<?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 Roger Chen on Medium]]></title>
        <description><![CDATA[Stories by Roger Chen on Medium]]></description>
        <link>https://medium.com/@rgrchen?source=rss-7cd948fcbf2b------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*EDsBuZSAuvTjJJspYaneYg.png</url>
            <title>Stories by Roger Chen on Medium</title>
            <link>https://medium.com/@rgrchen?source=rss-7cd948fcbf2b------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 27 May 2018 06:02:45 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/@rgrchen" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com" rel="hub"/>
        <item>
            <title><![CDATA[You mean you’d also build something?! Now I’m double interested. Apply away :)]]></title>
            <link>https://medium.com/@rgrchen/you-mean-youd-also-build-something-now-i-m-double-interested-apply-away-9285aae858c3?source=rss-7cd948fcbf2b------2</link>
            <guid isPermaLink="false">https://medium.com/p/9285aae858c3</guid>
            <dc:creator><![CDATA[Roger Chen]]></dc:creator>
            <pubDate>Fri, 25 May 2018 18:10:46 GMT</pubDate>
            <atom:updated>2018-05-25T18:10:46.986Z</atom:updated>
            <content:encoded><![CDATA[<p>You mean you’d also build something?! Now I’m double interested. Apply away :)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9285aae858c3" width="1" height="1">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Open Call: Fellowships @ Computable]]></title>
            <link>https://medium.com/computable-blog/open-call-fellowships-computable-21bdb5448f2c?source=rss-7cd948fcbf2b------2</link>
            <guid isPermaLink="false">https://medium.com/p/21bdb5448f2c</guid>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[distributed-systems]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[cryptography]]></category>
            <dc:creator><![CDATA[Roger Chen]]></dc:creator>
            <pubDate>Thu, 24 May 2018 19:28:02 GMT</pubDate>
            <atom:updated>2018-05-26T00:26:13.121Z</atom:updated>
            <content:encoded><![CDATA[<p>At Computable, our day job is trying to define and develop the next canonical open source web technologies. How cool is that? However, it involves a lot of hard work, and it takes a community. Actually, it takes <em>many</em> communities. Since designing and building blockchain protocols is multidisciplinary by nature, it’s important to integrate expertise across several fields. This is certainly true for our work at <a href="https://medium.com/computable-blog/introducing-computable-labs-48a69eb3aa8b">Computable — an intersection of cryptography, distributed systems, artificial intelligence</a> and more.</p><p>We thought a lot about how we can help bring together a diversity of talent and expertise. To that end, we’re launching the Fellowships @ Computable program to financially support short-term projects that we think will help not only us, but also the broader blockchain technology community. The fellowship will feature two tracks: Research Fellows and Data Fellows.</p><p><strong>Research Fellows</strong> will receiving funding to work on a well-defined technology problem that they feel is integral to the blockchain universe. This could include ideas on tackling scalability, privacy, and more.</p><p><strong>Data Fellows</strong> will build <a href="https://medium.com/computable-blog/introducing-computable-labs-48a69eb3aa8b">open data markets</a> on top of our protocol in their areas of interest and domain expertise. Perhaps these are data networks for genomics, weather, or financial data. Perhaps something else we haven’t thought about. Data Fellows are also entrepreneurs. They will have receive large ownership in the data markets they build, including the tokens associated with those blockchain data markets.</p><p>During their tenure, Fellows become full members of our team. They will work closely with Computable’s core developers who will provide advice, mentorship, and technical support. We’re even giving Fellows a home. Winners are invited (and encouraged) to work out of our office in San Francisco if possible. Details about the program and application process are below — we’re looking forward to seeing what sorts of fun and impactful proposals we get.</p><p><strong>Program specifics</strong></p><ul><li>Applications on a rolling basis</li><li>Minimum 3-month commitment</li><li>Full-time or part-time (preference for full-time)</li><li>Full-time fellows are funded up to $25k over the first 3 months with the possibility of extension</li><li>Remote is OK — check out our team <a href="https://www.computable.io/values">values</a> (“We are global.”) — but slight preference for Fellows to work out of our office (San Francisco).</li></ul><p><strong>Application process</strong><br>Send applications as well as questions and comments to <a href="http://fellowship@computable.io">fellowship@computable.io</a>. Make sure to address the questions below as part of your submission. We will accept applications on a rolling basis, but we will only support a handful of fellowships at a time, so get yours in soon. Top applicants will be selected for interviews before final funding decisions are made. We will select winners primarily based on the excellence of their proposals, independent of how they compare to other submissions. However, the bar is high, and only absolute standouts will receive funding.</p><p>In your application, directly answer these questions below, keeping your submission to 2 pages maximum:</p><ul><li>What fellowship are you applying for (Research Fellow or Data Fellow)?</li><li>Full-time or part-time? Remote or local (SF Bay Area)?</li><li>What do you want to work on?</li><li>Why are you well suited to work on this idea? Tell us about yourself.</li><li>Why is this important?</li><li>What can you accomplish in 3 months? (read: deliverables)</li><li>What deliverables and timelines do you envision over a full year?</li><li>Bonus: How did you hear about Fellowships @ Computable?</li></ul><p><strong>Suggested topics</strong><br>We are completely open to new and creative ideas, but here are some subject matters just in case it helps guide your proposal.</p><ul><li>Consensus protocols</li><li>Blockchain systems scalability</li><li>Distributed systems for machine learning and artificial intelligence</li><li>Cryptoeconomic design</li><li>Cryptoeconomic security</li><li>Privacy-preserving computations and transactions</li><li>Secure multiparty computing</li><li>Homomorphic encryption</li><li>Differential privacy</li><li>Federated learning</li></ul><p><strong>Good luck!</strong><br>Apply: <a href="http://fellowship@computable.io">fellowship@computable.io</a></p><p><em>Originally published at </em><a href="https://www.computable.io/blog/2018/5/24/open-call-fellowships-computable"><em>www.computable.io</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=21bdb5448f2c" width="1" height="1"><hr><p><a href="https://medium.com/computable-blog/open-call-fellowships-computable-21bdb5448f2c">Open Call: Fellowships @ Computable</a> was originally published in <a href="https://medium.com/computable-blog">Computable Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Introducing Computable Labs]]></title>
            <link>https://medium.com/computable-blog/introducing-computable-labs-48a69eb3aa8b?source=rss-7cd948fcbf2b------2</link>
            <guid isPermaLink="false">https://medium.com/p/48a69eb3aa8b</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[distributed-systems]]></category>
            <dc:creator><![CDATA[Roger Chen]]></dc:creator>
            <pubDate>Fri, 27 Apr 2018 18:21:30 GMT</pubDate>
            <atom:updated>2018-04-27T18:21:30.184Z</atom:updated>
            <content:encoded><![CDATA[<h4>Internet infrastructure and privacy technologies for decentralized data markets</h4><p><em>Resources:<br></em><a href="https://github.com/computablelabs"><em>GitHub</em></a><em>, </em><a href="https://github.com/computablelabs/wiki/wiki"><em>Wiki</em></a><em>, </em><a href="https://gitter.im/computablelabs/Lobby"><em>Gitter</em></a><em>, </em><a href="https://twitter.com/ComputableLabs"><em>Twitter</em></a><em>, <br></em>— —</p><p>“Who controls the past controls the future. Who controls the present controls the past.”<br><em>-Orwell sometimes by way of RATM</em></p><p>Who controls the data controls the present. Consider the era we live in. Information zips around us at unprecedented scale and velocity because of the Internet. Artificial intelligence has uncapped our ability to programmatically interpret and even generate information feeds using software. Given that nation-states, corporations, and we ourselves are information machines, it’s no wonder then that data has become the most potent resource of our time, from unlocking scientific discovery to toppling political regimes.</p><p>At Computable Labs, we’re working hard to make data a shared resource that is as openly accessible as possible. We believe it’s the biggest lever we have for creating social equity while delivering business value in the modern world. To that end, we’re building open source technology to make it not just possible, but also profitable to trade and exchange data. With the right trustless Internet infrastructure in place, we believe an entire industry will be created around an open global marketplace of data and algorithms.</p><p>Decentralized networks will be the heart (and perhaps also the soul) of this brave new world. Blockchain technologies offer new Internet primitives for economic ownership and secure transactions of digital goods online. As a result, concepts like fractional ownership of crowdsourced online datasets and machine learning models become realizable, and data markets can be created around them. Still, there is much work to be done to extend existing blockchain technology to fully implement these ideas, so that’s what we’ve set out to do.</p><p>In this post, we want to share a sneak peek at our development plans. We put a lot of consideration into establishing the right order of operations, resulting in roughly three phases of development. Our guiding light in building this roadmap ended up being pretty simple. Each phase outputs the technological building blocks we need to successfully complete the next one.</p><p><strong>Phase 1: Data Market Protocol</strong><br><em>We will develop a permissionless protocol that lets anyone easily construct online data markets in any domain.</em></p><p>The first phase of our work will involve the development of an open protocol for creating and operating data markets and exchanges. We have completed a preliminary design and will release the first of several formal technical publications soon. Anyone will be able to use this protocol to tokenize data and invite collaborators to build a crowdsourced data market together. We expect some projects will focus on verticals, e.g. genomics, while others will focus on functions, e.g. object detection, perhaps for self-driving cars.</p><p>While we will experiment by initializing some data markets ourselves, we intend to fully support other projects and companies that want to use our protocol for their own data market applications. Our primary goal in this first phase is simply to drive adoption amongst developers, whether as individuals or as part of organizations. We want to help pollinate the emerging ecosystem of data markets. We’re also looking for as much market feedback as possible, and there’s no better way than to have the market tell us what we can do to help them.</p><p>This first protocol is only the tip of the iceberg for what we ultimately want to accomplish. However, it delivers a couple technical primitives that our future work will rely on. Namely, we will be able to turn data into an asset well-suited for marketplace transactions and exchange, and we will formalize an incentive structure for crowdsourced construction of data markets.</p><p>Creating data markets for machine learning computations requires not only supplying data, but also curating and structuring it. These latter two functions are vital and often overlooked in many data market proposals. Curation ensures data is relevant and consistent. Structuring ensures it is actually machine learnable. Just as important is providing the necessary incentives for people and organizations to come together to collaboratively create these projects. Like any marketplace, data markets will face cold start challenges (perhaps more familiarly framed as the chicken-and-egg problem). Why would any data market contributor elect to contribute first? Historically, all sorts of interesting things happen when contractual ownership rights are offered on Day 0 in exchange for services. Banks loan money to homebuyers, wildcatters drill for oil, and employees work for startups. In the decentralized world, ownership is reflected by tokens, and (smart) contracts are enforced by software. During Phase 1, we will extend these concepts to work for ownership and transactions of data and algorithms, the latter of which can be simply thought of as data (model weights) as well.</p><p>Starting a data market with our protocol generates a custom non-fungible token associated with that market and lets its creators parameterize token issuance. We’ll save the juicy details for some rapidly upcoming posts and publications, but suffice it to say that we have architected the protocol to maximize for design flexibility. Understanding of token economics is rapidly evolving, so we believe it’s important to give data market creators leeway in how they want to incentivize the participants of their projects.</p><p>By the end of Phase I, it will be possible to effectively create crowdsourced data APIs owned on the permissionless Internet by their contributors. The economics of these data resources will be equitable. Those who contribute more to a data market receive more ownership in that market and therefore more financial payout as downstream users buy access into the data market.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RzESpB-IdLzcuGGEb6ZYLg.png" /></figure><p>One practical challenge with data markets is data replication. The ability to copy and paste data limits its value as a tradable asset. For some enterprises, it also holds them back from embracing more commercial data trade activities. While many businesses see the appeal in monetizing or exchanging data, they often remain leery of exposing their data to any potential loss of control. Our first generation data market applications enabled by Phase 1 and Phase 2, which we describe next, will only allow for one-way privacy. That is, either the buyer or seller will have to expose his or her asset (data or model) and trust the other side of a transaction. This will work just fine for certain data markets, but it will limit the overall opportunity until our privacy technologies of Phase 3 come online to enable fully trustless transactions.</p><p><strong>Phase 2: AI Smart Contracts<br></strong><em>We will make blockchain networks compatible with machine learning computations.</em></p><p>One modality of data markets involves selling computational access to data rather outright transfer of it (as is the case during Phase 1). For example, a data market could be constructed to sell training services. Buyers would submit models to a network that would use its data to train and return improved models. Alternatively, a model market could offer inference as a service. That is, buyers can rent access to state-of-the-art machine learning models without ever taking possession of those models. In effect, Phase 2 data markets keep sellers’ assets private and secure, whereas Phase 1 data markets protect buyers’ assets from exposure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jq9L8Ys0PODPK6UzoLRBzA.png" /></figure><p>There’s a lot of value in enabling Phase 2 designs. We’ve seen considerable market interest, for perhaps obvious reasons, from a number of businesses we’ve talked to that want to monetize their data without compromising it. Here’s the rub. Phase 2 data markets require blockchain coordination of machine learning computations, but blockchains and machine learning are at present quite incompatible. For example, it is non-trivial for heterogeneous nodes to deterministically arrive at the same machine learning outputs. This is obviously debilitating for decentralized networks that rely on consensus protocols. There are several other limitations, which we will discuss in future posts, but this is why the crux of Phase 2 will be making machine learning and blockchains more technologically compatible.</p><p>One thing that Phase 2 does not solve is the trust problem. It simply shifts trust requirements to buyers instead, and again only one-way privacy is guaranteed. Achieving bidirectional privacy becomes the focus for Phase 3, and the technology developed in Phase 2 plays a crucial role for that.</p><p><strong>Phase 3: Private Data Market Protocol<br></strong><em>Trustless third party network for private data transactions and computations</em></p><p>The untapped opportunity of data markets fully unleashes when the right privacy technologies come to fruition. We’ve talked to a number of businesses across several industries, and the message was clear. There is a strong appetite for finding new ways to monetize data and models, but they want to do so without giving those assets away. For example, is it possible to sell data to train a customer’s machine learning model without exposing the raw data (or model)? Conversely, how can we monetize a model and sell inference capabilities without revealing model weights and parameters? For now, companies acquiesce to exposing data through APIs or high-trust private data deals since there are simply no other alternatives. In the future, data commerce truly unlocks when both buyers and sellers can feel confident about exchanging data-related services without having to expose their precious assets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QvWUTxZrcCNXwzYSOY3l-A.png" /></figure><p>For this reason, we view a trustless blockchain network with privacy capabilities as a key foundation to an emerging industry built around data marketplaces. Decentralization will likely play the lead character in this story for a couple reasons. First, a single company brokering data transactions is not trustless. Quite the contrary, now both buyers and sellers would have to trust this centralized third party to neither maliciously nor mistakenly leak data. Second, decentralized networks reflect a technology paradigm that is particularly suitable for both robust and secure transactions. They eliminate single points of failure, and their distributed nature lends well to powerful privacy technologies e.g. secure multi-party computing. Our approach to privacy-preserving computations will combine a number of different methods, balancing between security and performance.</p><p>In order for a trustless blockchain network to guarantee data privacy, both data transactions and computations must privately execute within that network. This therefore requires tokenizing data and machine learning compatibility with blockchains, which is precisely why we’re working hard on Phase I and Phase II as we rapidly approach Phase III. We already have some exciting plans for making privacy-preserving computations and transactions a reality on the permissionless web, and we’re excited talk more about that in the coming months.</p><p>In the meanwhile, join the conversation on <a href="https://twitter.com/ComputableLabs">Twitter</a>, <a href="https://gitter.im/computablelabs/Lobby">Gitter</a>, and of course <a href="https://github.com/computablelabs">GitHub</a>. At Computable, some of our favorite conversations happen through pull requests!</p><p><em>-The Computable Labs Team</em></p><p><em>Originally published at </em><a href="https://www.computable.io/blog/introducing-computable-labs"><em>www.computable.io</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=48a69eb3aa8b" width="1" height="1"><hr><p><a href="https://medium.com/computable-blog/introducing-computable-labs-48a69eb3aa8b">Introducing Computable Labs</a> was originally published in <a href="https://medium.com/computable-blog">Computable Blog</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data liquidity in the age of inference]]></title>
            <link>https://zeropointnoise.com/data-liquidity-in-the-age-of-inference-1b814b93c9d2?source=rss-7cd948fcbf2b------2</link>
            <guid isPermaLink="false">https://medium.com/p/1b814b93c9d2</guid>
            <category><![CDATA[open-data]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[big-data]]></category>
            <dc:creator><![CDATA[Roger Chen]]></dc:creator>
            <pubDate>Thu, 21 Sep 2017 13:58:32 GMT</pubDate>
            <atom:updated>2017-09-21T13:58:32.173Z</atom:updated>
            <content:encoded><![CDATA[<h3>The age of inference</h3><p>It’s a special time in the evolutionary history of computing. Oft-used terms like big data, machine learning, and artificial intelligence have become popular descriptors of a broader underlying shift in information processing. While traditional rules-based computing isn’t going anywhere, a new computing paradigm is emerging around probabilistic inference, where digital reasoning is learned from sample data rather than hardcoded with boolean logic. This shift is so significant that a new computing stack is forming around it with emphasis on data engineering, algorithm development, and even novel hardware designs optimized for parallel computing workloads, both within data centers and at endpoints.</p><p>A funny thing about probabilistic inference is that when models work well, they’re probably right most of the time, but always wrong at least some of the time. From a mathematics perspective, this is because such models take a numerical approach to problem analysis, as opposed to an analytical one. That is, they learn patterns from data (with various levels of human involvement) that have certain levels of statistical significance, but remain somewhat ignorant to any physics-level intuition related to those patterns, whether represented by math theorems, conjectures or otherwise. However, that’s also precisely why probabilistic inference is so incredibly powerful. Many real world systems are so multivariate, complex, and even stochastic that analytical math models do not exist and remain tremendously difficult to develop. In the meanwhile, their physics-ignorant, FLOPS-happy, and often brutish machine learning counterparts can develop deductive capabilities that don’t nicely follow any known rules, yet still almost always arrive at the correct answers.</p><p>This is exciting because it means even without fully understanding the underlying physics, we can replicate in software powerful functions like perception, planning, and decision making. In fact, studying the inputs and outputs of learning models may even help us develop some of the missing physical models for complex systems in biology, chemistry, environmental sciences and more. When considering these capabilities, it’s no wonder why academics and industrialists alike are flocking towards ways to apply artificial intelligence. However, the exciting promise of this impending age of inference comes with an indisputable Achille’s heel. In this computing paradigm, running software applications requires data just as much as it needs memory and microprocessors. Data becomes a computing component as core as the physical parts that comprise actual computers. Suddenly, the notion of supply and value chains apply to not only physical goods, but also digital assets like data. Unfortunately, the economic ecosystems around data remain largely underdeveloped, and rather than being packaged and shipped seamlessly, data goods far too often stay raw and stuck.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AdgIA05NEJzmc3oPlbf_yA.jpeg" /><figcaption><em>Taste is a particularly interesting inference problem. On a high level, it’s simple to code rules about what dishes a person does or does not like. But what about different recipe variations of a particular dish? And then multiplex that across all the different kinds of foods someone might enjoy. It’s possible to break down food into molecular data to start understanding what compositions are most palatable, but hard to imagine a grand theory of everything distilling taste down to purely formulae. That is where statistics, deep neural networks, and the inference paradigm can shine, provided sufficient data exists for training a particular person’s gastronomic preferences.</em></figcaption></figure><h3>Data fiefdoms</h3><p>While the public web distributes information more freely and with more scale than ever before, sharing data valuable for training and running machine learning models runs contrary to how most Internet business models have evolved. That makes sense and there is nothing wrong with that per se. If a company has invested in building valuable data sets that help differentiate its product or service, its motivation is to preserve that moat by protecting that data against competition. Data network effects are a particularly powerful business and technology strategy that emerged along these lines. When we zoom outside the lens of an individual company and its singular interests, however, we start seeing several industry level benefits that breaking data silos would create for everyone. Here are just some of them.</p><h4><em>Efficiency</em></h4><p>Many redundant data sets are unnecessarily created because prior knowledge of their existence or accessibility are missing. The cost and time savings of reducing redundancy can be enormous.</p><h4><em>Reproducibility</em></h4><p>Sometimes it’s beneficial to purposely repeat data collection because reproducibility provides quality assurance. However, even in this case, data transparency and sharing is needed to actually compare independently collected data sets.</p><h4><em>Multiplier effects</em></h4><p>For narrow applications like cat recognition, the value of data can saturate rather quickly. For massive information spaces like genomics or driving around New York City, the value of data will continue to compound with aggregation for quite some time. Given that a culture of open source algorithms has already become widespread, data federation can have multiplicative effect as open models are combined with open data.</p><h4><em>Unlocking discovery</em></h4><p>Successfully solving many computational problems requires not just data, but also a critical mass of data, which can often be reached more quickly through aggregation or crowdsourcing. Different levels of critical mass unlock different levels of discovery, and we don’t even know what those levels are yet a priori.</p><h4><em>Big innovation from small players</em></h4><p>Economies of scale help large organizations amass data more efficiently, while smaller players are often on the outside looking in and find their innovative ideas data-starved. New tools giving small companies more leverage are needed to help close the data inequality gap so their innovations have a better chance of reaching markets.</p><p>Of primary concern here is not that all data should be fully federated. Forms of proprietary data will always exist because information asymmetry provides such strong strategic advantages. In fact, monopolistic ownership of data sometimes fosters innovation because it gives the innovator enough sense of security to invest in projects that require long time horizons. On the other hand, a plethora of data exhaust exists across fiefdoms that, when pooled and made accessible, could create a data commons that accelerates progress for everyone in an information intensive industry. These data sets usually create little value for individual owners when isolated, but aggregating them unlocks far greater value and grows the pie for everyone. The kind of data residing in a data commons is often necessary, but insufficient, and businesses require much more data and work to actually make their products competitive. Thus, competing by attempting to monopolize data at the data commons level seemingly amounts to a zero sum game, and resources spent doing so would likely be better used elsewhere. Instead, as machine learning proliferates across industries, sharing certain types of data can create a strong foundation on which everyone can build and drive progress. As data becomes more important than ever in the age of inference, it becomes imperative to create data liquidity cutting across organizational boundaries defined by corporations, universities, and even nation-states.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9gK9iKekfFP0qRJi_TBvRw.png" /><figcaption><em>Competing to hoard common data can lead to wasteful investment and a shaky foundation for all industry competitors. Conversely, establishing some level of data commons can create a strong foothold for an entire industry as it launches efforts to adopt machine learning. Companies still compete by developing proprietary tools and knowledge on top of the data commons, but they do so at a much faster clip by also partaking in data coopetition.</em></figcaption></figure><h3>Models of data exchange</h3><p>Sharing data for collective benefit is far from a novel concept. Calls for increasing information transparency have long existed, for example, in open science and open government. Industry has nonetheless shown limited initiative given understandable leeriness of competitors and hidden agendas. Yet, industrial markets scale to general populations, and forward-thinking altruism does not. Since markets operate according to supply and demand, market-driven incentives will likely need to take root for industry to widely adopt data sharing as it strives to incorporate machine learning competency. Here it’s worth exploring a few data exchange models to turn the gears on ways to design an industrial ecosystem that incentivizes sharing.</p><p>One way to think of data sharing is as a network where nodes represent data sets or repositories. From this perspective, the parameters that best define a data sharing network’s robustness would be its latency and uptime, where uptime corresponds to the visibility and accessibility of data nodes. Bandwidth would of course matter if real-time data sharing is critical, which it is in continuous learning applications, but here we will only consider batch learning applications where latency of procuring training data is more important. Using this framework helps us visualize the attributes and differences of three types of data exchange operating on different principles: open data, data brokerage, and data cooperative.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9MUhqKJ-9kIJ8XElZlhQNQ.png" /><figcaption><em>Lack of market incentives make the open data model hard to scale and particularly prone to challenges with data heterogeneity. Data brokerages gain financial reward for aggregating and selling data. This leads to more scale, reflected by the larger-sized nodes, and more data liquidity, though only for customers and partners that open paywalls (blue circles) with their wallets. Data cooperatives best align economic interests and coordinate efforts around data acquisition and sharing, but require the most amount of trust and suffer from cold start challenges.</em></figcaption></figure><h4><em>Open data</em></h4><p>Open data ideals are commendable and implementing them has yielded great results such as the Allen Institute’s ongoing work with the Allen Brain Atlas. Visibility and accessibility of data are high, presuming the organization publishing the data successfully markets availability and maintains infrastructure for data hosting. Significant latency exists in usefully combining data from different organizations because of data heterogeneity. This is challenging for any data exchange model, but it’s particularly hard in open data where initiatives tend to be bottom-up and fragmented across many participants without strong top-down prescription and enforcement of how to organize data. This is solvable, for example, with government agencies that can use funding as leverage in implementing standards. However, that does not solve the even bigger challenge with open data around scalability. It takes considerable work and expense to build data sets, structure them, and then host them for free public consumption. While some admirably bear this burden out of conviction in the power of openness, they form a minority, and most would-be data contributors would not take on the extra work even if they understand and agree with the value of a data commons. Ultimately, there needs to be more incentive.</p><h4><em>Data brokerage</em></h4><p>An obvious way to create incentive is introducing financial reward. The ability to monetize data has the natural effect of encouraging business formation around collecting and selling data for profit. Since that business model aligns supremely well with data aggregation, data brokers inherently solve for scalability and amass far larger amounts of data than open data projects, and they do so on a continuous basis. Visibility of data can be high because brokers are also incentivized to market their offerings, but actual data accessibility hides behind paywalls. As a result, data can move around fast and often on this network, but only for paying customers and partners. Nonetheless, data brokerage networks have worked extremely well and effectively power modern finance and e-commerce by providing the data backbone for end applications.</p><p>The data brokerage model faces limitations though, particularly pertaining to the types of data they handle. Namely, data with high acquisition costs are less palatable to data brokers because of higher capital intensity. There is no doubt, for example, that increased sharing and exchange around healthcare data is incredibly valuable. However, the process of establishing provenance and securing patient consent appears prohibitively expensive compared to the massive amounts of consumer and financial data that can be scraped from the web for free. Thus, businesses shy away from healthcare and other industries with high data acquisition cost (DAC), leaving them with poor data sharing networks as a result. Consider for a moment that important scientific enterprises in fields like medicine, chemicals, and materials feature high DAC (because data is generated from running actual physical experiments). These fields feature exactly the kind of inferential science for which machine learning can drive tremendous progress, yet lack of data liquidity puts them at a serious disadvantage as they scramble to adopt machine learning techniques. The data brokerage model also works best in areas with heavily fragmented data ownership. It struggles to work in industries featuring data monopolies who ultimately have far too much leverage and far too little incentive to share.</p><h4><em>Data cooperative</em></h4><p>Data cooperatives reflect a sort of membership model that has powerful advantages around coordination and alignment. In the ideal scenario, a virtuous cycle emerges as members collectively benefit from coordinating efforts, which attracts new members, which drives even greater collective value for members, and the pattern repeats. Alignment on the upside of collaboration (or even coopetition) is a very powerful thing. For example, members can agree on standardization to help alleviate challenges in integrating heterogeneous data. Data liquidity can be even higher than in the brokerage model because members may more transparently coordinate data exchange than partner brokers with opaque agendas. Perhaps the most distinct advantage of the cooperative model though lies in the ability to coordinate efforts and invest collectively in data generation and acquisition. This provides a key lever for solving the challenges facing industries where DAC is high and thus data liquidity is low, as described above. The cooperative can collectively prioritize what data sets to collect and pool resources to reduce effort, cost, and redundancy in procuring that data.</p><p>Impactful industry consortia like the Semiconductor Research Corporation (SRC) offer examples of how well this dynamic can work. A supreme example of successful coopetition, the SRC is governed by partners and competitors along the value chain of the semiconductor industry. Each year, members collaboratively set direction on R&amp;D focus areas to tackle the most critical technology challenges facing the industry. SRC funds some of those efforts and fosters knowledge transfer between member companies as they make progress. When a breakthrough is made, it is shared per membership agreement, with the notion that these technologies are pre-competitive. That is, their development is a common boon to everyone in the industry, and competition should ultimately focus on product offerings and differentiation. In the software world, the Linux Foundation provides another meaningful example. In the data world, this grow-the-pie mentality would be game-changing but has thus far been lacking. The primary challenge facing modern data cooperatives will be the cold start problem. Trust has to be established before companies will meaningfully contribute valuable data, but seeing others contribute data is what actually establishes trust.</p><p>The importance of data will continue to magnify, sometimes intensely so, as companies figure out how to derive value from data while privacy and security concerns simultaneously mount. I suspect we will soon see many innovative solutions to challenges around data ownership and exchange. However, we should not presuppose these issues will resolve themselves without considerable thought and effort, and there is too much on the line for us to get this wrong. The emerging paradigm of probabilistic computation holds too much promise for human progress and discovery to be stifled by playing zero sum games with data.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1b814b93c9d2" width="1" height="1"><hr><p><a href="https://zeropointnoise.com/data-liquidity-in-the-age-of-inference-1b814b93c9d2">Data liquidity in the age of inference</a> was originally published in <a href="https://zeropointnoise.com">zero point noise</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>