<?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 Ankur on Medium]]></title>
        <description><![CDATA[Stories by Ankur on Medium]]></description>
        <link>https://medium.com/@dopetard?source=rss-e060bb5c25e3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*pQ71KKE9UberWvy-6Ija0Q.jpeg</url>
            <title>Stories by Ankur on Medium</title>
            <link>https://medium.com/@dopetard?source=rss-e060bb5c25e3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 04:37:49 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@dopetard/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[Bitcoin: Lightning on the cusp of breakout growth.]]></title>
            <link>https://medium.com/prominent/bitcoin-lightning-on-the-cusp-of-breakout-growth-86d1942463af?source=rss-e060bb5c25e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/86d1942463af</guid>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[lightning]]></category>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[lightning-network]]></category>
            <category><![CDATA[bitcoin-news]]></category>
            <dc:creator><![CDATA[Ankur]]></dc:creator>
            <pubDate>Mon, 23 Aug 2021 02:49:34 GMT</pubDate>
            <atom:updated>2021-08-23T02:49:34.317Z</atom:updated>
            <content:encoded><![CDATA[<p><strong><em>TL;DR</em></strong><em> From Nassim Taleb’s black paper to Peter Schiff’s tweets, scaling has been a major criticism of Bitcoin. Because of the growth of the Lightning network, which is going parabolic since January this year — the skeptics might be in for a surprise.</em></p><p><strong>Context</strong></p><p>The Bitcoin “block space” is a free market. The fee required to send a Bitcoin transaction varies and depends on the supply and demand for the block space. Bitcoin has a “maximum block size” (4 MB) that limits the maximum amount of transaction data added to the block. The Bitcoin network targets that a new block is being mined every ten minutes. It makes Bitcoin’s block space a scarce resource, and with ever-increasing demand, more expensive. Evidently, it is impractical to use Bitcoin’s base layer to pay for your coffee. To the people who are new to the community and weren’t there to experience the days of <a href="https://masterthecrypto.com/bitcoins-civil-war-how-and-why/">Bitcoin civil war</a>, the most obvious solution to this problem would seem as easy as <em>increasing</em> the Bitcoin’s block size to meet demand, as Elon Musk recently suggested:</p><h3>Elon Musk on Twitter: &quot;Block size &amp; frequency should steadily increase to match broadly available bandwidth / Twitter&quot;</h3><p>Block size &amp; frequency should steadily increase to match broadly available bandwidth</p><p>There’s a fundamental flaw in this approach. Censorship resistance is one of the main properties that make Bitcoin special. Bigger blocks would increase the cost of verification, making running a full Bitcoin node expensive and dissuade people from verifying the Bitcoin transactions. The Bitcoin network would then become <em>less decentralized</em> in the long run as in a peer-to-peer network, more geographically dispersed nodes (or peers) are directly proportional to the network’s robustness against censorship. It is thus incumbent upon us that we work towards keeping the cost of running a full node as low as possible. Luke Dash Jr, a Bitcoin core developer, has even <a href="https://twitter.com/lukedashjr/status/1074768903593754627?lang=en">proposed</a> to shrink the Bitcoin blocksize further down to 300kb.</p><p>For Bitcoin to be used as a medium of exchange for billions of users — it needs to scale. It was clear that increasing the blocksize is not an acceptable solution, and the community needed to look further. Various scalability solutions have been proposed over the years. The lightning network got the most traction and has long been touted as a solution to Bitcoin’s scaling problem, with the entire “small block” community rallying behind it. Right from the days of blocksize war, it has had its fair share of criticisms. Roger Ver and the BCash proponents have dismissed it as too complex and hard to use. Others have often ridiculed it as an impractical solution with slow progress. The recent stats, however, tells a different story. The total value locked (TLV) in Bitcoin’s publically advertised lightning channels has doubled since January this year and crossed 2300 Bitcoin ($115m) as depicted below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mTEZdA_zo0JcGFOFxeX1rg.jpeg" /><figcaption>Bitcoin locked in public lightning channels</figcaption></figure><p>From its inception, the lightning network is architectured with privacy-preserving ethos. So we don’t have any quantitative way to find out the total capacity of private lightning channels. Nevertheless, the public channel capacity is a good proxy to gauge the growth of the lightning network. As depicted, the numbers look promising, indicating that the double-side network effect is starting to get established and is poised for breakout growth.</p><p><strong>Custodial Exchanges</strong></p><p>Many factors contributed to the unprecedented growth of the network in the past few months. Major Bitcoin exchanges like Bitfinex, OkCoin, and Kraken have all already integrated or are in the process of integrating lightning-powered deposits and withdrawals. The resulting increase in efficiency also helps the network and bootstraps the liquidity for the users of lightning. Speculative use-cases of these exchanges will result in the protocol maturing and hitting a critical mass of liquidity, which would improve the protocol’s user experience. This feature also helps a regular user as cheap and instantaneous, deposit and withdrawal mitigate the custodial risk of keeping the funds on an exchange wallet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/965/1*bJF4NLCMLRiqkFmue_FL3g.jpeg" /><figcaption>LNMarkets Total aggregated. trading volume</figcaption></figure><p>LNMarkets, a Bitfinex funded Bitcoin derivative platform, was the first to bring this feature to the market. It has now pulled over $200m in aggregated trading volume since its launch in Q1 of 2020. As the protocol matures more, we will see other large-scale derivative exchanges like FTX, Binance, and ByBit integrate lightning as well — which will work as a catalyst for user adoption by helping the network hit a critical mass of second-layer liquidity. It will also make Bitcoin arbitrage faster and less capital intensive, making the market more efficient in the process.</p><p><strong>Remittances — </strong><a href="https://strike.me/"><strong>Strike</strong></a></p><p>Strike is operating as a Bitcoin-lightning-native neo-bank. Within six months of public beta, they have five-figure registered users and millions of dollars in volume per month. Strike is operational in the US, with Strike Global soon to be launched. They have partnered with crypto exchange Bittrex, and plan to onboard 1M+ Bittrex users onto Strike and the lightning network. They also plan to issue Strike VISA cards in the US. Unlike Cashapp or Venmo, which curates a walled garden and only allows for intra-network transactions among its users, Strike’s use of an open lightning network makes it interoperable will all lightning nodes and wallets in existence <em>by default</em>. On the other hand, Strike also “talks with the user’s bank” — making fiat onboarding seamless. Stripe leverages globally ubiquitous Bitcoin “the network” to settle fiat denominated transactions fast and make remittances cheap.</p><p><strong>The lightning backend — </strong><a href="https://galoy.io/"><strong>Galoy</strong></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*b9F4c--7pRehjOeDGCuWkA.png" /></figure><p>Galoy is the official technology and wallet provider for the pilot <a href="https://www.bitcoinbeach.com/">Bitcoin beach project </a>in El Salvador. Bitcoin Beach is an initiative designed to create a sustainable Bitcoin economy — utilizing the Lightning Network. The project is based in El Zonte, a town of roughly 3,000 residents on the Pacific Coast of El Salvador. Galoy is currently in the proof of concept stage and has &gt; 6000 registered users. In the US, merchants accept card payments, which have an interchange rate, a fee that a merchant must pay with every credit card and debit card transaction. Financial service firms charge this fee in return for accepting the credit risks and handling charges inherent in credit card transactions. We believe that lightning provides a better alternative to the merchants, with fewer inherent fees. B2B services like Galoy can become a “Stripe for the lightning economy” — by providing developer toolings to SMEs. Galoy is entirely <a href="https://github.com/GaloyMoney">open source</a>, and we are very excited about the prospects here.</p><h4>Closing Thoughts</h4><p>The various node implementations of the lightning network are under active development, and the protocol is maturing with every passing day. Many challenges are yet to be tackled, like better user experience, higher liquidity, more robust security, and even a favorable regulatory landscape. Optimism doesn’t wait on facts, it deals on prospects, and we have a lot of it. Twitter and Square (Cashapp) have hinted at the plans to integrate the lightning network. Jack Dorsey recently said that it is “<a href="https://twitter.com/jack/status/1403174877444587523?lang=en">only a matter of time</a>” before Twitter launches their Lightning network integration. Many prominent podcasters are already accepting Bitcoin streamed to them per second on the apps like Breez and Spinx. El Salvador will soon airdrop $30 on lightning bitcoin to all of its 6 million citizens. We at HCM Capital are excited about the prospects here and believe that the network will continue to grow orders of magnitude in the coming years, ushering in the new era of financial sovereignty and freedom for the world.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=86d1942463af" width="1" height="1" alt=""><hr><p><a href="https://medium.com/prominent/bitcoin-lightning-on-the-cusp-of-breakout-growth-86d1942463af">Bitcoin: Lightning on the cusp of breakout growth.</a> was originally published in <a href="https://medium.com/prominent">HCM Capital</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Zhuhai Summit]]></title>
            <link>https://medium.com/prominent/zhuhai-digital-finance-summit-keynote-address-by-jack-lee-5be48410c8ab?source=rss-e060bb5c25e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/5be48410c8ab</guid>
            <category><![CDATA[crypto]]></category>
            <category><![CDATA[defi]]></category>
            <category><![CDATA[china-crypto]]></category>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <dc:creator><![CDATA[Ankur]]></dc:creator>
            <pubDate>Fri, 26 Mar 2021 03:50:21 GMT</pubDate>
            <atom:updated>2021-03-26T18:50:16.621Z</atom:updated>
            <content:encoded><![CDATA[<h3>Zhuhai Digital Finance Summit, Keynote address by Jack Lee.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vPM7tPdm_FR5uzTD_-dJvA.png" /></figure><p>On March 26th, 2021, <a href="https://hcm.capital/">HCM Prominent</a> founder Jack Lee gave a keynote speech on digital assets and the benefits of programmable financial infrastructure at the Digital Finance Summit in Zhuhai, China. This blogpost features the slides and builds upon the ideas shared during the address.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MAzLySdjBXwiAuOl4cDSrg.png" /></figure><ul><li>Bits and bytes are the atomic units of data. When money is encodable like data in form of “bits” — it inherits the properties that make it programmable. The decentralized digital assets are tracked on a public ledger called “blockchain” (which is a database that stores digital information in a data structure called “blocks”) Public blockchain allows this data representing money to be independently verifiable by anyone.</li><li>Public blockchains are engineered so that no one entity or node in the system can have any undue advantage over another. Irrespective of whether you are a bank or a normal participant, you have to play by the same rule.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SROgwSXGMDR2ZWztI2HQxQ.png" /></figure><ul><li>For the unbanked population of the world: removing third-party intermediaries helps with bypassing inadequate financial infrastructure. The permission-less infrastructure of Bitcoin and DeFi enables equitable financial services for the unbanked and underbanked population of the world.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xlBoI0ZgC69LBF1OfLNtSg.png" /></figure><ul><li>For rest of the already banked world, these new permission-less financial infrastructure provides a much needed upgrade over the legacy financial system. In the US, legacy financial system is a big obstacle in building a fast and cheap monetary network. According to Financial Times, 43% of US banks still use COBOL — a programming language invented in 1959. Bypassing this legacy infrastructure via open finance with benefits like no delays, transparency around fees, etc is a viable alternative.</li><li>The innovation around text and pictures in the last decade has resulted in building numerous multi-billion dollar companies. Facebook, Twitter, Snapchat, Instagram, etc is only possible because texts and pictures are encoded as bits and programmed to function under different social constructs. Similarly, programmable money will provide an additional primitive for developers to build innovative applications with.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JL6XT2Rw7BZ1EmoSVQXKSg.png" /></figure><ul><li>The current status quo is where banks using legacy financial infrastructure act as a middleman and run a “rent-seeking business”. This is inefficient and not only results in increasing transaction lag but also makes something as trivial as money transfer expensive for the users. It’s 2021, and yet something as basic as normal wire transfer cannot be done over weekends and holidays. Furthermore, there are counterparty risks inherent with using banking services. In this era where software is eating the world, we think finance will not be an anomaly.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EbfIn4QBUd2d_WJERHKGzg.png" /></figure><ul><li>Users interacting with DeFi apps don’t interact with humans but verifiable code, that mitigates any counterparty risk. The infrastructure is also global and transparent, which results in increased efficiency.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wMrMFtFNdorh-xjTmXg3hQ.png" /></figure><ul><li>The consequent slides will explore the idea maze and innovation in the space we are excited about!</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SyKzKaCJJ_CF7wW-6-_5mw.png" /></figure><ul><li>These open financial primitives must scale to 7.8 billion people worldwide. Visa processes around 70,000 transactions every second, by comparison, a Bitcoin on-chain transaction takes around 10 minutes. Scaling the Bitcoin and other PoW based blockchains have been debated for the better part of the last decade. From block size wars to chain forks, scalability has been one of the major points of contentions in the community. Layer 2 networks like Lightning provides a very viable alternative to Bitcoin’s scaling problem. Bitcoin works as a store of value right now, with lightning — it will also work as a medium of exchange. Startups like Strike by Zap are working on using lightning and the Bitcoin network to build products that provide cheap and fast global remittance.</li><li>There have been myriad attempts to scale Ethereum as well. The community seems to have found consensus around using zk-rollups to scale the Ethereum PoW chain. <a href="https://zksync.io/">zkSync</a> , that uses zk-rollups, is currently live on Ethereum mainnet.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S2ihh6yvcocj8h3qm2kb0w.png" /></figure><ul><li>Bitcoin “the asset” has already been well established as a store of value or digital gold. Bitcoin “the network” has benefits like 24/7 availability and international liquidity — which startups like Strike leverage as a “settlement layer” They use Bitcoin and the lightning network to offer cheap and fast remittances globally. This will make legacy systems like SWIFT obsolete. In 2019, more than 11,000 financial institutions sent over 30,000,000 payments per day via the SWIFT network. Today, over 50% of existing international transfers are done via the SWIFT network. However, 50+ years after its inception, SWIFT remains one of the more expensive ways to transfer funds internationally and settlement times still range from days to weeks. We envision the new paradigm making SWIFT obsolete.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iYZYYF6AznmkGhg5V1UDcw.png" /></figure><ul><li>Complaint services help with bringing more users to the ecosystem. CashApp by Square is doing well in this regard. Recently they allowed people to send and receive bitcoin using $cashtag. It’s similar to being able to send and receive Bitcoin using WeChat ID. CashApp currently has 30 million monthly active users, this exposes them to Bitcoin and helps escalate the timeframe for mass-adoption. Other complaint services like Coinbase are also building easy-to-use products for retail users. These services work as a bridge and onboard new users from traditional finance to the decentralized finance world, we will see more such services built.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jH3Mkbi7vssIXzVVJY-Iow.png" /></figure><ul><li>Financial services offered by DeFi like loans, insurance, etc will bring more fair and equitable services to the masses. It will also enable use cases like PoolTogether (no loss prize games) and Flash loans — that has no equivalent in the traditional financial world and is only possible where money is programmable in the DeFi world. Many startups are working with new technologies like DLC to bring private smart contracts to Bitcoin.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Qvay3pDU6f_xO21iE1MhVg.png" /></figure><ul><li>Internet’s business model will evolve from ad-based to micropayments based. When transactions are cheap and money can be “streamed” like live videos, it allows users to efficiently pay for services according to the usage without needing to trust the creator. Prominent Bitcoin podcasters are using Spnix chat on lightning where the user stream “sats per minute” they listen to the podcast. Ethereum’s Sablier finance also allows for similar use cases.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*smwry6bPL2BNaw60M2zUDg.png" /></figure><ul><li>Composability is a major advantage of Ethereum’s DeFi ecosystem. This “plug and play” model of DeFi protocols works as LEGO blocks, where any developers can combine them in innovative ways to build novel apps. Since these “money LEGOs” are permission-less, developers just have to concentrate on the core business logic and not worry about access to key infrastructure. This opens up a lot of space to build products that were previously unimaginable.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uN5sS8ltkyr7ILHirLRFaQ.png" /></figure><ul><li>The keynote concludes with re-emphasizing the need for a programmable financial infrastructure and its importance in building a more sovereign and equitable future.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lhZ5pQF_kYHMQ8LZDI_ZPA.png" /></figure><p>Signing off with a quotable tweet from Technoking of Tesla himself!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5be48410c8ab" width="1" height="1" alt=""><hr><p><a href="https://medium.com/prominent/zhuhai-digital-finance-summit-keynote-address-by-jack-lee-5be48410c8ab">Zhuhai Summit</a> was originally published in <a href="https://medium.com/prominent">HCM Capital</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Atomic Swap of ERC20 Tokens on the Raiden Network]]></title>
            <link>https://medium.com/@dopetard/atomic-swap-of-erc20-tokens-on-the-raiden-network-a9890c4957d7?source=rss-e060bb5c25e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/a9890c4957d7</guid>
            <category><![CDATA[lightning-network]]></category>
            <category><![CDATA[crypto]]></category>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[ethereum]]></category>
            <category><![CDATA[raiden]]></category>
            <dc:creator><![CDATA[Ankur]]></dc:creator>
            <pubDate>Thu, 01 Mar 2018 07:56:37 GMT</pubDate>
            <atom:updated>2018-03-02T06:48:49.106Z</atom:updated>
            <content:encoded><![CDATA[<iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F1Q3q2o4DKZE%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D1Q3q2o4DKZE&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F1Q3q2o4DKZE%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/e31e3a7d8d1e36e78b6e32733f3b6f71/href">https://medium.com/media/e31e3a7d8d1e36e78b6e32733f3b6f71/href</a></iframe><p><a href="https://www.washingtonpost.com/news/the-switch/wp/2013/11/12/bitcoin-needs-to-scale-by-a-factor-of-1000-to-compete-with-visa-heres-how-to-do-it/?utm_term=.0f95068b0c92">VISA processes around 70,000 transactions every second.</a> By comparison, a single transaction on Bitcoin’s blockchain takes around 10 minutes. In a world where over seven billion people execute billions of micro-transactions every day — resulting in scalability being a huge bottleneck on our quest to widespread adoption of cryptocurrencies. The debate around making cryptocurrencies scalable and practically feasible for everyday use is not new and has resulted in the development of various approaches such as <a href="https://www.coindesk.com/sidechains-breakthrough-almost-everyone-bitcoin-missed/">sidechains</a> and myriad different <a href="https://en.wikipedia.org/wiki/List_of_bitcoin_forks">forks</a></p><blockquote>Blockchain’s decentralised nature is amazing but they are slow, expensive and not scalable</blockquote><p>One of the main challenges for us at Exchange Union is to realise the potential of bridging different exchanges around the world by making the trading of digital assets among them secure and instantaneous. Various off-chain scaling solutions like the <a href="https://lightning.network/">Lightning Network</a> have been under heavy development and promise to solve the latency problem. The idea of settling transactions off-chain and hitting the blockchain <strong>only</strong> when it’s <strong>absolutely necessary </strong>was already envisioned by Satoshi back in 2010:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-oZTT00RtUKmwM99QevdoQ.png" /><figcaption>Satoshi’s Bitcointalk comment on “time” limited transactions akin to today’s lightning and Raiden network</figcaption></figure><p><a href="http://raiden.network/"><strong>Raiden Network</strong></a> is similar to Bitcoin’s Lightning Network, but for Ethereum’s blockchain. Once production ready it will enable instant, low-fee and scalable micro payments of ERC20 compliant tokens.</p><p>MIT technology review has a good <a href="https://www.technologyreview.com/s/609621/cryptocurrency-exchanges-can-be-pretty-sketchy-places-the-solution-a-blockchain-of-course/">article</a> on problems of centralised exchanges, but decentralised exchanges also have their own fair share of problems to deal with. Many of these problems arise from sacrificing transaction speed for settlement on-chain, which might take up to an hour. Not having high enough latency results in less users and ispo facto, less liquidity. Technologies like the Lightning Network and the Raiden Network are on the verge of solving this. Combined with Atomic Swaps, they enable decentralised exchanges to be instant, which is one of the biggest argument against using decentralised exchanges nowadays.</p><p>The video demonstration above shows a proof of concept of this principle: an instant, trustless exchange of two ERC20 tokens on the Raiden Network.</p><p><strong>If you are keen to try all this on your own, check out our </strong><a href="https://github.com/ExchangeUnion/Raiden-ERC20-Swap-PoC"><strong>guide on github</strong></a><strong> detailing all the steps to execute the swap yourself! Happy swapping! </strong>🍻</p><p>We would like to thank the<a href="https://brainbot.li/"> Brainbot team</a> and all the other open source contributors who are working on Raiden’s technology. We are amazed by its fast paced development and intend to actively participate inthe development of Raiden Network and play our part in the building new global financial infrastructure.</p><p>PS: We are still searching for talented folks to join us in building Exchange Union. Take a look at our <a href="https://www.exchangeunion.com/jobs">open</a> positions and join our <a href="https://www.forbes.com/sites/kashmirhill/2012/05/24/sheryl-sandberg-to-harvard-biz-grads-find-a-rocket-ship/#3567b3d23b37">rocketship!</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a9890c4957d7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Tensorflow for absolute beginners]]></title>
            <link>https://medium.com/data-science/tensorflow-for-absolute-beginners-28c1544fb0d6?source=rss-e060bb5c25e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/28c1544fb0d6</guid>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[tensorflow]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Ankur]]></dc:creator>
            <pubDate>Mon, 13 Mar 2017 01:06:26 GMT</pubDate>
            <atom:updated>2017-09-08T06:05:48.524Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>Follow <a href="https://www.tensorflow.org/install/">this</a> official Tensorflow guide to install Tensorflow</blockquote><p>If you follow me on <a href="https://twitter.com/dopetard">Twitter</a>, you’d know that I’m a big fan of open-source community. I’ve always admired researchers like <a href="https://medium.com/u/ac9d9a35533e">Andrej Karpathy</a> who take their precious time out to blog and write tutorials, which has helped me immensely over the years. So, when Google China reached out to me with a volunteering opportunity of writing tutorials to help their new online education initiative in China, I knew it was time for me to give back. This is first part of tutorial which will build upon very hands on knowledge of using Google’s Tensorflow library. I’ll assume basic familiarity with Python programming language and basic high school level of mathematical maturity. In case you aren’t familiar with Python, but know how to code in general, you should be able to pick up the syntax and follow along. Alternatively, you can look into <a href="https://www.codecademy.com/learn/python">Codecademy</a> to learn Python. I’ll start by building and training perceptrons performing boolean logic like AND, OR, XOR and by the end of this tutorial series, we will be classifying cats! or dogs! or even <a href="http://letmegooglethat.com/?q=angelababy"><em>Angelababy</em></a>! using a technique called <a href="https://en.wikipedia.org/wiki/Inductive_transfer">Transfer learning</a> with Google’s pretrained deep neural network. This blogpost will be later translated into Mandarin for better accessibility to Chinese students.</p><p>Before learning Tensorflow, Let’s start with, what is a neural network? And Why the hell everyone can’t stop talking about it?</p><p>The human brain consists of billions of neurons which are interconnected by synapses. If “enough” synaptic inputs to the neuron fires, then the neuron will also fire. This process is called <em>thinking. </em>Here’ a question,<em> c</em>an we model this phenomenon, using our laptop and a bit of math, so that we can make our Machines, “<em>think</em>”? Absolutely Yes!<em> iso facto</em>, <em>Artificial Neural Networks. (</em><a href="https://en.wikipedia.org/wiki/Skynet_(Terminator)"><em>SKYNET</em></a><em> anyone?) </em>These networks started to perform really well in recent years- partly because of insane amount of data available these days and partly due to <a href="https://en.wikipedia.org/wiki/Moore&#39;s_law">Moore’s law</a>. For the purpose of building intuition, I like to think of Neural Networks as <em>Universal Function Approximators. </em>If you know Calculus, This analogy will be easily apparent to you. Think of it as approximating a curved function by putting together many line segments. Everything from Google Translate to FB/Weibo uses this technology. They are used in cancer detection, drug discovery and even <a href="https://blogs.nvidia.com/blog/2015/02/24/deep-learning-drive/"><em>Self-driving cars</em></a><em>! </em>Chances are, you come across some form of neural network every single day in your life. Google’s Tensorflow library provides very high level abstraction for implementing NN’s. If you really want to understand Neural Networks well- You should know about how algorithms like <a href="http://neuralnetworksanddeeplearning.com/chap2.html"><em>Backpropagation</em></a> and various A<a href="https://en.wikipedia.org/wiki/Activation_function"><em>ctivation functions</em></a> works, It’s a good idea to write them from scratch in pure Python/numpy once to gain better intuition. We can always use open-source libraries like Tensorflow to prototype and build them very fast, hence avoiding writing error-prone algorithms from scratch. Lets start building a very basic single layer perceptron. A basic form of neural network that can model the logical AND gate, Shall we?</p><p>Dont worry, I’m going to explain each line of code below. I’d advise you to open this post in two windows, so you could look at the code and explanations simultaneously. Before that let’s talk about about how Tensorflow works; To make our code execute faster and avoid drawbacks of languages like Python, which inherently trades speed with readability, Everything in TensorFlow is based on creating a computational graph. Think of a computational graph as a network of nodes, with each node known as an operation, running some function that can be as simple as addition or subtraction to as complex as some multi variate equation. Then the graph is executed in a Tensorflow session with fast C++ backend. Variables in TensorFlow are managed by the Session.</p><p>Now we know enough to dive in and get our hands dirty with code, which is the fastest way to learn.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/4601a13cbd55ea120a607fb17fb5f0e5/href">https://medium.com/media/4601a13cbd55ea120a607fb17fb5f0e5/href</a></iframe><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/be198933db47653b70f6dd5034b07f51/href">https://medium.com/media/be198933db47653b70f6dd5034b07f51/href</a></iframe><p>Line 1: It simply imports the Tensorflow library where all the awesomeness resides.</p><p>Line 2: We simply assigned the value of True to 1 and False to -1, quick brain teaser: Can we assign Value of False to Zero?</p><p>Line 3: <em>Bias</em> is a constant we choose (typically 1 or -1) and let our neural net decide the weight with which the constant term should be multiplied to get the desired function we intend to approximate. To learn more about bias; refer to this <a href="http://stackoverflow.com/questions/2480650/role-of-bias-in-neural-networks"><em>Stackoverflow thread</em></a></p><p>Line 4 to 9: This is input data with with we want to train our neural network. All four possible values of True and False being TT, TF, FT, FF with the added bias term ‘b’.</p><p>Line 10 to 15: This is output data of our training set, T, F, F &amp; F respectively. Our neural network should be able to look at the training input from Line 4–9 and training output from line 10–15 , henceforth be able to approximate the desired ‘<em>AND’</em> function</p><p>Line 16: Input (x) and output (y) training data will remain constant so our neural network is going to approximate the function by approximating the ‘weight’ (w) , In this case 3 * 1 since 3 inputs corresponds to 1 output, NN will keep adjusting the weight to find the optimal value during the training process, hence its’s a <em>Variable</em> object. The values are initialized to normally-distributed random numbers, for that reason we used <em>tf.random_normal </em>module. I hope the commonly used equation for neural networks “<em>y = Wx”</em> makes more sense to you now.</p><p>Line 18–22: Now we can define our model. Starting with the step function, step activation functions are useful for binary classification schemes. In other words, when we want to classify an input pattern into one of two groups, we can use a binary classifier with a step activation function. Note that when using deep neural networks, step function will not be useful, we will introduce non-linear activation functions like <em>sigmoid</em>, <em>tanh</em>, <em>relu</em> and my favourite <em>leakyrelu</em> (which happens to be my <a href="https://twitter.com/leakyrelu">twitter</a> handle). This piece of code just returns 1 if parameter to the function is &gt; 1 and returns -1 otherwise</p><p>Line 24–26: After defining our step function, We can simply define output, Error and Mean squared error in three lines. Error basically computes, how “off” our model output predictions are from the real “output” of training set. We will then proceed to compute the Mean-squared error, Why? Well, There is very beautiful piece of math behind it, You’d already know if you took Probability 101 class in college, but <a href="https://www.probabilitycourse.com/chapter9/9_1_5_mean_squared_error_MSE.php">here</a> is a nice short tutorial explaining it.</p><p>Line 28–29: The evaluation of certain functions like weights (w) will help adjust the value of our predefined <em>Variable, </em>So, in this piece of code- we computed the desired adjustment (delta) based on the error we earlier computed. And then proceed to add it to our weights.</p><p>Line 31–32: The model has to be evaluated by a TensorFlow session, which we instantiate before initializing all variables to their specified values, Remember? We got to run our model using fast C++ backend of Tensorflow. The magic is behind <em>sessions</em></p><p>Line 34–40: We can now run our model through training epochs, adjusting the weights each time by evaluating <strong>train</strong>. Since we’re using a binary output, we can expect to reach a perfect result with a mean squared error of <strong>0</strong>. We will reach it very quickly. Note that <strong>Session.run</strong> will return the result of whatever is evaluated till then. On each epoch, we evaluate <strong>mse</strong> in order to track progress, and <strong>train</strong> to actually adjust the weights.</p><p>Now go ahead and run this code, You will be able to arrive at a solution very fast in just few epoch because of the simplicity of this network.</p><p>I’d leave it as an exercise to the reader: Implement a ‘<strong>OR</strong>’ and ‘<strong>XOR</strong>’ gate by playing with the code (<em>hint: change the training_output dataset to the desired output results</em>).</p><p>Congratulations! You just coded and trained your first Neural Network. Till next time, adios.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=28c1544fb0d6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/tensorflow-for-absolute-beginners-28c1544fb0d6">Tensorflow for absolute beginners</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>