle. So you would need about 14 of them to get to around <strong>2500 calories</strong> (the suggested amount for adult males according to: <a href="https://www.nhs.uk/common-health-questions/food-and-diet/what-should-my-daily-intake-of-calories-be/">https://www.nhs.uk/common-health-questions/food-and-diet/what-should-my-daily-intake-of-calories-be/</a> — of course in reality, this is <strong>highly individualized</strong> and depends on one’s daily routine/activities as well — you have to figure out what the rough figure is for you body).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/502/1*uGg8biGYtYYLxHL9xMX9uA.png" /></figure><p><strong>14 of those</strong> is too much for a day, so I combined each shake with an <strong>mct oil</strong> pouch from <strong>fbomb </strong>(pictured left). This is an additional 250 calories, so each <strong>owyn shake + mct oil</strong> pouch packs 430 calories, that’s a decent hit from one drink and a great option in general when you want to quickly down a decent hit of calories i.e. <strong>high calorie to volume ratio</strong> — on the flip side: can be hard on the stomach after you’re 2+ of these in, but doable/bearable.</p><p><strong>Side note:</strong> need to confirm this from the literature, but fats are likely good for not putting on weight in the form of fat (i.e. potentially a signal to your body that the <strong>environment</strong> is plentiful in fat and therefore no need to pack it on — remember hearing something along these lines at some point — in any case, a pet hypothesis atm which I will need confirm in a future post or tweet — feel free to chime in if you have any thoughts here).</p><p>I had <strong>4 of these per day</strong>, for a total calorie count of <strong>1720</strong> + a small amount of <strong>pistachios</strong> and a pack of <strong>siren protein bites</strong> to bring up the calorie count into the <strong>1800–1900</strong> range (for both days).</p><p><strong>1800–1900 calories</strong> should be enough for me (i.e. it’s enough, I shouldn’t be starving on that, I also went for a quick run on both days — but nothing too intense).</p><h3>Supplements to facilitate the new dietary conditions</h3><p>During the 2 days, I also supplemented with a <strong>sodium, calcium, &amp; potassium beta-hydroxybutyrate (BHB) </strong>supplement (meant to facilitate <strong>ketosis </strong>→ when your body burns primarily <strong>fats </strong>for fuel rather than <strong>carbs</strong>): <a href="https://www.amazon.com/Vinegar-Capsules-Formula-Appetite-suppressant/dp/B07GNZNNFN">https://www.amazon.com/Vinegar-Capsules-Formula-Appetite-suppressant/dp/B07GNZNNFN</a> (not advocating any particular brand here, just sharing what I took).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*kKD0LNrmPf_JSiGg" /></figure><p>The <strong>keto bhb pills</strong> did help quite a bit in keeping hunger at bay.</p><p>I also supplemented with a <strong>multivitamin </strong>(to ensure my body had all the essential <strong>vitamins &amp; minerals</strong> needed), I usually go with the <strong>performix </strong>brand (which doesn’t seem to be available on amazon.com atm for some reason, shown left).</p><h3>Getting enough calories → does not imply you won’t be hungry</h3><p>Another interesting finding from this experiment as well as from prior experience is that even if you objectively get all the calories and nutrients that you need per day, you may still be hungry. I.e. your body (stomach) does not get all the right <strong>satiation cues</strong> it needs when you’re consuming mostly protein shakes (or similar, such as: shakes based on meal replacement powders) + snacks (though the <strong>keto bhb pills</strong> do help out quite a bit here).</p><p><strong>To restate once more (i.e. the main finding)</strong>: at some point you will get sick of the limited set of items you are consuming and won’t want to have them even if you’re hungry.</p><h3>Possible future experiment / potential modifications:</h3><ul><li><strong>Increase shake &amp; snack variety</strong> (also: the “snack set” should include both sweet &amp; salty snacks, sweet only gets repetitive very quickly).</li><li><strong>Supplement with fiber powder</strong>, to aid in giving your body all the right satiation cues, for example: <a href="https://www.amazon.com/GoBiotix-Prebiotic-Fiber-Boost-Unflavored/dp/B077MFYLSS/">https://www.amazon.com/GoBiotix-Prebiotic-Fiber-Boost-Unflavored/dp/B077MFYLSS/</a></li></ul><p>In any case, I finally got my first shipment of meals from <strong>trifecta </strong>(<a href="https://www.trifectanutrition.com/">https://www.trifectanutrition.com/</a> — <strong>chef prepared, vacuum sealed meals, delivered in bulk to your home</strong>), trying out the keto plan at the moment. Much healthier than most (if not all) of the eating options out here, so I should be good for some time.</p><p>At some future time point, I will need to give this experiment a take 2 with the modifications mentioned above + anything else that makes sense.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/616/0*sI17Y1ZaieVZSCbV.jpg" /><figcaption>Bon Appétit! (Image taken from: <a href="https://www.trifectanutrition.com/meal-plans">https://www.trifectanutrition.com/meal-plans</a>)</figcaption></figure><blockquote>And just to quickly mention again: I am not sponsored by any company — all the products chosen here were based on quick (very quick) research on my part, alternatives could have been chosen/used.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a36bcffbd07b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[On religion as an evolutionary phenomena, what is “god”?]]></title>
            <link>https://medium.com/@shuvrobiswas/on-religion-as-an-evolutionary-phenomena-what-is-god-7fd76ce5c128?source=rss-4762af85f9e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/7fd76ce5c128</guid>
            <category><![CDATA[evolution]]></category>
            <category><![CDATA[religion]]></category>
            <dc:creator><![CDATA[Shuvro Biswas]]></dc:creator>
            <pubDate>Wed, 06 May 2020 18:13:55 GMT</pubDate>
            <atom:updated>2020-05-11T15:18:27.178Z</atom:updated>
            <content:encoded><![CDATA[<h3>Warning:</h3><p>This article is speculative. Please excuse any spelling / grammar mistakes (time is always of the essence).</p><h3>What is “god”?</h3><p>It is the idea of something being beyond any single individual (be it some force, entity, or otherwise).</p><p>It is further freedom from rule by “other individuals” and therefore an embodiment of autonomy. When “god” lays down the rules, the same rules apply to everyone, no one is above anyone else, and all answer only to god. In a scene from Knightfall, shown below, we hear the following:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FJ8e1C0i1bqs%3Fstart%3D47%26feature%3Doembed%26start%3D47&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DJ8e1C0i1bqs&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FJ8e1C0i1bqs%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/0f22e75fdc63be3da3cab34361b8241a/href">https://medium.com/media/0f22e75fdc63be3da3cab34361b8241a/href</a></iframe><h4>“And remember that we answer to god, just as we fight for god, and we win by the grace of god”</h4><p>The knights are not fighting for kings or queens or for some individual <strong>apparently</strong> “higher up” — men* who do, are not autonomous, they’ve given their life to someone else.</p><p>Rather they’re charging forward for the eternal embodiment of the values they hold dear. This is worth fighting for. However this is not enough, as the specific embodiment of values one holds dear matters (in other words, the specific value system in question matters).</p><h3>Examples of common value systems:</h3><p>Christianity<br>Judaism<br>Islam <br>Hinduism<br>Buddhism<br>…</p><p>If the value system (when shared among a sufficient number of people) is well put together (i.e. consistent within its own terms [1], similar to the concept of consistency in formal systems theory) and when implemented approximates a valid or sensible strategy from an evolutionary perspective (i.e. increases the likelihood of gene propagation of participants subscribed to said value system, and is able to defend its members), then this value system will persist and potentially grow over time in one or both of the following ways:</p><ol><li>Via the now higher offspring production rate enabled by the value system.</li><li>Via self-propagation in the form of “converting” those currently not subscribed to the value system.</li></ol><p>And so if a value system is well put together, and is able to defend and propagate itself, we get at last at: “and we win by the grace of god” i.e. the value system is sufficiently strong enough to survive &amp; spread.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0h7nTVthpIZ3584-.jpeg" /><figcaption>A scene from the Mahabharata — Krishna &amp; Arjuna riding into battle</figcaption></figure><h3>Religion as information on evolutionary patterns and successful survival strategies</h3><p>Religion and religious texts are highly concentrated distillates of information from our ancestors on the patterns of natural selection humanity has had to deal with in the past (these patterns have likely recurred at various scales and levels many times throughout history, making religious texts outliers in information density relevant and useful for survival and propagation of genes).</p><p>However useful these texts may be in the “<strong>form</strong>” &amp; “<strong>analogy</strong>” based knowledge they contain, they are not literal truths (nor does this imply they’re useless, i.e. approximate guides with anomalously high information density encoding past survival scenarios + corresponding behavioral/cultural survival strategies to navigate those scenarios indeed help propagate genes, and therefore are quite useful from a practical point of view).</p><p>^ This gives me a new found appreciation for the world’s religions — each densely packed with information about the evolutionary process &amp; our nature, each telling us something about our story.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*1C9sT5Fu-JzJXuwp" /><figcaption>Icarus flying too close to the sun</figcaption></figure><h3>Towards a new value system for humanity</h3><p>We need something new. There is a disconnect between the knowledge &amp; wisdom religion is able to provide, and the current times we live in. This is fine.</p><p>The world progresses, we learn, we discover, we create, and with that we enter new worlds where we may find old value systems outdated. We do need new value systems. Value systems are important since they embody autonomy, i.e. for everyone subscribed to a “fair” value system, everyone is autonomous &amp; sovereign and does not report to anyone else. But what is the right value system? Or for starters, what is a “good” value system?</p><p>Let’s eliminate the degenerate case: value systems that include axiomatic statements such as: do what person X says.</p><p>Nondegenerate value systems are “symmetric” in the sense that any person subscribed to the value system has the same rights and constraints imposed upon them as anyone else — only symmetric value systems enable autonomy.</p><p><em>A side: some symmetric value systems can eventually over time be manipulated / hijacked by persons who skew the given value system to empower themselves (i.e. charismatic religious leaders who selectively pick &amp; choose ideas/values from a symmetric value system to propagate aims that were not of the original value system itself but confounded with the given individuals aims).</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FsspnvYKFKDM9LbeQVAs_g.jpeg" /><figcaption>Symmetry</figcaption></figure><h3>In search of a fair, rational, &amp; symmetric value system</h3><p>We need to create a new value system, one that rational individuals would consider “fair.” Something to take the place of the religions we have lost and that is scientifically and rationally grounded. In doing so, we have a greater shot at being truer to our nature, freedom, and autonomy.</p><p>The way to do that in a “multi-agent” setting is with a “fair” &amp; symmetric value system that maximizes some collective utility or reward function.</p><p><strong>Work to be done:</strong> defining more concretely what a “fair” value system might look like, one based on our current knowledge of the world and of ourselves, and defining more concretely what our collective utility function <em>is </em>or should be. The latter will of course be of much debate, but I argue it is derivable axiomatically from a few simple principles taken axiomatically, but this (as with all things) must be corroborated / verified via peer review.</p><p>————————————————————————————————</p><p>[1] “More or less” for most adopted real world value systems.</p><p>* Men used here for historical context, a proxy for individuals in general.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7fd76ce5c128" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Towards AGI: understanding and math > compute power and data]]></title>
            <link>https://medium.com/@shuvrobiswas/towards-agi-understanding-and-math-compute-power-and-data-8d16c640fe89?source=rss-4762af85f9e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/8d16c640fe89</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[agi]]></category>
            <dc:creator><![CDATA[Shuvro Biswas]]></dc:creator>
            <pubDate>Mon, 17 Feb 2020 19:32:28 GMT</pubDate>
            <atom:updated>2020-02-17T19:32:28.985Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Warning:</strong></h3><p>This article is speculative, and an initial description of some ideas I’ve been playing around with that I intend to flesh out in full detail over time.</p><p>In this article, I argue that the next step towards building AGI (artificial general intelligence: <a href="https://en.wikipedia.org/wiki/Artificial_general_intelligence">https://en.wikipedia.org/wiki/Artificial_general_intelligence</a>) will not require more compute power than a modern day laptop, nor will it require massive quantities of data — in fact, it may not require any data at all.</p><p><strong>How so?</strong></p><h3><strong>On Data</strong></h3><p>Data is not needed because we can autogenerate the data needed to build “intelligent” systems. For example, DeepMind’s achievements with the game of go (with their AlphaGo Zero system: <a href="https://deepmind.com/blog/article/alphago-zero-starting-scratch">https://deepmind.com/blog/article/alphago-zero-starting-scratch</a>) does not rely on external training data, but rather uses the paradigm of “self-play” to improve over time.</p><p>More generally, in the reinforcement learning framework, an “environment” with multiple agents (the game of go for example, or call of duty perhaps) is a potentially infinite source of data.</p><p>The “<strong>right</strong>” environment, with the “<strong>right</strong>” agents, will produce an evolving dynamic, where the <strong>N-agent system</strong> overall is far from equilibrium and exhibits “<strong>perpetual novelty</strong>” (which is the case for all complex systems in general, see: <a href="https://www.sfipress.org/books/worlds-hidden-in-plain-sight">https://www.sfipress.org/books/worlds-hidden-in-plain-sight</a>).</p><p><strong>“right” agent</strong>, can be further articulated into, agents with the “right” action-set, and “right” utility function.<br><strong>“right” environment</strong> is one that is not too “simple”, and not too “chaotic” either.</p><p>Both of these notions need to be further articulated and detailed which I will take a stab at in later articles in this series.</p><p>In fact, if we think our universe is a simulation, then all data in it is autogenerated (this is a separate issue, but the statement here holds as a conditional).</p><p><strong>Side note:</strong> I’ve referred to AGI without giving a formal definition of what that means. Some detractors of AGI would claim intelligence is not definable, but there is no reason to believe so, one working definition of general intelligence is given by computer scientist Marcus Hutter: <a href="http://www.hutter1.net/ai/aixigentle.pdf">http://www.hutter1.net/ai/aixigentle.pdf</a></p><h3>On Compute Power</h3><p><strong>Assertion:</strong> General intelligence will be solved *with respect to* compute power. That is, we must understand how different regimes of compute power limit and affect intelligence. Given X amount of compute power, what is the most intelligent system that can be built? Answering this question is more important than jumping into the practice of building intelligent systems without considering compute power (which many do, and is common practice at the moment). Answering this question also gives us insight into how much compute power will be required to accomplish various tasks as well as achieve human (or let’s take an easier target: chimp) level intelligence.</p><p>One of the next steps in AGI research is understanding the relationship between compute power and various “<strong>levels</strong>” of intelligence. This step can be taken on any modern laptop or desktop because we do not even understand small neural networks well (let’s say with just 1000s of parameters). More effort should be directed at studying small neural networks, and how they “learn” — so the next step in AGI research does not really require more horsepower, but more dissection and understanding of what we have.</p><p>Further the focus on <strong>neural networks</strong>, and deep learning is limiting. We should consider other fields and disciplines that may have some insight to add; for example: complex systems theory (much progress has been made here), evolutionary biology, neuroscience, information theory, ontology, epistemology and so on.</p><h3>Final Comment: can capitalism produce “interesting” AI?</h3><p>My final comment is that, at least in the U.S., where machine learning and AI work has surfaced into the economic sector (making its way out of academia), efforts are directed mostly at improving accuracy by marginal amounts, to increase the bottom line. This limits what we look at and study.</p><p>Our current desire for more data and more horsepower to “increase accuracy” by a marginal percentage will likely soon hit a wall. This is not to say that AI research is stagnating, but to suggest that perhaps we are going about it in the wrong way.</p><p>Notable exceptions include: deepmind/google, facebook, and a few other big tech companies who are spending money on solving fundamental problems (but again, the general pattern of research here seems to be more around scale and data, rather than fundamentals and understanding).</p><p>Nontheless, obscure academics and researchers continue to tackle the bigger questions, such as: Juergen Schmidhuber, Marcus Hutter, and so on. The next major breakthroughs towards AGI will likely come from non-obvious/obscure sources.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8d16c640fe89" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cryzen’s Machine Learning and Algorithmic Trading Interface]]></title>
            <link>https://medium.com/@shuvrobiswas/cryzens-machine-learning-and-algorithmic-trading-interface-7d5baad0d7d8?source=rss-4762af85f9e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/7d5baad0d7d8</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[algorithmic-trading]]></category>
            <category><![CDATA[xen]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Shuvro Biswas]]></dc:creator>
            <pubDate>Mon, 20 Aug 2018 17:29:57 GMT</pubDate>
            <atom:updated>2018-08-20T18:17:45.901Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2pBZKMi9DdWIZTYwpnOB0g.jpeg" /></figure><h3>The Future of Trading</h3><p>The future of trading (be it stock trading or cryptocurrency trading) is machine learning and AI based algorithmic trading. Code can react to changing market conditions faster than any human can. And machine learning can bring in data driven / pattern recognition based approaches to trading.</p><p>While it is impossible to perfectly predict the future, probabilistic insights can be gained and used to generate <a href="https://www.investopedia.com/terms/a/alpha.asp">alpha</a>. As the field of machine learning continues to break new ground, there will be greater and greater opportunities to apply the latest techniques to algorithmic trading.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nwf53MNs1ZXSszcRyMLIWA.jpeg" /><figcaption>Machine vision system detects people — convolutional neural networks, which typically power machine vision systems, are general pattern recognizers which can also be applied to markets.</figcaption></figure><p>To this end, Cryzen’s algorithmic trading platform will prioritize the development of machine learning and AI based trading strategies.</p><h3>Cryzen — Bot Development Environment</h3><p>The simplest way to describe Cryzen’s platform is as a “bot development environment” that allows users to develop, tweak, test, and deploy bots from an easy to use online <strong>code studio</strong> — further, nonprogrammers will have the ability to tweak and run premade bots from a simplified UI.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GoZkGPzdjctY5Eo3" /><figcaption>Cryzen Code Studio Depiction</figcaption></figure><p>Cryzen’s code studio is meant to express trading logic that takes into account various data streams to execute trades and transfers. For example, one can very simply write trading logic to buy ETH if a tweet contains the hashtag #ethereum, is tweeted by an account with more than 20k followers, and if the tweet is of positive sentiment (we will provide sentiment scores as part of the twitter data package so the user isn’t required to write the code to do so, though they could if they wanted to).</p><p>However, what if one wanted to do something more complex like train a Neural Network to predict price volatility over the next hour, given past price and twitter data? It’s hard to train and validate a machine learning model within a backtesting system (it’s just not the right fit).</p><p>To address this issue, Cryzen will release a separate machine learning interface to train, validate, and test models.</p><h3>Cryzen’s Machine Learning Interface</h3><p>The Cryzen machine learning interface will have access to the same data streams available for algorithmic trading. Instead of writing trading logic, however, users will specify some aspect of the future to predict given some aspect of the past.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6q7wLLi8mK-qmObaUIL9tw.png" /><figcaption>For each timepoint <strong>t</strong> in our available data streams, we can look at the data before it (“past” data relative to <strong>t</strong>), to tell us something about the data coming after it (“future” data relative to <strong>t</strong>). This is how we generate training data for machine learning models in the context of time-series data. The image above depicts one instance or row of training data. By moving timepoint <strong>t </strong>around, many instances or rows of training data can be generated. Typically, some features would be derived from the “past” data window, to predict some feature derived from the future “data” window.</figcaption></figure><p>Going with our previous example, the aspect of the future we may care about is: ETH volatility over the next hour as measured by the standard deviation. The aspects of the past we may think is relevant may be: ETH price over the last 24 hrs, and any tweets containing the hashtag #ethereum from an account with more than 20k followers. Let’s get into the details and fully describe our model inputs and model outputs:</p><p>Inputs:</p><ul><li>ETH price change over 24 hrs expressed as a percentage</li><li>Standard deviation of ETH over past 24 hrs</li><li>Mean of ETH over the past 24 hrs</li><li>Average sentiment of all tweets containing #ethereum over the last 24 hrs</li><li>Max sentiment of all tweets containing #ethereum over the last 24 hrs</li><li>Min sentiment of all tweets containing #ethereum over the last 24 hrs</li><li>Standard deviation of sentiment of all tweets containing #ethereum over the last 24 hrs</li></ul><p>Output:</p><ul><li>Standard deviation of ETH for the next hour</li></ul><p>Given these input / output dynamics (easily specifiable via the Cryzen machine learning interface), we can generate standard training data for supervised machine learning models— at this point it’s off to the races with standard Python machine learning libraries such as Scikit-learn, Keras, and Tensorflow.</p><p>Possible model choices include:</p><ul><li>Random forest (regression version)</li><li>Standard feedforward neural network</li><li>Custom model created with Tensorflow</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jg8rAi0AknJSwugQjoEXBg.png" /><figcaption>Various Neural Network Architectures</figcaption></figure><p>Once your model is specified, we will take care of training, time-series cross-validation, testing, and hyperparameter optimization over the date range specified.</p><p>What you are returned with is a fully trained model, and associated prediction accuracy. Once satisfied with the model’s performance, the user can hop over to the code studio and create trading logic that depends on predictions from their newly created machine learning model.</p><p>Specifically, the user can now augment their trading algorithm with the model’s predicted ETH volatility over the next hour, making logic such as the following possible: cancel all ETH orders if the model’s prediction of upcoming volatility is high (i.e. maybe you want to play it safe in volatile conditions — though these conditions can also be used to make large/quick profits).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wwNjvfglCCaBKzUDedfOUw.jpeg" /></figure><p>The ability to develop and train predictive machine learning models independently from the development of trading logic is powerful. Users can generate a repertoire of predictive models and mix and match them in any way with existing or upcoming algorithms they plan to develop.</p><p>To kickstart the injection of machine learning into cryptocurrency trading, Cryzen will provide some premade predictive models that anyone can use without going through the machine learning interface, such as:</p><ul><li>volatility predictors (a more involved version of the model discussed in this article)</li><li>network congestion predictors</li></ul><p>Visit <a href="http://www.cryzen.com/">www.cryzen.com</a> and subscribe to stay in the loop on feature releases, machine learning, and more! Also follow us on social media to keep an eye on the horizon of cryptocurrency algorithmic trading:</p><ul><li><a href="https://twitter.com/cryzenllc">Twitter</a></li><li><a href="https://www.facebook.com/cryzenllc">Facebook</a></li><li><a href="http://t.me/cryzen">Telegram</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7d5baad0d7d8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[XENB — Cryzen’s Bounty Token]]></title>
            <link>https://medium.com/@shuvrobiswas/xenb-cryzens-bounty-token-c975021e3882?source=rss-4762af85f9e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/c975021e3882</guid>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[ethereum]]></category>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[algorithmic-trading]]></category>
            <category><![CDATA[token]]></category>
            <dc:creator><![CDATA[Shuvro Biswas]]></dc:creator>
            <pubDate>Fri, 10 Aug 2018 21:49:06 GMT</pubDate>
            <atom:updated>2018-08-10T22:12:40.446Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yx2pkhyhLXYetgkNSqpq3g.jpeg" /></figure><h3><strong>XENB Launched August 8, 2018</strong></h3><p>This week we deployed XENB to the Ethereum blockchain: <a href="https://etherscan.io/token/0x9aa060f5ce3c994772adf9deb8418a0d47a90d26">https://etherscan.io/token/0x9aa060f5ce3c994772adf9deb8418a0d47a90d26</a></p><p>In this short post, I’ll describe XENB in detail, why we’ve created it, and how it will convert to XEN dividend token in late 2019.</p><p>XENB is our bounty token — that is, a placeholder token which we can send to partners and the community while XEN and the Cryzen algorithmic trading platform gets up and running.</p><h3>XEN Dividend Token — Short Summary</h3><p>To summarize earlier posts by Cryzen, XEN is a dividend token which will use an <a href="https://medium.com/@mustwin/building-an-oracle-for-an-ethereum-contract-6096d3e39551">oracle / smart contract</a> system to distribute Cryzen’s profits to XEN token holders. In specific, the XEN system will take the profits displayed on <a href="http://www.cryzen.com/profits">cryzen.com/profits</a> (fully available to the public), and distribute 40% of it to XEN token holders.</p><p>To learn more about where exactly Cryzen’s profits will come from (something you may be interested in as a XENB holder or future XENB / XEN holder), please see our <a href="https://medium.com/@cryzen/introducing-cryzens-xen-dividend-token-aa523052c213">previous article</a> on XEN token.</p><h3>XENB Bounty Token</h3><p>XENB in the interim will be used to fuel our bounty campaign program on <a href="https://www.cryzen.com/bounty">cryzen.com/bounty</a>, onboard partners, and soon fuel machine learning and backtesting competitions on our platform.</p><p>In late 2019, XENB will be convertible to XEN dividend token in a 1:1 manner. Once XEN dividend token is legally compliant with securities laws in the U.S. we will launch XEN to the Ethereum blockchain — our current estimate is that this will take place in mid-2019.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/832/1*_O-rdAmXBURPGLxp-zndUg.png" /></figure><p>After the XEN token ICO is complete, we will deploy a <em>XENB to XEN</em> conversion smart contract. Users holding XENB will be able to send their XENB to our conversion smart contract and receive XEN in return.</p><h3>XENB Transfer Block</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EG7RpDEeZ7RhlXfjL5Aqqw.png" /><figcaption>XENB Code Snippet</figcaption></figure><p>For the time being XENB is nontransferable. That means, anyone in our bounty program receiving XENB will not be able to move it out of their receiving wallet. This has been done for compliance reasons (with respect to our upcoming presale and ICO).</p><p>The XENB transfer block will be lifted after 1 year (this has been hard coded into the logic governing XENB) or if we override the lock, an action we will take after our SEC compliant XEN token ICO — we are currently working with a great legal team to make all of this happen!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c975021e3882" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>