I am an Electronics and Communications engineer by qualification. What I have learnt technically by completing a degree in electronics is highly debatable. But the fact that I have learnt a lot in life from Electronics is not debatable :) Today I would like to share what Sine wave taught me. https://upload.wikimedia.org/wikipedia/commons/0/0c/Vale_sinusoidale.gif 1. Keep it simple. Sine wave is the simplest waveform that can be generated naturally. It says keep simple things simple. Do not get into the habit of aking imple things omplex, which I generally call MSc degree. M S C Whatever is natural is simple and long lasting. 2. Develop a long term view When delta t is very small we can assume the wave to be smooth. If we take a look at short ranges it appears to be increasing monotonically (Eg 0 to PI/4 a) or monotonically decreasing in the other ranges. But when you consider the whole range it becomes clear that the wave has both positive and negative slopes. It says when we consider in small patches it may either appear as a string of good times (or a string of bad times) but when you consider the whole life you will get to know that every trough is followed by a crest and every crest is followed by a trough. So be humble when you are going upward and be optimistic when you are going down. 3. Keep going Cosine wave is a variation of a sine wave. It is just a sine wave with phase shift. It starts with peak value at the beginning, whereas a sine wave starts with a zero value. Some may start with a high and some may start with a low. But each one of us has got to go through highs and lows. 4. Break it down http://www.peterstone.name/Maplepgs/images/Fourier_series.gif Fourier wave says that any complex waveform can be broken down into sine waves ( as increases the approximation get better) of different frequencies and amplitudes which in turn gives rise to the complex waveform when they are superposed. n n Every complex and difficult task can be broken into a number of simple and easy tasks. But in real life the actual value should not be very high, it must be optimal. This is true up to a certain extent in electronics also. As n increases the approximations become more accurate. But after a certain value the increase in accuracy is not worth the extra time and efforts. 5. Your altitude matters The average output of a sine wave is zero. So just by increasing the amplitude or frequency you cant increase the total power output. The only way of increasing the average output power is by adding a positive offset. Since you know you have to go through the highs and lows make your attitude independent of them. Your output {or success} is dependent upon the offset and that is nothing but the altitude. So increase your altitude in the positive direction. Remember that your attitude determines your altitude. So, develop an attitude where in you can handle the crests as easily as you handle the troughs.

Waves

Declutter Your Mind

Read My Stories

Simplicity and Longevity: Life Lessons Sine Wave Taught Me

My Bootstrapping Journey Week 5 — Retrospective Read last week’s article here. Read the first part of this series here: .This is my product: . Read my entire journey . In this series, I’ll try to bring my product to profitability. What I Learnt Building a SaaS Find Better Questions here I’d like to talk a little more about individual customers, but it would be too much of a pain to ask them individually whether I can write about them here, and I respect their privacy. But I can show you anonymized data. This is a look at my CRM: Let me explain these stages: Sign Up in App:They downloaded the application and registered In Trial:They started the trial. Almost everyone converted to this stage since I changed the trial to no credit card required. I might still revert this in the future. But for now, I want as many people as possible to try . Find Better Questions Used the Product:People are tagged this way as soon as they create a list, which is the first and only thing you can do after you created an account. Paying Customer:I don’t have to explain this one. There are pretty much no customers, which is a pain, but even worse, there are more than 10 people who downloaded the application but never actually used it. I wonder how that can happen. They download 50MB+, but don’t use the app? This needs to be improved. How to Have People Use Find Better Questions Here are the things I did so far to improve the situation: Ask them why they don’t use the thing.Even if they didn’t answer, that’s good feedback. They’re simply not interested. I fixed a bug.This is embarrassing. There was a bug which didn’t allow you to create a list. Two people fell victim to this. They never returned thus far :( I guess that they deleted the app. It would be hard to get them to download it again. Improve the onboarding a little.I just changed the text a little. I think I’m on the right path to get people to actually use the thing. If this continues, I’ll have to make more major changes. What else did I do this week? Finished the new animation at https://findbetterquestions.com/ Answered two Quora questions:1: 2: What are some examples of great marketing? Why is entrepreneurship considered a lonely journey? I sent some cold emails. I signed up to . I like it thus far. Bluetick.io Write many more emails. That’s pretty much it for this one. → Week 6: Coming at the end of this week. Make sure to 👋 ← Week 4 follow me

How I Convert Trials to Customers

An Android Developer’s Roadmap to React Native - Part 4 Photo by on Edgar Chaparro Unsplash If you come from native development background, you might be aware of a feature that both Android Studio and XCode provide to create different applications from the same code. These are known as flavors in Android Studio and targets in XCode. The idea is to create different packages with same core functionality but with minor difference in features. This can include different color schemes, branding or some functionalities unique to a package. A common example is creating a free and paid version of a same application, wherein the paid version has same core functionality as the free version with some added functionalities and features. Since the major functionality is the same there is no need to create different projects, but handle flavor/target specific changes at run-time. I hope you get the idea. If not follow these links to learn more. _Get started using Android Flavors to easily create variations in your app_proandroiddev.com Advanced Android Flavors Part 1 — Building White Label Apps on Android _How to create multiple applications that differentiate and share parts of the source code_medium.com Xcode “Targets” with multiple build configurations For simplicity, I will be using the Android term _flavors_ to represent different product builds in React Native. Now, lets move to the problem at hand The project I was working on relies heavily on build flavors, and React Native does not have inbuilt functionality to handle the same. Sure, you can create different flavors from but that only decides the build process, not the actual run-time changes that need to be made. So here is what I came up with, to bridge this gap between native and React Native. react-native-cli Setup Step 1: Configuration ( Native ) The first step is to setup product flavors in Android Studio and build targets in XCode. For the purpose of the tutorial, our application will be called . This application has two flavors: and . You can read the above mentioned articles to help you setup the configuration on Android Studio and Xcode. NoteTaker lite pro Step 2: Native Modules Once we have setup the flavors on the native side, we need a way to pass the product flavor identifier from native to react-native at run time. This, definitely requires us to create native modules or bridge methods for both Android and iOS. For simplicity, we will keep our flavor identifiers as strings. In Android, the flavor in stored in the and for iOS you can use from BuildConfig TargetName info.plist Take a look at the following documentation if you do not know what native modules are in React Native. _Sometimes an app needs access to a platform API that React Native doesn't have a corresponding module for yet. Maybe…_facebook.github.io Native Modules · React Native Step 3: Flavor.js Next, we create a JavaScript file that serves as a utility class for flavors in React Native. This file exposes a method to get the current product flavor from native and a variable that stores this information once it is fetched from native. Step 4: Colors.js Since our and versions have different color schemes, we create a file, that contains the flavor specific colors. Also this file exports a method to get the relevant colors based on the product flavor. lite pro Colors.js Our setup is now almost complete. We just need to call the build flavor method at the start of our react native application. But there is a catch. Since the build flavor is asynchronous, we need to ensure that the first component is mounted only when the flavor has been fetched from native. For this, we pass null in render method of our component initially, and on receiving callback from method we re-render the component by using buildFlavor setState Implementation Implementing flavor specific features Now that our flavor has been fetched from native and have access to our product flavor in any component, writing flavor specific code is a piece of cake. In fact, we have already seen such implementation in method. Consider a feature of adding images to notes which is only available in the flavor getThemeColors pro Implementing color schemes When defining styles for a component you can use two approaches:1. Directly apply styles using the styles prop2. Use StyleSheets The first method poses no problems since the styles are evaluated and applied at run time when that particular component is mounted. Thus the method is evaluated properly and we get the required colors. getThemeColors However, when using StyleSheets, the styles are evaluated when the file is loaded into the JavaScript run time environment. This means that even if the component is not mounted its styles are already evaluated. Thus, in our case, the delayed render of the component in after the flavor is fetched, fails to serve its purpose since even though the component is not mounted, its styles have already been evaluated. Thus, returns the wrong set of colors. To overcome this, we need to create StyleSheets for the component only when it is going to be mounted. We achieve this by creating StyleSheet in lifecycle method. App.js getThemeColors componentWillMount So now we have product flavor specific features and color schemes. The only drawback with this setup is that the styles and color schemes are now generated at run time rather than at compile time. Not that it has much of an impact on performance, but still, is a step behind. If you have any ideas to improve this setup, please feel free to share them with me either here in responses or via e-mail. This is the fourth part in the series “An Android Developer’s Roadmap to React Native”. You can read previous articles in the series by following these links: Part 1: Part 2: Part 3: https://medium.com/@rohanx96/an-android-developers-roadmap-to-react-native-part-1-5ec20cf93757 https://medium.com/@rohanx96/set-up-your-own-react-native-ide-f7d52c90da43 https://blog.usejournal.com/redux-an-intro-and-beyond-92d694bc314f

Facebook

Target

Setting up Android like product flavors in React Native

An infant being administered a Vaccine It was a Friday night, 20th July & Zhi Ruo world came crashing down. She was alerted to an article that was spreading like wild fire on Weibo by her friend. Her friend made sure that she read the article immediately by calling her mobile & asking her to check the article right away. She didn’t give any more details. As she opened the link, her heart sank. She became increasingly numb as she read each word. ‘Please!’ She kept murmuring. ‘Not my child! Please! No!’ as she recalls the horror that unfolded in her life over the past week. A Chinese vaccination manufacturer, Changsheng Biotechnology, was fined on July 20th 2018 for producing 252,000 defective vaccines for infants. She began to frantically search her daughter’s medical records. As she retrieved her file & cross referred the DPT vaccine serial number with the batch numbers on Weibo, she found that her infant was adminstered a defective DPT vaccine. She called her husband who can only spend time with them on alternative weekends as he works in Shanghai, 5 hours train ride away. He rushed back home to Zhi Ruo & his daughter vowing action against the company. Anger is the over riding emotion amongst all parents across Shandong province, China. Outrage spread over the Chinese social media as regulators & the government tried to contain the fallout over information that once of the countries largest vaccine manufacturers was making & distributing defective infant vaccines. Both the Prime Minister & the President of China made public comments that they will not tolerate this occurence & that punishments will be handed out to those who violated the safety standards. Some posts on social media appeared to be removed. According to a project run by the Journalism and Media Studies Centre at the University of Hong Kong, which monitors censorship on Weibo, China’s Twitter-like service, the Chinese word for “vaccine” was one of the most restricted on Sunday and Monday. Fu King-wa, an assistant professor who heads the project, known as Weiboscope, said also that the rate at which posts related to the scandal were being screened peaked on Sunday. For every 10,000 posts made across the 120,000 accounts monitored, an average of 63 were blocked, he said. Yet, the hashtag #Changsheng vaccine case# had been viewed over 470 million times as of Monday, 23rd July. Changsheng Biotechnology is owned by Chinese billionaire Gao Junfang. Last week shares for Changsheng, listed in Shenzhen, dropped 10%, the maximum amount . On Monday, trading of its shares was temporarily halted. The company has since lost 40% of its share value last week, when the case has become public. they are allowed to fall Shandong is a coastal province & located nicely halfway between Beijing & Shanghai. The modernisation that swept China from the 1970s served well to modernise the coastal provinces including Shandong. Living standards grew rapidly. Luxuries, modern appliances & gadgets appeared. The western way of life was adapted to the Chinese culture to create a new way of life. But substandard practices are rife. This applies to both Manufacturing & Services. . According to the Jilin Food and Drug Administration, where Changsheng Bio-technology is based, the company sold about 252,600 substandard DPT vaccines to the Shandong Centre for Disease Control and Prevention, the agency in charge of public health in a province of about 100 million people. The drugs were administered to children as young as three months old. China’ s vaccination system is a nightmare. It has been affected by corruption, staff shortages & weak regulations/enforcement What stunned many parents & the society is that this is not the first instance of such an incident. In March 2016, in the same province led to 37 people getting arrested. a similar incident in November 2017, Changsheng Bio-technology & another company failed to meet the national standards for 2 batches of DPT vaccine. The official watchdog reported that though the vaccines were ineffective at immunisation, there would be no danger to human health from administering them. On July 15th 2018, a surprise inspection by the official watchdog led to the uncovering of forged data related to the defective vaccines. The manufacturerer’s license was immediately revoked. The company recalled the defective vaccines the next day. Fines were announced a couple of days later & the story got out to social media spreading like wildfire. Manufacturing & Consumer Trust TO TRUST or NOT ? As a consumer, Trust is built within the intermeshed framework of standards, regulation, enforcement, quality control, , brand reputation & feedback systems supply chain In the early days of any industry or service, regulators play a guiding role in establishing dependability from manufacturing to consumption. As a critical support function, supply chain is automatically regulated as part of the process. Quality checks are established at several points of the supply chain to test for variance from the standard. Bad samples are weeded out. Depending on the seriousness of the product, this could mean an entire batch being recalled. For example, Passenger Car Brakes or Air Bags cause recalls on a massive scale. Such instances are not uncommon during the initial days of establishing consumer trust. But with feedback loops, regulatory policing & tighter quality control, recall frequency decreases & consumer trust increases over time. In this case, despite the presence of several checks & balances, Quality Controls were bypassed & records were forged. Public trust had been misused. What if there was a mechanism that can do the following? Ensure that records are tamper proof Ensure that records remain open for public inspection anytime Display Audit results on demand Track & record the vaccine from the time it was manufactured, stored, tested, packed, shipped and up to the time it was administered? Track & Record any changes in storage conditions (Ideal temperature, Away from Strong sunlight etc) Record & Display all test results along with the testing authority’s certificate to anyone with an internet connection Well, this is possible when we implement a solution that leverages IoT data. This debacle in China could have been averted. Faking of records could have been impossible had there been a blockchain solution in place to record & track the supply chain Blockchain Blockchain is one of the better technology solutions that are currently available for supply chains The case for Blockchain integration into Supply Chains Though Blockchain was originally intended for financial transactions, businesses of all kinds are getting creative with the blockchain ledger, as it can be used to record information, track, and verify trades of virtually anything that holds value. Depending on the product, the supply chain can span over tens or even hundreds of stages, multiple geographical locations, a large number of invoices and payments, have several individuals and companies involved, and stretch over months / years of time from start to end. Since every transaction is captured & recorded on a block and across multiple copies of the ledger that are distributed over many nodes, it is highly transparent & easily accessible to anyone with an internet access. It is also highly secure since every block is linked to the one before it and after it. Since every node has a copy of the entire blockchain from the very first block to the most recent one, it is highly improbable for any tampering to occur after the data has been written on the block. There is not one central authority over the blockchain, though certain permissioned types of blockchain have some elements of the legacy spoke & hub design. Supply Chains, even complex ones have taken to Blockchain like ducks to water. Walmart uses blockchain to keep track of the pork it sources from China. The blockchain records where each piece of meat came from, processed, stored, its temperature and its sell-by-date & have Blockchain Solutions for Diamond supply chains. Though, takes the limelight in implementing a blockchain solution for ensuring that blood diamonds are not traded in their value chain. Blockverify.io Everledger.io De Beers Dole, Nestle, Unilever, Driscoll’s, Golden State Foods, Kroger, Wegmans are all leveraging the Blockchain for their perishable produce supply chains. Blockchain has been touted as the counterfeiter’s worst nightmare Public health & safety are 2 ideas that can never or should be compromised. Blockchain allows tracability, Proof of custody & transparency as its core properties. Impact Areas in the Supply Chain the quantity and transfer of assets — like units, boxes, containers, etc. — as they move across the supply chain Recording purchase orders, change orders, receipts, Bills of Lading, packing lists, Invoices, shipment notifications, inventory data or other trade-related documents. Tracking can also be done internationally over border customs Tracking or verifying certifications or certain properties of physical products; for example determining if a coffee is organic or fair trade Assigning physical goods to serial numbers, bar codes, digital tags like RFID, Chassis numbers, Vehicle Identity Numbers (VIN)etc. Linking information about manufacturing process, assembly, delivery, and maintenance of products with suppliers and vendors Sharing Circling back to the story at heart of this article, Changsheng Biotechnology would have found forgery of its records impossible had there been a Blockchain based solution. Of course, the past cannot be undone but the course of future remains firmly in our hands. We have a duty of care towards our next generation & blockchain can help us deliver it. If you are wondering what is Blockchain’s best use case, there is none better than improving Public Health & Safety by using Blockchains in Supply Chains

Assembly

Kroger

Twitter

Babies, Supply Chains & Blockchains

Alongside

BUNCH

Google

Netflix

Tesla

Trello

Uber

Visa

YouTube

An In-depth Review of Andrew Ng's deeplearning.ai Speciliazation

The Double Spending problem, Consensus mechanisms on distributed networks, and Establishing “trust” in the “trustless” network! The Double Spending problem If Alice has $10 in her wallet, she should be able to buy goods only up to $10. So for example if Alice goes to a store, and there is a lamp for $10 and table for $10, she should be able to spend the $10 only on one of them. The double spend issue does not arise in day to day cash transactions since the exchange of currency for the goods happens together (i.e. it is a centralized system). So Alice gives up the $10 and gets the lamp, and now does not have the $10 any more. However the issue is a little different in distributed systems. In case of distributed systems, the record of the transaction is broadcast to all nodes in the network . (i.e. Alice broadcasts the transaction on the network so that every node on the network is made aware that “Alice has used up $10 to buy a lamp”) Each node records this transaction message and then passes the message onto the next node in the network, and the process continues until all nodes in the network have recorded the message . “Alice has used up $10 to buy a lamp” However as this message propagates through a large network following issues may arise. Messages may take different paths as they propagate through the networks and consequently arrive at different nodes at different times. Due to node failures, some nodes may be unable to forward the message to the next node and the message may be lost. Thus, we are left with a situation where at a given time, certain number of nodes , and certain number of nodes are are aware that Alice has spent the $10 on the lamp not aware that Alice has spent the $10 on the lamp. For the nodes that are not yet aware that Alice has spent the $10, i.e. the message has not yet reached them; they are still under the impression that Alice has $10 in possession to spend on anything else. Thus, it is possible for Alice to broadcast another message to the network that and if this message reaches a node before the first message i.e. , they will now assume that Alice has used up the $10 on the table instead. “Alice has spent $10 on the table ” “Alice has spent $10 on a lamp ” This opens up the possibility of Alice being able to spend the $10 on the table, and the $10 on the lamp; which should not be allowed as Alice only has $10, not $20. same This is the double spend problem. The double spend problem arises in distributed systems since the transaction messages take time to propagate through a large network. Because of differences in propagation time over the network, there is no guarantee that the order in which a transaction message arrives at a node is the order in which they were created. Note: It may be argued that one could include a universal timestamp on the message data, along with the hash to ensure data integrity, and this would easily address the issue of transaction messages arriving at different times at a particular node. However, Alice could easily fake her timestamp before signing the message, and put in an earlier timestamp for the second message, creating a whole deal of confusion on the network. At a deeper level, the network is now left in an inconsistent state where some nodes have validated that and others have validated that . “Alice has spent $10 on the lamp ” “Alice has spent $10 on the table ” In order to resolve this inconsistent state of the network (where few nodes have validated one transaction and others have validated a conflicting transaction), we need some consensus mechanism on the order of the transactions, thereby bringing the network back to a consistent state. Consensus mechanisms in (Distributed Ledger Technologies) DLT and Blockchains “The truth isn’t a thing of fact or reason. It is simply what everyone agrees on.” ― Gregory Maguire, Wicked: The Life and Times of the Wicked Witch of the West There are two ways in which this consensus can be established, where each node on the network is known and each node votes “yay” or “nay” on the validity of the transaction. Consensus is achieved by tallying up the majority votes and endorsement policy (eg. PBFT algorithm). The endorsement policy may dictate that the transaction is valid if 2/3 of the nodes on the network validate “yay” on the transaction. Voting based consensus (as in case of trusted consortium nodes or private distributed network, eg. Hyperledger): where only one member of the network, chosen based on PoW or PoS, is allowed to decide if the transaction is valid or not, and this decision is agreed to by all on the network. In effect whoever wins the competition (wins the lottery) and the network agrees that transactions validated by the winning node are valid. Lottery or Competition based consensus (as in case of public or trustless node network, eg. Ethereum): The “competition” by which the next node is elected may be in the form of solving a cryptographic puzzle eg. PoW, or may be in the form of increased chances based on contribution of investments into the network eg. PoS. Consensus mechanism (either voting or lottery), are a way for the network to determine which of the conflicting transactions are to the be considered valid by the entire network. All the nodes of the network then invalidate the conflicting transaction, and proceed only with the consensus agreed upon validated transaction. It is interesting to note that the validated transaction may not actually be the “correct” transaction out of the conflicting transaction. In our example if the group consensus votes “Alice spends $10 on the table” as the validated transaction, the “correct” transaction i.e. “Alice spends $10 on the lamp” will get invalidated by all the nodes of the network. In fact the purpose of the consensus algorithms is not to ascertain the “truth” as to which was the “correct” transaction between the two transactions. Instead consensus algorithms are merely a mechanism to prevent the double spending problem in distributed networks (i.e. In our example using the consensus mechanism we are ensured that Alice can spend that $10 once and only once); and ensures that only one of the conflicting transaction is agreed upon by the entire network, and the any conflicting transactions are invalidated by the entire network. Establishing “trust” in a “trustless” network Bodhi: “You don’t trust me?”Johnny Utah: “You gotta earn trust.” — Point Break (1991) Proof of Work (PoW) algorithm is a means by which a winner emerges from within the network (based on the solution to a cryptographic puzzle), and that winner decides which of the conflicting transaction in the network are valid and to be a part of the next block in the blockchain. Could not any node be picked at random and designated the next winner (true random lottery), and this node be elected to decided which of the conflicting transactions are valid. However the question begs, why do we need nodes to compete with each other to solve a complex cryptographic puzzle in the first place to select a winner? i.e. Why do we need a complex PoW? The answer to this is as follows, If lottery winners are picked without it being computationally intensive to add blocks (or some token of computationally intensive i.e. PoS), then it becomes easy for anyone to add the next block on the Blockchain. This means that several parties can add their versions of block to the blockchain, and the one with the greatest single computing power will be able to scale up the block chain, and have the longest blockchain. The genius of Satoshi Nakamoto’s solution to the issue of “establishing trust in a trustless network”, is to ensure that it is impossible for an individual, or a even a large group of individuals (as long as the group size is <50%), to computationally outpace the collective network, i.e. to create new blocks to add to the blockchain, and (continuously) maintain the longest chain on the blockchain. Thus, the basic idea is to make it and introduce some mechanism to do this. The most common of these mechanisms is the PoW algorithm. computationally hard to add blocks to the blockchain However there is also Proof of Elapsed Time (PoET), which needs node to computationally “wait” for a period of time, before the next block is added to the blockchain, thereby again artificially making it computationally “hard” to add blocks to the Blockchain. In case of PoS, the escrow of tokens as a mechanism to be chosen as the next member to create a block, again makes it computationally hard for any one individual to invest enough number of tokens to outpace the collective network. 👏 Found this post useful? Hit the button below to show how much you liked it :) Follow me on Medium for the latest updates and posts! Read Next: Making sense of the Blockchain Landscape in 2018 Other Articles by Prashant Ram: BlockChain Demystified: Looking Under the hood of the BlockChain Understanding Digital Signatures in Blockchain and Cryptocurrency The Distributed Autonomous Organization (DAO) and how Blockchain/AI can take over the network — and why you cannot stop it!

Ledgers

Why do we need Proof of Work (PoW) in Blockchain?

Different

How many claps does the average Hacker Noon post get? 604.

Last fall I was hired to build an economic model for as part of my . Since then, I encountered many people who expressed interest in this model, so I decided to share the basic theory and implementation. token-curated registries consulting job Contents 1. Philosophical introduction2. Model summary3. Agents and motivations4. System statics5. System dynamics: simulation in R6. Conclusion 1. Philosophical introduction The great expositive summary of can be found on , so I will dive into my points straight away. Token economy Wikipedia As I see it, a token economy is an attempt to quantify life and give a price to everything. This aligns perfectly with classical economic thinking, where even externalities, such as air pollution, can be compensated by sufficient taxes. Wikipedia says that tokens have been successfully used at schools and psychiatric hospitals, but the recent resurgence in interest has begun with the rush. Since the creation of universally accepted token was problematic before, nobody tried to tie all services to some price. However, the open source blockchain algorithms made this process really easy and suddenly many people decided that it is a good idea to (read ) literally everything. cryptocurrency tokenize monetize Being an economist myself, I believe that this thinking is terribly wrong. I understand where it comes from — everybody took at some point and was inspired by elegance of models. But I also know that is not the only mechanism out there. , , all exist for a reason. Not every is . Sadly, not everybody takes advanced econ courses, while almost everybody wants to profit from ’s. Econ 101 invisible hand free market Centralized matching mechanisms social planner’s problem lotteries equilibrium efficient ICO Not every equilibrium is efficient Of course, tokens can be justified in appropriate cases. Actually, the great working area for tokens is online games, where every crystal is given a price and has an obvious value to players. In general, whenever any kind of labor is involved, a money making procedure is good. Otherwise, it is some kind of . rent seeking Ryan Selkis of the difference between and , and explained why he believes the former will be worthless in the long run. I will go even further and claim that most so-called utility tokens are as well bubbles and don’t give any extra utility. Messari greatly summarized bubble utility tokens Examples Below I reviewed a couple of examples that appear in the first results page when you google . utility tokens claims to help users to prevent social networks from collecting their data “for free” and sell it for money instead, of their information. This certainly sounds fun, but it totally neglects the fact that Facebook or Youtube are already providing a great service for free in exchange for collecting users’ data, moreover they offer all kinds of privacy settings for all their users. It is a fair trade. What Datum wants is to take those services for granted and try to profit from that, providing no utility whatsoever. Datum i.e. “take control” wants to make another LinkedIn but to replace reputational structure with token structure and give tokens for truthful references. That’s wrong on many levels, including: Talenthon impossibility to quantify references possibility that worker can change, making references obsolete the assumption that businessmen would care about tokens more than they care about their reputation This is exactly what I meant above, when I said that monetary relationships are not always efficient. And finally, these people hypocritically encourage openness and decentralization while using proprietary algorithms. The right decision is to only decentralize those services that actually support the decentralization. Another issue with is that historically every decentralization was tightly related to . All examples of successful decentralization used to be altruistic and didn’t involve monetary goals. Decentralized apps (DAPPs) I care for are etc. token-backed decentralization disruption, hippie philosophy, anarchy BitTorrent, Wikipedia, Linux, GNU Project, OpenStreetMap Decentralized projects used to be the very definitions of , where corporations like would be put to shame by the community for not contributing code back to but still get away with it, and nobody would attempt to financially incentivize them to do so. trust-based systems Apple BSD These old days, which most of us still remember, should raise everyone’s concerns about the longevity of trustless structures. There are, however, appropriate use cases for tokens and some developers get this. does this well by aiming at a peculiar area — ranking ads services. This is compatible with token system and is not something that has already been done on a massive scale. AdChain does even better job by providing a service of renting storage space, which works at the same time as a mining tool via . They have problems with modelling incentives but at the end of the day they provide legit utility and thus are . Filecoin proof-of-storage not bubbles In this article I will try to build an economic model for exactly those . utility tokens 2. Model summary This economic model of tokens is very generic, so that it can work as a blueprint for future token creators and as a reasonable abstraction for applied economists. Ideally, this model can also serve as a guideline for writing whitepapers. In case you have skipped the previous section, let me remind you that are any tokens that produce a legit service for consumers and are not tools for middlemen or speculators (although such agents do exist in our model). utility tokens primarily Since tokens aim at structures, we will need some kind of proof that every corresponds to a single person instead of some . Since I am a huge opposer and since most of the tokens are built on Ethereum platform, which is anyway, I will use the latter in the model. trustless node botnet proof-of-work moving towards proof-of-stake The service is paid for in fixed number of tokens, not in cash. Pricing service in a fixed token rate will correctly capture the proportionate relationship between demand and price — the greater the demand for the service, the more people will buy tokens, the price of tokens will increase, and so will the cash equivalent of the service; and vice versa. There are consumers, investors, and honest and malicious miners in our model, each having different and . Consumers and malicious miners want token price to depreciate, while honest miners and investors want token price to appreciate. motivations types The incentives should be modelled in such a way that no player will want to deviate from the honest sutainable system as long as there is an honest majority ( ). We will assume the ’s voting model, as it satisfies the and as long as of votes are honest. We will model the resulting economy but not dig into the technicalities or revisit what and have already spent some time to model. conditional equilibrium Ethereum Casper checkpoint tree accountable safety plausible liveness 2/3 Vitalik Buterin Virgil Griffith 3. Agents and motivations In a general there are four types of agents: utility token economy Honest miners Honest miners are people with domain expertise, who actually do the job and create the utility. They are incentivized by a system to do the best job they can to earn more tokens as long as they constitute of the token holders. 51% active They do not play any , they naively vote for the correct version in every quorum. They earn (mine) rewards as long as malicious miners don’t take over, and at the same time they provide a utility to customers. Therefore the main quantitative feature of honest miners is their proportion among all miners. strategies i.e. Main quantitative feature of honest miners is their number. The corresponding issue is that it is impossible to tell for certain which miners are honest and which are not. Therefore customers and investors choose their strategies based on the expectations. The prior probabilities which form these expectations are exogeneous in this model but can be endogenized. If you are a researcher interested in extending the model in this area, please , so that we can discuss and share the ideas on that topic. contact me Customers Customers are people who use and benefit from the service created by the and pay for it using tokens they bought for cash. honest miners As mentioned in the introduction, utility tokens only work good in peculiar services like job referencing or ads ranking. Therefore, customers not only care about the quality of service, but also about its popularity/public acceptance. This can be modelled using the following version of : Cobb-Douglas utility function Customer’s utility function where is the expected demand and is the expected number of experts (honest miners) in the system. E[x] E[z] Note that even when there is zero demand but nonzero number of experts ( ) the customer still receives a non-zero utility from the service. Conversely, when the service is very popular and has high demand but zero expert knowledge ( ) the utility from the service is zero despite its popularity. E[x]=0; E[z]>0 E[x]>0; E[z]=0 The demand for tokens is equal to the expected total number of customers The customers are not uniform though — there are , and . They differ in reservation utility levels — each type of customer will only subscribe to the service if the utility will exceed respectively, where . early adopters late majority laggards U(x,z) Uearly, Ulate, Ulag Uearly ≤ Ulate ≤ Ulag According to , the consumer types are distributed as early adopters, late majority and laggards: product adoption life cycle 16% 68% 16% Simplified Roger’s curve You can see that I have simplified the actual Roger’s curve. If you are interested in extending the model to use actual technology adoption lifecycle, you can and I will share my ideas. contact me Investors Investors are people who buy tokens for /market value and hold them in expectation of value appreciation. We will assume that there are identical investors with equal wealth level , who are using a version of to choose between investing in stocks and our tokens. ICO N w Markowitz’s mean-variance formula Modified Markowitz’ mean-variance formula where and are coin and stock returns respectively, and and are risk-aversion coefficient and coin volatility. μc μs γ σc Note that we simplify and consider stock returns , so the model can be extended here. You can also replace Markowitz with more complex and more realistic models like , , , or or add other cryptocoins into the decision. Since our model aims to be as generic as possible, we leave out these options. If you are interested in financial economics and want to extend the model in this area, please so that we can share ideas and/or collaborate. deterministic Merton Munk Bodie Samuelson Campbell contact me We will also assume that all the investors are risk neutral. For further simplicity, we will omit from our model — the investors who will strategically predict the quorum outcome and vote in that direction to mine extra tokens. Those, who are interested in that, can enhance on this model themselves and/or for ideas and suggestions. gamblers contact me The more investors a system has, the more cash is in the system. Malicious miners Malicious miners are people who buy tokens for /market value and vote at quorum to force their version of blockchain. ICO In this model we don’t consider hypothetical ( system trolls who want to sabotage the honest blockchain just for the sake of it), so all malicious miners have a rational agenda — it’s almost always the profit maximization outside of the system. Jokers i.e. Malicious miner’s profit where the first term is the initial profit a company makes by selling their good or service at price to their initial consumers plus a share of total demand that comes from the token vote with probability of winning the vote. The second term is the which completes probability of demand from tokens to at the cost of buying tokens necessary to win the majority vote at price per token. The second term is equal to zero in case of non-interference, no malicious token activity. f α interference term 1 pc i.e. The more malicious miners a system has, the easier it is for them to influence the votes. Malicious miners participate in the token economy only if spending cash on tokens and winning the votes will increase their outside profit. Otherwise, they won’t have any incentive to meddle with the blockchain. In this model we assume that all malicious miners want to lobby the single version o the blockchain and not conflict among themselves. Moreover, malicious miners can preemptively buy tokens while they are cheap to increase profits in the future, even if this decision does not increase short-term profits. If you are interested in extending the model in this area, please feel free to to discuss this and exchange ideas. contact me In the next section we will quickly see how these agents interact at a fixed point in time. 4. System statics At any point in time, the agents solve their corresponding utility maximization problems from the previous section. The generalized static interactions can be visualized as follows: System statics in a utility token economy Note that we differentiate between and tokens, where former participate in votes and the latter do not. active passive To model the price formation of a token we use a simplified version of with money market equilibrium modelled in (Ciaian et al. (2014)). quantity theory of money “Economics of Bitcoin Price Formation” Price formula for token where is the dollar value of a token economy, is the opportunity interest rate (in our case, ), is the size of token supply and is an exogeneous stabilization parameter. PY r μs T η Note that in a proof-of-stake economy, the token supply is fixed, so the token is . T deflationary To calculate the nominal GDP ( ) we add all the cash that customers and investors inject in the token economy. PY If you are interested in more complicated pricing formulas, please read Ciaian’s paper, mentioned above, and/or to exchange ideas. contact me 4. System dynamics: simulation in R Previous two sections summarize the economy at any fixed point in time. To see how this system works over time we need to perform a dynamic analysis. The system dynamics can be modelled to follow the timing described below: The economy begins when developers create the protocol and generate all the tokens (see Section 2). , we assume that there is only one creator. Without loss of generality Creator distributes a certain share of all tokens to early experts who agree to work for tokens instead of cash and not sell them for at least a certain period of time. We make a weak assumption that all these early experts are miners. She sells other tokens in an . The decision how much to keep, how much to sell and how much to distribute involves a whole optimization process on its own, but we exogenize it for simplicity. As always, any . honest ICO extensions are welcome Consumers decide whether to subscribe to the service by buying tokens or to opt for other service providers. Investors decide whether to buy tokens or stocks instead. Malicious miners decide whether buying tokens and joining the vote will increase their short-term profit. All market participants are determined and active tokens are bidden in a vote. Majority vote wins, and losers’ stakes are — equally redistributed among winners. slashed Consumers’ and investors’ cash injections are calculated and added to price formula. Token’s price is updated. Current token rate of return is calculated as Token’s attractiveness parameter is updated. Current stock rates of returns are observed. (1 + Δpt/pt). η r Loop steps 3 to 6. R simulation We are ready to see how the economy will work in a simulation. First we will define the default parameters. Then we initialize the main results matrix: Assuming that 160 early adopters each bought 5 tokens in the first period for exchange rate and each of 200 investors also bought 10 tokens, we have the size of the token economy in the first period equal to 2800. Plugging that into price equation gives the real token value of , but we use this information in an unusual way and assign it as an appreciation in the first period: 1 token = 1 $ 0.14 What I just did, doesn’t really make any sense — I just had to arrive at a workable Markowitz solution somehow. So, if anyone wants to propose a good price appreciation mechanism, and I will correct that. please contact me And finally, we run a simulation for periods. N This basic implementation gives results, like price and demand dynamics etc. rudimentary 6. Conclusion In this article I have discussed my opinion on the current state of crypto-tokens and built a universal economic model of utility tokens. Since this model is very generic, . However, this framework and the corresponding R script are the great start to model any utility token economy. the realistic results can only be obtained by tailoring it to your specific use case Whether you are a researcher or token developer, I believe this framework will be useful to you in analyzing token economies. I welcome every discussion on model extensions and/or improvements. I strongly believe that open knowledge is the only way to the progress. You are free to use any of the components of the article, but and let other people read this article and contribute to the discussion. The publishable version of this paper is not ready yet, but I can send you a draft version . I kindly ask you to cite the source on demand

Filecoin

Utility tokens: discussion, economic model and simulation in R

The plethora of knowledge involved in Machine Learning is the most fabulous thing about the subject. The theoretical and coding balance requires a steady and disciplined approach. In this five series tutorial, we saw CNNs, where we saw various approaches to different scenarios, and then worked on word embeddings, which was our gateway to Natural Language Processing, and finally ended with Support Vector Machines(SVMs) which were as powerful as Artificial Neural Networks, during the time of their inception. Is this all there is to Machine Learning? If we have this questions running, then we are absolutely right to think the same! The above mentioned topics are just a glimpse into the supervised learning models, that are available for various domains. CNNs capture deep learning, embeddings capture NLP & SVMs capture supervised ML algorithm, but broadly, all these fit the supervised learning approach. Thus, we should also realise that initially, when we talk about Machine Learning, we are told that there are three classes of ML. Supervised LearningUnsupervised LearningReinforcement Learning Unsupervised Learning: It’s supervised learning revisited, but this time with no ground truth data. The dataset has no information about the correctness of data, and thus, the algorithms need to find patterns in the data. There are various types of unsupervised learning algorithms, ranging from clustering to auto-encoders in deep learning. k-means clustering The goal of clustering is to create groups of data points such that points in different clusters are dissimilar while points within a cluster are similar, k being the number of classes. In k-means the way these groups are defined is by creating a centroid for each group. Hierarchical Clustering: The challenge is here is unique as we intend to build a hierarchy of clusters, that is, get in a sense of granularity in various classes. The algorithm goes like this… Start with the initial k clustersMerge the two cluster that are closest to each other to obtain k-1 clustersRecompute the distances between the clustersrepeat till a single cluster of k data points is obtained. Principal component analysis (PCA) Dimensionality Reduction: First, a little linear algebra refresher — let’s talk about and . spaces bases You’re familiar with the coordinate plane with origin O(0,0) and would be i(1,0) and j(0,1). The basis vectors tell us the span of the coordinate plane it covers. Hence we can have multiple vectors covering the same space. basis vectors This means we can change the basis of a space. Now imagine much higher-dimensional space. Like, 50K dimensions. You can select a basis for that space, and then select only the 200 most significant vectors of that basis. These basis vectors are called , and the subset you select constitute a new space that is smaller in dimensionality than the original space but maintains as much of the complexity of the data as possible. principal components Another way of thinking about this is that PCA remaps the space in which our data exists to make it more compressible. The transformed dimension is smaller than the original dimension. Singular Vector Decomposition: SVD is a matrix decomposition technique which is robust and efficient. It works on rectangular matrices as well, unlike the other techniques like eigenvalue decomposition techniques. The pseudoinverse is the generalization of the matrix inverse for square matrices to rectangular matrices where the number of rows and columns are not equal, aka, Moore-Penrose Inverse. numpy scipy.linalg svd # define a matrix A = ([[ , ], [ , ], [ , ]]) U, s, VT = svd(A) from import array from import array 1 2 3 4 5 6 NumPy provides the function pinv() for calculating the pseudoinverse of a rectangular matrix. numpy numpy.linalg pinv A = ([[ , ],[ , ],[ , ],[ , ]]) # calculate pseudoinverse B = pinv(A) from import array from import array 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Application in Dimensionality Reduction Data with a large number of features, such as more features than observations (rows) may be reduced to a smaller subset of features that are most relevant to the prediction problem. The result is a matrix with a lower rank that is said to approximate the original matrix. To do this we can perform an SVD operation on the original data and select the top k largest singular values in Sigma. These columns can be selected from Sigma and the rows selected from V^T. What next? There are energy based models under unsupervised learning methods, which are fascinating, for the reason that, these probabilistic models are being paralleled to energy distributions of particles in physics… Well certainly do explore these areas! Restricted Boltzmann Machines, Deep Belief Nets and Auto-Encoders. Reinforcement Learning: This type of ML is one of the most interesting ones, for this represents the way we take actions. Dopamine is the reward that we get for doing an action successfully. In technical terminologies, the reinforcement learning agent decides how to perform a given task, be it playing games, or finding optimal path between two points, or learning how to fly a drone and do backflips. Trial and error is the best way to put it. For readers interested in going deep into RL and the math behind it, I suggest you to go through David Silver’s Lectures and Richard Sutton’s book on RL. Some amazing math backs this beautiful subdomain of Machine Learning. Terminologies: Inspired from finite state machines, these are models with finite states. The process is modelled probabilistically such that at each state, the future state is predicted based on the probabilities. Makes sense right? During the training of the agent, the rewards are associated with each transition that takes place. The definition if MDP is that the current state is independent of the previous states. Hence, it is memoryless. Markov Decision Process (MDPs): We have a function Q that takes a state and action as input and return the reward of that action at the current state. Before we explore the environment, Q gives the same (arbitrary) fixed value. But then, as we explore the environment more, Q gives us a better and better approximation of the value of an action at a state . We update our function Q as we go. Q-Learning: a s Discount factor helps in deciding important states and rewards, also an hyperparameter. Coding Time: Let’s have a look at Andrej Karpathy’s code of 130 lines, that implements Reinforcement Learning from scratch using numpy only, built on OpenAI gym’s ATARI 2600 Pong. I suggest you read his blog here to gain insights into RL. numpy np cPickle pickle gym

# hyperparameters
H = # number hidden layer neurons
batch_size = # every how many episodes to a param update?
learning_rate = gamma = # discount factor reward
decay_rate = # decay factor RMSProp leaky sum grad^ resume = False # resume previous checkpoint?
render = False

# model initialization
D = * # input dimensionality: x80 grid resume:
 model = pickle.load(open( , )) :
 model = {}
 model[ ] = np.random.randn(H,D) / np.sqrt(D) # initialization
 model[ ] = np.random.randn(H) / np.sqrt(H)
 
grad_buffer = { : np.zeros_like(v) k,v model.iteritems() } # update buffers that add up gradients over a batch
rmsprop_cache = { : np.zeros_like(v) k,v model.iteritems() } # rmsprop memory

def sigmoid(x): / ( + np.exp(-x)) # sigmoid { :dW1, :dW2}

env = gym.make( )
observation = env.reset()
prev_x = None # used computing the difference frame
xs,hs,dlogps,drs = [],[],[],[]
running_reward = None
reward_sum = episode_number = True: render: env.render()

 # preprocess the observation, set input to network to be difference image
 cur_x = prepro(observation)
 x = cur_x - prev_x prev_x is not None np.zeros(D)
 prev_x = cur_x

 # forward the policy network and sample an action the returned probability
 aprob, h = policy_forward(x)
 action = np.random.uniform() < aprob # roll the dice!

 # record various intermediates (needed later backprop)
 xs.append(x) # observation
 hs.append(h) # hidden state
 y = action == # a dlogps.append(y - aprob) # grad that encourages the action that was taken to be taken (see http: # step the environment and get measurements
 observation, reward, done, info = env.step(action)
 reward_sum += reward

 drs.append(reward) # record reward (has to be done after we call step() to get reward previous action) done: # an episode finished
 episode_number += # stack together all inputs, hidden states, action gradients, and rewards episode
 epx = np.vstack(xs)
 eph = np.vstack(hs)
 epdlogp = np.vstack(dlogps)
 epr = np.vstack(drs)
 xs,hs,dlogps,drs = [],[],[],[] # reset array memory

 # compute the discounted reward backwards through time
 discounted_epr = discount_rewards(epr)
 # standardize the rewards to be unit normal (helps control the gradient estimator variance)
 discounted_epr -= np.mean(discounted_epr)
 discounted_epr /= np.std(discounted_epr)

 epdlogp *= discounted_epr # modulate the gradient advantage (PG magic happens right here.)
 grad = policy_backward(eph, epdlogp) k model: grad_buffer[k] += grad[k] # accumulate grad over batch

 # perform rmsprop parameter update every batch_size episodes episode_number % batch_size == : k,v model.iteritems():
 g = grad_buffer[k] # gradient
 rmsprop_cache[k] = decay_rate * rmsprop_cache[k] + ( - decay_rate) * g** model[k] += learning_rate * g / (np.sqrt(rmsprop_cache[k]) + )
 grad_buffer[k] = np.zeros_like(v) # reset batch gradient buffer

 # boring book-keeping
 running_reward = reward_sum running_reward is None running_reward * + reward_sum * print % (reward_sum, running_reward) episode_number % == : pickle.dump(model, open( , ))
 reward_sum = observation = env.reset() # reset env
 prev_x = None reward != : # Pong has either + or reward exactly when game ends.
 print ( % (episode_number, reward)) + ( reward == ) "" " Trains an agent with (stochastic) Policy Gradients on Pong. Uses OpenAI Gym. " "" import as import as import 200 of 10 do 1e-4 0.99 for 0.99 for of 2 from 80 80 80 if 'save.p' 'rb' else 'W1' "Xavier" 'W2' k for in k for in return 1.0 1.0 "squashing" [0,1] ( ):
 """ 210 6400 ( ) 1 """ = [35:195] # = [::2,::2,0] # 2 [ == 144] = 0 # ( ) [ == 109] = 0 # ( ) [ != 0] = 1 # ( ) 1 . ( ). ( ) ( ):
 """ 1 """ = . ( ) = 0 ( )): [ ] != 0: = 0 # , ( ) = * + [ ] [ ] = ( ): = . ( ) [ <0] = 0 # = . ( ) = ( ) , # 2, ( ):
 """ . ( ) """ = . ( ). ( ) = . ( ) [ <= 0] = 0 # = . ( ) function to interval def prepro I prepro x160x3 uint8 frame into x80 80 D float vector I I crop I I downsample by factor of I I erase background background type 1 I I erase background background type 2 I I everything else paddles, ball just set to return I astype np.float ravel def discount_rewards r take D float array of rewards and compute discounted reward discounted_r np zeros_like r running_add for t in reversed xrange( , r.size 0 if r t running_add reset the sum since this was a game boundary pong specific! running_add running_add gamma r t discounted_r t running_add return discounted_r def policy_forward x h np dot model[ ], x 'W1' h h ReLU nonlinearity logp np dot model[ ], h 'W2' p sigmoid logp return p h return probability of taking action and hidden state def policy_backward eph, epdlogp backward pass eph is array of intermediate hidden states dW2 np dot eph.T, epdlogp ravel dh np outer epdlogp, model[ ] 'W2' dh eph backpro prelu dW1 np dot dh.T, epx return 'W1' 'W2' "Pong-v0" in 0 0 while if if else from 2 if else 3 for 1 if 2 else 0 "fake label" //cs231n.github.io/neural-networks-2/#losses if confused) new for if 1 for this with for in if 0 for in 1 2 1e-5 if else 0.99 0.01 'resetting env. episode reward total was %f. running mean: %f' if 100 0 'save.p' 'wb' 0 if 0 1 -1 'ep %d: game finished, reward: %f' '' if -1 else ' !!!!!!!!' (Source: http://karpathy.github.io/2016/05/31/rl/) Well, OpenAI was founded to fasten the progress of RL and make it at par with supervised learning. We shall talk about using OpenAI gyms in another blog. Their documentation is understandable easily too. I suggest you go through Andrej Karpathy’s blog, the best article to get you started with RL. This was tutorial 5 in the 5 part series. For any queries, suggestions or feedback, contact me at lalith@dataturks.com.

Ball

Where next? After SVMs, CNNs and Word Embeddings

React is one of the most popular frameworks in modern day javascript. Every week there’s new updates, libraries and tools released to improve your workflow and your skills. These 5 newsletters are a great way to keep up with it all. Awesome React Newsletter If you’re looking for comprehensive coverage of everything in the React world this is the newsletter for you. Each week it features the most popular React news, articles and recently released libraries. React Native Newsletter React Native lets you use React to build cross-platform mobile apps. This newsletter covers updates to React native itself as well as popular libraries and apps for it. React Digest React Digest focuses on just 5 interesting links each week. It covers everything React related including Flux, Redux, MobX, and React Native. Fullstack React Each issue of Fullstack React focuses on insightful tutorials, libraries, and React code. They cover React itself, Flux, GraphQL, Relay, and more. React Status React Status rounds up React and React Native news, tutorials, and libraries each week as well as new React jobs. Want More? You can find more great and newsletters at as well as newsletters for every other topic you might be interested. React Javascript InboxReads.co

Keep

5 Must-Read React Newsletters

HackerNoon Tech Stories Archive July 28th, 2018

Simplicity and Longevity: Life Lessons Sine Wave Taught Me

@nkgokul
Gokul N K

How I Convert Trials to Customers

@EmilBruckner
Emil Bruckner

Setting up Android like product flavors in React Native

@rohanx96
Rohan Agarwal

Babies, Supply Chains & Blockchains

@SZ
Suraz

An In-depth Review of Andrew Ng's deeplearning.ai Speciliazation

@mrdbourke
Daniel Bourke

Why do we need Proof of Work (PoW) in Blockchain?

@prashantramnyc
Prashant Ram

How many claps does the average Hacker Noon post get? 604.

@mschleiss
Martin Schleiss

Utility tokens: discussion, economic model and simulation in R

@ravshansk
Ravshan S.K.

Where next? After SVMs, CNNs and Word Embeddings

@dataturks
DataTurks: Data Annotations Made Super Easy

Where next? After SVMs, CNNs and Word Embeddings

@dataturks
DataTurks: Data Annotations Made Super Easy

5 Must-Read React Newsletters

@inboxreads
InboxReads

HackerNoon Tech Stories Archive July 28th, 2018

@nkgokulGokul N K

@EmilBrucknerEmil Bruckner

@rohanx96Rohan Agarwal

@SZSuraz

@mrdbourkeDaniel Bourke

@prashantramnycPrashant Ram

@mschleissMartin Schleiss

@ravshanskRavshan S.K.

@dataturksDataTurks: Data Annotations Made Super Easy

@dataturksDataTurks: Data Annotations Made Super Easy

@inboxreadsInboxReads

@nkgokul
Gokul N K

@EmilBruckner
Emil Bruckner

@rohanx96
Rohan Agarwal

@SZ
Suraz

@mrdbourke
Daniel Bourke

@prashantramnyc
Prashant Ram

@mschleiss
Martin Schleiss

@ravshansk
Ravshan S.K.

@dataturks
DataTurks: Data Annotations Made Super Easy

@dataturks
DataTurks: Data Annotations Made Super Easy

@inboxreads
InboxReads