<?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[DAIN Studios - Medium]]></title>
        <description><![CDATA[Data Science, Artificial Intelligence and Advanced Analytics for Business. - Medium]]></description>
        <link>https://medium.com/dain-studios?source=rss----1bbed93041bb---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>DAIN Studios - Medium</title>
            <link>https://medium.com/dain-studios?source=rss----1bbed93041bb---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 09:38:38 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/dain-studios" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Data strategists accelerate digital transformation]]></title>
            <link>https://medium.com/dain-studios/data-strategists-accelerate-digital-transformation-35a5e1a2dcbf?source=rss----1bbed93041bb---4</link>
            <guid isPermaLink="false">https://medium.com/p/35a5e1a2dcbf</guid>
            <category><![CDATA[digital-transformation]]></category>
            <category><![CDATA[data-strategy]]></category>
            <category><![CDATA[management]]></category>
            <dc:creator><![CDATA[György Paizs]]></dc:creator>
            <pubDate>Fri, 21 Oct 2022 09:32:13 GMT</pubDate>
            <atom:updated>2022-10-21T09:56:57.911Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HLMW2ZD1tY0hsauOZpRt_Q.jpeg" /></figure><ul><li><strong>Data strategists are central</strong> to unlocking untapped potential in the field of data and AI</li><li>In recent times, <strong>the role has become more specialized</strong>. Three new data strategist profiles have emerged as organizations accelerate their digital transformations.</li><li>In the coming years, <strong>even more specialization is expected</strong></li></ul><p><strong>Companies that want to become data-driven need data strategists.</strong> These new technical business experts have specialized skills that bridge the gap between strategy and data science, allowing firms to unlock the potential of data and AI.</p><p><strong>Data strategists connect business and data teams, allowing each element to focus on what they do best.</strong> When we wrote about the topic in<a href="https://dainstudios.com/insights/the-rise-of-the-data-strategist-part-i/"> <strong>2019</strong></a> and<a href="https://dainstudios.com/insights/the-rise-of-the-data-strategist-part-ii/"> <strong>2020</strong></a>, we discovered that organizations can’t just rely on the generalist talents of existing staff to fulfill all the roles of data strategists. Firms need the support of individuals with increasingly specialized skills.</p><p>In this article we will discover three new data strategist profiles, offering some key information on the characteristics of each.</p><h4>Three new types of data strategist</h4><p><strong>The increasingly complex demands</strong> asked of the strategist has meant the evolution of the role into three profiles: Transformation Leader, Operation Leader, and Product Owner.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*0FeOetfG6bKsqrBwiXc-sw.jpeg" /></figure><ul><li><strong>Transformation Leader</strong>: These people work at a large scale. As they explore a company’s overall data and AI strategy, they push for the right data culture — defining priorities for an entire organization.</li><li><strong>Operation Leader</strong>: These specialists turn thinking into practical action. Hungry for the processes and workflows that create maximum data impact — they make the opportunity for a business to chase the most efficient and robust form of management and data governance.</li><li><strong>Product Owners</strong>: These experts shake up products. Not only do they identify new, innovative features and set priorities for developers — they also help to discover the broader picture of where a company should go next with their product.</li></ul><h4>Even more specialization is around the corner</h4><p>Over the next five years, as companies confront the next stages of the digital transformation, data and AI will have to do more complex tasks, more quickly. These <strong>increasing demands will lead to specialization of the role of data strategist beyond the three profiles mentioned above.</strong> So, while there will still be a need for generalists, the emphasis will be on larger firms deploying individuals with expertise in specific functions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZmDrB9j0YymdLg5eLYo1wQ.jpeg" /></figure><h4>Build awareness about data strategist profiles</h4><p>It can also be mentioned that, as many organizations look to enhance their data and AI capability at the same time, experienced talent will become more expensive and harder to find. In this context, <strong>awareness of the potential of data strategists is crucial.</strong> The more businesses know about the role, the easier it is for them to create the teams they need fast.</p><p>Do you want to find out more about data strategists? <a href="https://dainstudios.com/insights/evolution-of-the-data-strategist/">Read</a> our article about the evolution of the role.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=35a5e1a2dcbf" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dain-studios/data-strategists-accelerate-digital-transformation-35a5e1a2dcbf">Data strategists accelerate digital transformation</a> was originally published in <a href="https://medium.com/dain-studios">DAIN Studios</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Will Artificial Intelligence augment or disrupt the role of the composer?]]></title>
            <link>https://medium.com/dain-studios/will-artificial-intelligence-augment-or-disrupt-the-role-of-the-composer-f604feeb137c?source=rss----1bbed93041bb---4</link>
            <guid isPermaLink="false">https://medium.com/p/f604feeb137c</guid>
            <category><![CDATA[bach]]></category>
            <category><![CDATA[creativity]]></category>
            <category><![CDATA[music]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[DAIN Studios]]></dc:creator>
            <pubDate>Tue, 13 Oct 2020 14:17:11 GMT</pubDate>
            <atom:updated>2020-10-13T14:17:11.515Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DsEBDMJsUoueRBrEPOaxIQ.png" /></figure><h4>Music and AI</h4><h4>PART 1: Resuscitating J.S. Bach</h4><p>The music industry gets disrupted often. Industry 3.0 was the domination of automation, computers and electronics with composers like Stockhausen and Xennakis, as well as many others, experimenting with electronic sound within the context of music. But who will shape the creation of music in Industry 4.0? Will it be the large Googles, Apples and digital platforms that provide the tools for anyone to create music using artificial intelligence by broadening and augmenting the role of the composer? Or will society reject the role of Artificial Intelligence (AI) in creating music and hold to the cultural and national place of composers that has reigned for hundreds of years ?</p><p>As a musician, trained in classical music performance, analysis and composition, I find a strong emotional and internally conflicting response to calling creators of music using artificial intelligence “composers”. Apparently, I am not alone either. David Cope, the American scientist, composer and former Professor of Music, knows all too well as the developer of Experiments in Music Intelligence (EMI) software and Emily Howell project. Cope has experienced two extremes. On one hand, there has been a lot of interest in AI generated music composition over the past two decades with praise from the scientific community. But the music industry had a different viewpoint. To quote Cope:</p><blockquote>“A number of big-name classical performers expressed interest, but their agents wouldn’t let them touch it with a ten-foot pole, citing industry controversy over the work. They thought it would blemish the name of the performer”.</blockquote><p>There are many digital tools that could potentially disrupt the role of the composer. Over the past year, I have tried different AI Music programs and have felt fairly skeptical that AI might be replacing the art of composition anytime soon. I had fun with the <a href="https://bachbot.com/#/?_k=v56xb1">Bachbot</a> — which played samples of Bach, one original, the other generated by AI — and <a href="http://www.musi-co.com/">Impro AI -Musico</a> generates endless streams of repetitive music and beats with a game-like user experience. There are many more applications and machine learning projects that explore how to create a musical experience using AI, and until recently, I felt pretty safe to say, machines would not be replacing the creativity of humans any time soon.</p><p>That was until OpenAI released its prototype MuseNet in May 2019.</p><p>In this blog, I will explore the capabilities of OpenAI’s MuseNet with a focus on the style of J.S.Bach. Using a series of music demonstration experiments, I’ll reflect on the extent that AI technology is likely to drive composers and creators of music, and to adopt AI applications in their music making.</p><p>J.S. Bach’s music is highly complex in terms of its modulation formula, the use of both tonal and real fugue formulas and complex rhythms through polyphonic voices. However, thanks to extensive <a href="https://en.wikipedia.org/wiki/MIDI">MIDI</a> transcriptions of Bach’s compositions, there is sufficient data that can be used to train machines to compose like Bach. The aim of this experiment is to test the capabilities of the musical machine and compare the extent to which human creativity plays a role in generating Bach like music. All experiments will use the advanced settings on MuseNet using the style: Bach, with no seed notes for introduction. The piano will be used as the testing instrument throughout all experiments.</p><p>Our first demonstration clip uses eight AI generated segments to construct Bach. You can hear the except below:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fw.soundcloud.com%2Fplayer%2F%3Furl%3Dhttps%253A%252F%252Fapi.soundcloud.com%252Ftracks%252F621452481%26show_artwork%3Dtrue&amp;display_name=SoundCloud&amp;url=https%3A%2F%2Fsoundcloud.com%2Fuser-16902346%2Fmusenet-and-dain-studios-improvises-bach-1&amp;image=https%3A%2F%2Fsoundcloud.com%2Fimages%2Ffb_placeholder.png&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=soundcloud" width="800" height="166" frameborder="0" scrolling="no"><a href="https://medium.com/media/bec5179a82df9ff22417aef6bdeeabc7/href">https://medium.com/media/bec5179a82df9ff22417aef6bdeeabc7/href</a></iframe><p><strong>Analysis: </strong>While there are some Bach-like moments, the machine has trouble knowing how to start and therefore it sounds like the music starts in the middle of a piece. Modulations and rhythm were somewhat erratic, and it seems that the machine has not been able to sufficiently predict patterns in Bach’s music and make a Bach-like phrase.</p><p><strong>Verdict: </strong>Limitations are evident in the pattern prediction and structure, but stylistically relevant.</p><p>Let us try another simulation, this time taking the time to select an opening that starts well. This means reseting the MuseNet API a few times until we find an opening musicial segment that sounds like an opening segment of a J.S. Bach composition. Our next except is here:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fw.soundcloud.com%2Fplayer%2F%3Furl%3Dhttps%253A%252F%252Fapi.soundcloud.com%252Ftracks%252F621451470%26show_artwork%3Dtrue&amp;display_name=SoundCloud&amp;url=https%3A%2F%2Fsoundcloud.com%2Fuser-16902346%2Fmusenet-improvises-bach&amp;image=https%3A%2F%2Fsoundcloud.com%2Fimages%2Ffb_placeholder.png&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=soundcloud" width="800" height="166" frameborder="0" scrolling="no"><a href="https://medium.com/media/ff14c399642c0161d6d673d9956f3510/href">https://medium.com/media/ff14c399642c0161d6d673d9956f3510/href</a></iframe><p><strong>Analysis:</strong> With careful choices on the excerpt options, a Bach-like composition is possible. The clip has a clear start and phrases are constructed. The clip starts so that the rhythm is stable and the modulations and harmonic structure is quite similar to Bach. As the clip develops, the pitch range is extreme and uncommon for Bach who wrote the for harpsichord, however these extreme ranges work well for the piano. By the time the fifth segment is generated, the piece starts to lose its consistent structure and the Bach feel is lost — the MuseNet algorithm is limited when it comes to a complex musical transition and/or ending. <strong>Verdict:</strong> When it comes to J.S. Bach, the ability of the MuseNet as a composition tool requires human intervention for starting and ending and creating appropriate transitions in the musical composition process. Once the start of the AI-generated musical segments is decided, complex stylistic features in harmony and modulation are possible with AI.</p><p>Each of the default clips above, and the subsequent clips generated with the MuseNet API have clear stylistic musicial features that make the music distinguishabley trained on J.S. Bach data. Providing options for human judgement and choice so that phrases can be developed is a strength of MuseNet, which also can support the creativity process.</p><p>In this case, listening and deciding on the musicial segments each time an excerpt is generated requires solid knowledge of J.S. Bach’s musical features to choose the best segment to complete the phrase. The MuseNet experience is like attempting to put a Bach jigsaw puzzle together without knowing how the picture will look or how many pieces the puzzle has.</p><p>I do think this prototype of MuseNet can be a novel way to explore Bach like music, and can empower those with basic music knowledge to develop musical creativity — what will determine signficance is what the MuseNet user chooses to do with the creation made.</p><p>You can find more information about the online lecture <em>Exploring Artificial Intelligence in Classical Music</em> on Youtube <a href="https://www.youtube.com/watch?v=0RCMjPpUMLk&amp;t=2280s">here</a>.</p><p>You can follow DAIN Studios on LinkedIn <a href="https://www.linkedin.com/company/10454699/admin/">here</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f604feeb137c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dain-studios/will-artificial-intelligence-augment-or-disrupt-the-role-of-the-composer-f604feeb137c">Will Artificial Intelligence augment or disrupt the role of the composer?</a> was originally published in <a href="https://medium.com/dain-studios">DAIN Studios</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Essentialism of your Data Strategy]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/dain-studios/the-essentialism-of-your-data-strategy-48ae7e9c5d67?source=rss----1bbed93041bb---4"><img src="https://cdn-images-1.medium.com/max/1267/1*2woCDpcgNIX3hiliZpAqnA.jpeg" width="1267"></a></p><p class="medium-feed-snippet">How a use-case-driven approach provides more value.</p><p class="medium-feed-link"><a href="https://medium.com/dain-studios/the-essentialism-of-your-data-strategy-48ae7e9c5d67?source=rss----1bbed93041bb---4">Continue reading on DAIN Studios »</a></p></div>]]></description>
            <link>https://medium.com/dain-studios/the-essentialism-of-your-data-strategy-48ae7e9c5d67?source=rss----1bbed93041bb---4</link>
            <guid isPermaLink="false">https://medium.com/p/48ae7e9c5d67</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[data-strategy]]></category>
            <category><![CDATA[data]]></category>
            <dc:creator><![CDATA[Tizian Kronsbein]]></dc:creator>
            <pubDate>Thu, 08 Oct 2020 07:48:53 GMT</pubDate>
            <atom:updated>2020-10-08T07:49:33.405Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Prototyping with data and AI in service design]]></title>
            <link>https://medium.com/dain-studios/prototyping-with-data-and-ai-in-service-design-841e1d4820b2?source=rss----1bbed93041bb---4</link>
            <guid isPermaLink="false">https://medium.com/p/841e1d4820b2</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[service-design]]></category>
            <category><![CDATA[prototyping]]></category>
            <category><![CDATA[data-strategy]]></category>
            <dc:creator><![CDATA[James Keirstead]]></dc:creator>
            <pubDate>Thu, 28 Nov 2019 12:54:16 GMT</pubDate>
            <atom:updated>2020-10-08T06:47:35.354Z</atom:updated>
            <content:encoded><![CDATA[<p>Last month, I spoke at IXDS’s Pre-Work talk in Munich on the theme “Opportunities and Pitfalls of Prototyping with Artificial Intelligence”. IXDS is a leading innovation and service design studio in Germany, dedicated to ‘prototype the future’, and their monthly Pre-Work Talks bring together a community of service designers, engineers and researchers to learn about the latest developments in the field. Together with DAIN Studios CEO Dirk Hofmann, who spoke at the parallel event in Berlin, we thought this was the perfect forum to share some of our thoughts on how data and AI innovation might affect service design.</p><p>Advances in AI technologies and platforms make it easier than ever to develop products and services that respond intelligently to users and their contexts. But with all the hype and buzz about how AI <em>could </em>be used in various markets and industries, one shouldn’t forget to ask whether it actually makes sense to use AI for a given problem.</p><p>Here are some questions that a service designer might ask themselves when assessing if AI could be useful when prototyping a new service (or product):</p><ul><li>How would a service provider perform the task today?</li><li>Which elements of the task are key to improving the overall user experience? For example, are there repetitive steps that could be automated? Or are there decisions to be made that would benefit from AI-provided insights?</li><li>What assumptions is the user making when interacting with the service? Could AI be used to make those assumptions more visible to the service provider?</li></ul><p>With this understanding of the problem, service designers and AI experts can identify the task elements where AI can make the biggest contribution and suggest suitable AI methods and tools. But to develop an actual prototype, data is required. This can be a challenge at the prototyping stage, where data specific to the use case might not be available. Fortunately there are strategies to work around this, be it transfer learning, the use of public data sets, or adapting algorithms to cope with this <a href="https://en.wikipedia.org/wiki/Cold_start_(computing)">‘cold-start’ problem</a>. Finally the expertise of an analytics strategist is invaluable to pull the pieces together, linking questions about technology and infrastructure with an understanding of the organisation’s overall goals.</p><p>Prototyping with AI therefore requires an intensive collaboration between service designers, data and AI experts and strategists — not to mention users — and these interactions should start from the very beginning of a project. As Dirk puts it, this “Data Driven Design” approach has close parallels with the iterative design processes familiar to both the data science and service design communities.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XXgWxVu7duhkYaSh.png" /><figcaption>Data Driven Design: a hybrid of the traditional iterative design processes familiar in both the data science and service design communities.</figcaption></figure><p>During and after the presentations, the IXDS audience asked insightful questions that touched on a number of other pitfalls we’ve seen in developing AI-supported prototypes. For example:</p><ol><li><strong>Belief in the Black Box</strong>: Failure to understand what is going on within your AI system can limit the potential to deliver product and service innovation in the long run. In particular, be wary of outsourcing all decision-making to AI systems and explicitly consider the role of humans in AI-augmented systems.</li><li><strong>Edge case gotchas</strong>: It’s inevitable that a prototype solution will fail in certain situations. The key is to identify the most critical edge cases, to design supplementary processes and systems that can deal with these failures gracefully, and to communicate these processes and limits with stakeholders in order to build trust and engagement.</li><li><strong>Fail to Scale</strong>: A prototype is not a production system. Project and leadership teams should consider early on how to build the organisational capabilities that support both prototyping activities and deployment to live systems. Similarly, both organisations and technology are highly dynamic so think about how your proposed processes and solutions might evolve over time.</li><li><strong>Taking shortcuts</strong>: While new technologies and tools have made it easier to develop AI systems, it is still hard work! Trying to shortcut the process of data collection, algorithm development, and testing is a sure-fire way to create problems that will ultimately undermine the success of your service or product.</li></ol><p>The enthusiasm of the IXDS crowd suggested to us that there is a demand for further elaboration on these topics, so stay tuned for upcoming posts! In the meantime, you can catch Dirk’s summary on data-driven innovation via the video <a href="https://www.linkedin.com/feed/update/urn:li:activity:6592361387597549568">here</a> and the full presentation slides <a href="https://www.slideshare.net/DAINshares/opportunities-and-pitfalls-of-prototyping-with-artificial-intelligence-berlin-15102019-dirk-hofmann">here</a>.</p><p>For more information on how DAIN Studios can help your organisation make the most of data and AI opportunities, please read more about our <a href="https://www.dainstudios.com/services/">services</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=841e1d4820b2" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dain-studios/prototyping-with-data-and-ai-in-service-design-841e1d4820b2">Prototyping with data and AI in service design</a> was originally published in <a href="https://medium.com/dain-studios">DAIN Studios</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[To what extent does Computer Vision AI need ethics?]]></title>
            <link>https://medium.com/dain-studios/to-what-extent-does-computer-vision-ai-need-ethics-84c0816f565?source=rss----1bbed93041bb---4</link>
            <guid isPermaLink="false">https://medium.com/p/84c0816f565</guid>
            <category><![CDATA[ethics]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[DAIN Studios]]></dc:creator>
            <pubDate>Tue, 19 Nov 2019 12:58:29 GMT</pubDate>
            <atom:updated>2019-11-19T12:58:29.715Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lu2FNONJWeXcP6lRn-u-5Q.jpeg" /></figure><p><em>Photo by </em><a href="https://unsplash.com/@christopher__burns?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText"><em>Christopher Burns</em></a><em> on </em><a href="https://unsplash.com/s/photos/artificial-intelligence?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText"><em>Unsplash</em></a></p><p>Computer Vision as a technology employing Artificial Intelligence (AI) has been around for decades. However, the capabilities in processing large volumes of data in a moment has increased to the point that Computer Vision technologies can, and are, being used to capture more detailed and personalized information in a specific point of time. However, concerns on the ethics of Computer Vision AI have emerged, requiring much more general public awareness and discussion as the technologies develop.</p><p>Computer Vision AI for people can have significant impacts on people’s lives, culture and society — both in positive and negative ways. Unfairness by algorithms and errors may account for the unintended consequences of AI, but when placed together with Computer Vision technology applied to humans, the negative potential consequences raise highly urgent ethical questions and issues. Potential negative impacts of Computer Vision AI include identify theft, malicious attacks against identity, discrimination based on identity, identity error and misinformation, espionage and copy right infringement.</p><p>On the other hand, Computer Vision AI provides such a compelling opportunity for positive life changing technologies that can be used for social good — such as providing helpful information to blind and vision impaired people, it could also be used to support equality and inclusion. Emotional recognition takes the technology a step further — on the positive side, emotional recognition could provide valuable feedback in education, health diagnostics and social democratization, but could quickly be used for controlling and manipulating mass-behavior, in such a case where society enters an era of automatic quantitative reality.</p><p><a href="https://www.dainstudios.com/2017/09/20/saara-hyvonen/">Saara Hyvönen</a>, co-founder of <a href="http://www.dainstudios.com/">DAIN Studios</a> and DAIN Studios’ AI ethics expert, answers some questions about using AI in Computer Vision and emotional facial recognition, and the discussions it has stimulated with DAIN’s facial emotion recognition Naama that will be at Slush.</p><p><strong>To start the discussion on the ethics of Artificial Intelligence (AI), tell us about the activities DAIN Studios has been involved in with respect to AI ethics.</strong></p><p>In May 2017, the Minister of Economic Affairs, Mika Lintilä, launched the Artificial Intelligence Programme to think about how to ensure that Finland becomes one of the frontrunners among countries that apply Artificial Intelligence. I have been a member of the ethics subgroup of this programme. The focus of this group was to create discussion around AI ethics and push for renewal of operating models to take ethical guidelines into consideration.</p><p>But already before this, compliance and ethics has been a topic of interest: When innovating around data and AI one always needs to think about the broader context: user experience, legal aspects, and, yes, ethical considerations. Data and algorithms do not live in a void.</p><p><strong>DAIN Studios will have a Facial Emotion Recognition API called Naama (Finnish word for face) using Computer Vision AI at </strong><a href="https://www.slush.org/events/helsinki/#/"><strong>Slush</strong></a><strong> (world leading start-up event)— what was your initial reaction to the suggestion of using Computer Vision AI?</strong></p><p>My reaction is twofold: this kind of technology is really interesting but of course facial recognition also has this uncomfortable “Big Brother” feeling. What we want to showcase here are the possibilities of this type of technology, but also demonstrate the explainability part: how emotions are detected, what is significant to the algorithm.</p><p><strong>From an AI ethics viewpoint, what are some of the issues the programmers and developers using similar technologies should consider?</strong></p><p>In AI development, we can roughly divide ethical questions in two parts: intention and implementation. Intention deals with what we want to do with AI; this is more of a strategic/business question. Developers more run into implementation related questions: how we develop AI. Here it is important to think about what kind of biases may be hidden in the data and in the outcome of the development. Is the algorithm fair? How is fairness defined, how do we prove it?</p><p><strong>What are the main concerns for an AI ethics expert such as yourself, in terms of Computer Vision AI use cases that use Facial Recognition?</strong></p><p>The right to privacy would be my main concern with the use of facial recognition software. We already know from the widely published case of the social credit system in China that this kind of technology can be used for mass surveillance in ways that impact all aspects of an individual’s life. This highlights the importance of thinking about what we, as a society, want and how these types of technologies should and should not be used.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=84c0816f565" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dain-studios/to-what-extent-does-computer-vision-ai-need-ethics-84c0816f565">To what extent does Computer Vision AI need ethics?</a> was originally published in <a href="https://medium.com/dain-studios">DAIN Studios</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Creating a Computer Vision API in 60 minutes]]></title>
            <link>https://medium.com/dain-studios/creating-a-computer-vision-api-in-60-minutes-658ff64ae4f7?source=rss----1bbed93041bb---4</link>
            <guid isPermaLink="false">https://medium.com/p/658ff64ae4f7</guid>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[api]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Thomas Nguyen]]></dc:creator>
            <pubDate>Wed, 04 Sep 2019 08:34:05 GMT</pubDate>
            <atom:updated>2019-10-14T12:59:09.833Z</atom:updated>
            <content:encoded><![CDATA[<p>How to create a Computer Vision API in 60 minutes using fast.ai library, Flask, and Google Colab?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Qzb7m_JTZd5bungsao7PKg.jpeg" /></figure><h3>Requirements</h3><p>The requirements are straightforward. You need a Google account in order to able to access your <a href="https://drive.google.com/">Google Drive</a> and use <a href="https://colab.research.google.com/">Google Colab</a>.</p><h3>1. Create your Image Dataset</h3><p>We are going to classify flowers but, instead of taking existing datasets available from the Internet, we are going to build our own labeled image dataset using Google Images.</p><p>First, open your new notebook Google Colab and connect it to your drive with the following code:</p><pre>from google.colab import drive<br>drive.flush_and_unmount()<br>drive.mount(&#39;/content/gdrive&#39;, force_remount=True)<br>root_dir = &quot;/content/gdrive/My Drive/&quot;</pre><p>You will be asked to click a link and copy/past the code that will be given to you in the box appearing in the output of the cell.</p><p>Now you can access the data in your Drive from your notebook.</p><p>Download the images and store them into your Drive. To do this, go on Google Image and search for the objects you want to classify. In our case, we want to classify Sunflower, yellow Daisy and yellow Tulip. Let’s start with the Sunflowers. Search Sunflowers on Google image and scroll to the bottom until you see <em>show more results</em>. Then, open the developer mode of your browser and go the Web Console.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZD3eedCTaJU1km6dfnEb0Q.png" /></figure><p>There, past the following code to download a CSV of all the image urls:</p><pre>urls = Array.from(document.querySelectorAll(&#39;.rg_di .rg_meta&#39;)).map(el=&gt;JSON.parse(el.textContent).ou);<br>window.open(&#39;data:text/csv;charset=utf-8,&#39; + escape(urls.join(&#39;\n&#39;)));</pre><p>This triggers the download of the file. Save it and upload it using the widget on the left of the Notebook: Files &gt; Upload. Do the same for the other flowers.</p><p>We are going to now use the <a href="http://fast.ai/"><em>fast.ai</em></a><em> </em>library. Create a cell and paste:</p><pre>!curl -s https://course.fast.ai/setup/colab | bash</pre><p>This will install the library and configure your Colab notebook to run smoothly with it. Then, activate the GPU going to <em>Runtime &gt; Change Runtime Type &gt; GPU.</em></p><p>Now, import the library and download the images:</p><pre>from fastai.vision import *<br>from fastai.metrics import error_rate</pre><pre>folders_files = [(&#39;sunflowers&#39;, &#39;sunflowers.csv&#39;), (&#39;yellow_daisy&#39;, &#39;yellow_daisy.csv&#39;), (&#39;yellow_tulips&#39;, &#39;yellow_tulips.csv&#39;)]<br>for (folder, file) in folders_files:<br>  path = Path(root_dir + &#39;DeepLearning/Datasets/&#39;)<br>  folder = (path/folder)<br>  folder.mkdir(parents=True, exist_ok=True)<br>  download_images(path/file, folder, max_pics=200)<br>  verify_images(folder, delete=True, max_size=500)</pre><p>Here we create a folder and download 200 images for each class and verify they are not corrupted.</p><p>Next, we create an <em>ImageDataBunch</em> from the downloaded images. This object will represent our data with their labels. To optimise the process, we also:</p><ul><li>Split the data to have 20% of validation</li><li>Resize the data in squares of 224 pics</li><li>Operate image augmentation with <em>get_transforms</em></li><li>Normalize the data</li></ul><pre>np.random.seed(42)<br>data = ImageDataBunch.from_folder(path, train=&quot;.&quot;, valid_pct=0.2,<br>        ds_tfms=get_transforms(), size=224, num_workers=4).normalize(imagenet_stats)</pre><p>Let’s visualize our data:</p><pre>data.show_batch(rows=3, figsize=(7,6))</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/893/1*v7f1GG5SzlJCPzossFGOww.png" /></figure><p>Everything seems correct. However to ensure optimal performance, you should check the images manually and remove the non-consistent images from your dataset. The Google Drive UI is quite convenient for validating and filtering images manually.</p><p>Now we are going to train our model. The <a href="http://fast.ai/">fast.ai</a> library is very concise and is set with good default features.</p><pre>learn = cnn_learner(data, models.resnet34, metrics=error_rate)<br>learn.fit_one_cycle(4)</pre><p>With these lines, we are downloading a pre-trained model ResNet34, passing our data and setting the metric as the error rate.</p><p>Then, we fit the model with the one cycle policy, as this mode of approach usually performs well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/869/1*8_pc3iU8BLIwmY8da4B9OA.png" /></figure><p>We get an error rate of 6% which is not bad, but we could do better. We were training the top layers of the pre-trained model. Let’s unfreeze all layers so their parameters are able to be modified during the training phase.</p><p><a href="http://fast.ai/">fast.ai</a> library provides <em>lr_find </em>which will launch an LR range test that will help you select a good learning rate. Plotting the curve:</p><pre>learn.unfreeze()<br>learn.lr_find()<br>learn.recorder.plot()</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/465/1*m8rIV2A5M4RNN1PH1VGOvQ.png" /></figure><p>A rule thumb is to spot the strongest downward slope. Therefore we pick the beginning of the range at 1e-5 and choose to stop at 1e-4 because we were already training at 1e-3 by default. That way, the first layers are will be trained with a learning rate of 3e-5 and the last ones at 3e-4.</p><pre>learn.fit_one_cycle(10, max_lr=slice(3e-5,3e-4))</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/460/1*Cgulh_hWXVUaH7e0nN6M7w.png" /></figure><p>3.5% of error, which is better.</p><p>Let’s analyze the classification errors:</p><pre>interp = ClassificationInterpretation.from_learner(learn)<br>interp.plot_confusion_matrix(figsize=(12,12), dpi=60)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/824/1*MolWc2k_plhK1EgZ5ss1RA.png" /></figure><p>We see most of the errors come from the confusion between yellow daisy and sunflowers which can be understandable due to their visual similarities. Plotting the errors which had the biggest top loss can help us to further understand the causes of the error.</p><pre>interp.plot_top_losses(9, figsize=(15,11))</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/575/1*6mzHbhNvGpOWBQWT_-OEMg.png" /></figure><p>Some normal images have been misclassified, but others don’t look like flowers at all. This process demonstrates that we have not cleaned the dataset to the extent that we should have. Therefore, to improve performance, we should clean the image dataset and rerun the process to have a consistent score.</p><p>After cleaning and retraining our model, it should be a shame to keep it in the notebook, let’s put it now in production!</p><p>Save your model:</p><pre>learn.export()</pre><p>Download it and save it at the base of your API local folder. We are going now to build a basic API which will allow the user to upload an image and get the prediction.</p><p>Create a virtual environment with Python 3 having Flask and <a href="http://fast.ai/">fast.ai</a> library.</p><p>Create api_endpoint.py and past the following code:</p><pre>import os<br>from flask import Flask, flash, request, redirect, url_for, send_from_directory, jsonify<br>from werkzeug.utils import secure_filename<br>from fastai.vision import *<br></pre><pre>app = Flask(__name__)</pre><pre>UPLOAD_FOLDER = os.getcwd() + &#39;/files/&#39;<br>ALLOWED_EXTENSIONS = {&#39;png&#39;, &#39;jpg&#39;, &#39;jpeg&#39;, &#39;gif&#39;}</pre><pre>app = Flask(__name__)<br>app.config[&#39;UPLOAD_FOLDER&#39;] = UPLOAD_FOLDER<br></pre><pre>def allowed_file(filename):<br>    return &#39;.&#39; in filename and \<br>           filename.rsplit(&#39;.&#39;, 1)[1].lower() in ALLOWED_EXTENSIONS</pre><pre>@app.route(&#39;/&#39;, methods=[&#39;GET&#39;, &#39;POST&#39;])<br>def upload_file():<br>    if request.method == &#39;POST&#39;:<br>        # check if the post request has the file part<br>        if &#39;file&#39; not in request.files:<br>            flash(&#39;No file part&#39;)<br>            return redirect(request.url)<br>        file = request.files[&#39;file&#39;]<br>        # if user does not select file, browser also<br>        # submit an empty part without filename<br>        if file.filename == &#39;&#39;:<br>            flash(&#39;No selected file&#39;)<br>            return redirect(request.url)<br>        if file and allowed_file(file.filename):<br>            filename = secure_filename(file.filename)<br>            img = open_image(file)<br>            pred, _, losses = learner.predict(img)<br>            print(pred)<br>            file.save(os.path.join(app.config[&#39;UPLOAD_FOLDER&#39;], filename))<br>            return jsonify(str(pred))<br>    return &#39;&#39;&#39;<br>    &lt;!doctype html&gt;<br>    &lt;title&gt;Upload new File&lt;/title&gt;<br>    &lt;h1&gt;Upload new File&lt;/h1&gt;<br>    &lt;form method=post enctype=multipart/form-data&gt;<br>      &lt;input type=file name=file&gt;<br>      &lt;input type=submit value=Upload&gt;<br>    &lt;/form&gt;<br>    &#39;&#39;&#39;</pre><pre>@app.route(&#39;/uploads/&lt;filename&gt;&#39;)<br>def uploaded_file(filename):<br>    # return send_from_directory(app.config[&#39;UPLOAD_FOLDER&#39;],<br>    #                            filename)<br>    return &#39;File updated!&#39;</pre><pre>if __name__ == &#39;__main__&#39;:<br>    defaults.device = torch.device(&#39;cpu&#39;)<br>    learner = load_learner(&#39;.&#39;)<br>    print(&#39;OK&#39;)<br>    app.run(host=&quot;0.0.0.0&quot;, port=int(&quot;80&quot;), debug=True)</pre><p>We load our model line 58.</p><p>Launch your API with python api_endpoint.py and access with <a href="http://0.0.0.0/">http://0.0.0.0/</a></p><p>There you can upload your image and get your classification. For example with this image:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/225/1*GW057v_YwY_VSyyWcUHfiw.png" /></figure><p>You get a JSON response:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/330/1*sDYtZmB7dbtyDgSZGCy8Ow.png" /></figure><p>To go deeper into this topic, I highly recommend you try the excellent courses of <a href="https://www.fast.ai/">https://www.fast.ai/</a>.</p><p><a href="https://github.com/dainstudios/api-flower-classifier">You can also find the code in the repository</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=658ff64ae4f7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dain-studios/creating-a-computer-vision-api-in-60-minutes-658ff64ae4f7">Creating a Computer Vision API in 60 minutes</a> was originally published in <a href="https://medium.com/dain-studios">DAIN Studios</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Exploring the Data Realm: Running a Data Audit]]></title>
            <link>https://medium.com/dain-studios/exploring-the-data-realm-running-a-data-audit-52fc1d4e2d6e?source=rss----1bbed93041bb---4</link>
            <guid isPermaLink="false">https://medium.com/p/52fc1d4e2d6e</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[big-data]]></category>
            <category><![CDATA[data-management]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[digital-transformation]]></category>
            <dc:creator><![CDATA[Thomas Nguyen]]></dc:creator>
            <pubDate>Tue, 13 Aug 2019 11:36:17 GMT</pubDate>
            <atom:updated>2019-08-13T11:36:17.056Z</atom:updated>
            <content:encoded><![CDATA[<p>Many companies want to embrace digital transformation and use Artificial Intelligence (AI). They have the ideas, resources, and a vision, but the foundations are missing. That is, a deep knowledge of their data, from how they use and define terminologies, to how they store, use and archive data. <strong>Most AI and data analytics use cases will require connecting multiple and different data sources and types to gain deep insights and business optimization outcomes. However, a lack of knowledge and narrow vision of the data ecosystem can be a threat to the success of these projects.</strong></p><p>This post aims to give an overview of the <strong>Principles of the Data Audit</strong> which will be the first milestone for sustainable <strong>Data Management.</strong></p><h3>Principles</h3><p>The Data Audit is composed of 3 main iterative steps: the data asset identification, the measurement/evaluation of the data quality and the enrichment of the data through better quality processes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7VjBdZUs1nqip7GjYnxFcg.png" /></figure><h3>Identify the data assets</h3><p>First things first, get familiar with the data ecosystem. There are data and it can be described in many different ways. The characteristics of the data asset need prudence in definition to be fully understood. It is important to ask questions such as:</p><p>How to identify a data asset? What is the right level of detail?</p><p>To identify a data asset in the most efficient way, it is necessary to know its life cycle.</p><p>First, let’s define a <strong>data asset</strong>.</p><blockquote><strong>Any entity that is comprised of data. A data asset may be a system or application output file, database, document, or Web page, also a service that may be provided to access data from an application.</strong></blockquote><p><em>National Institute of Standards and Technology — U.S. Department of Commerce</em></p><ol><li><strong>Planning</strong>: Identify objectives, plan information architecture, develop standards and definitions;</li><li><strong>Obtainment</strong>: Ways to acquire data. For example, by creating records, purchasing data, or loading external files;</li><li><strong>Storage and Sharing</strong>: Form of storing and sharing data. Data may be stored electronically such as in databases or files. Data can be shared through such means as networks, emails or APIs;</li><li><strong>Maintenance</strong>: Update, change, and manipulate data. Cleanse, wrangle and transform data. Match and merge records, or optimize data;</li><li><strong>Use</strong>: Retrieve data. Use information. Use can be completing a transaction, writing a report, making a management decision from a report, and/or running automated processes;</li><li><strong>Disposal</strong>: Archive information or deleting data or records;</li></ol><p>For each phase, <strong>four aspects have to be taken in account</strong>:</p><ul><li>the <strong>What: </strong>what data is concerned during these phases?.</li><li>the <strong>How</strong> — <strong>Processes: </strong>how are these phases executed from a procedural point of view?</li><li>the <strong>Who</strong>: who is involved in these phases?</li><li>the <strong>How — Technology</strong>: how are these phases executed from a technological point of view?</li></ul><p>This framework gives a holistic view of the analyzed data assets: <strong>it will be the pedestal for improving the data asset, imagining new use cases and making data governance implementable.</strong></p><h3>Measure and Evaluate Data Quality</h3><p>Now that the data asset can be identified, the next step is to determine if the data is usable? Not all data is usable or relevant to the AI and analytics use cases, and therefore it is important and necessary to assess data quality. To do this, we ask the following questions:</p><p>What makes data quality? Is it necessary to have an impeccable data asset?</p><p>Data quality relies on <strong>metrics</strong> that will be confined to <strong>business rules</strong>. These metrics will be the same for all data assets but it will not be the case for the business rules. The comparison of the two will give the <strong>validation of the data quality</strong>.</p><h3>Data Quality Metrics</h3><p>Data quality can be measured with 6 main metrics:</p><ul><li><strong>Timeliness: </strong>Describe the latency between the event and the availability of the information — e.g. the contracts data set is refreshed once per year;</li><li><strong>Completeness</strong>: Percentage of values that are considered as complete — e.g. only 80% of the timestamps are complete;</li><li><strong>Validity</strong>: Respect of a certain syntax, range or set of rules — e.g. age has to be between 18 and 110 for some cases;</li><li><strong>Uniqueness</strong>: Presence of duplicates or missing values — e.g. if a business serves 100 clients, there shouldn’t be more than 100 in the data set;</li><li><strong>Consistency</strong>: Consistency in the way to represent data — e.g. dates are always encoded in the same way;</li><li><strong>Accuracy</strong>: Exactitude of the data in comparison with the reality — e.g. the billing system should be the same, and consistent with, the financial data sets.</li></ul><p>At the exception of timeliness, all the other metrics are measured at the level of fields.</p><h3>Business Rules</h3><p>It’s good to be able to quantify the data quality but <strong>the final goal is to qualify data reliability</strong>. Indeed, <strong>not all the fields of a data set have the same importance</strong>: inaccuracy, incompleteness, and non-validity can be accepted for some of them.</p><p>To qualify this data reliability, one first step is to define business-related data quality rules. These rules are defined by the owner of the data asset and they represent the standard from which data can be qualified as reliable. Each metric we’ve seen previously should have its own business rule but it can happen in cases where there is none.</p><p>For example, in a telecom company, there is a data set containing all the mobile customer contracts. Some of the fields are <em>contract ID </em>and <em>email address</em>. Let’s elaborate the business rules for both of them.</p><p><strong>Timeliness </strong>is the same for the whole data set — it should be refreshed daily.</p><p><em>Contract ID:</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4CLXz8mZ4NlfQoS3o_jKPg.png" /></figure><p><em>Email Address:</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cj_l-vQQzqXGYUUUY0Ifbw.png" /></figure><p>The key idea here is that not all the fields have the same value. For those which are central in the core business, the business rules should be strict and carefully defined. However, for secondary fields, it may not be worth spending too much time on the determination of rules.</p><h3>Data Quality Evaluation</h3><p>The goal here is to <strong>compare the metrics measured and the defined business rules.</strong> If all the metrics respect the rules then the field is validated.</p><p>In the best case of fields being evaluated, all fields will be validated and therefore the entire data asset is validated too. In the worst but simple case, all fields are wrong and the entire data asset will need correction.</p><p>The reality usually presents a mix of both valid and invalid fields in the data quality evaluation with some fields being compliant, and others not. In that case, the mandatory fields should be defined and their compliance required for validating the data asset. This validation should, however, go with some indication about <strong>how the problematic fields could affect (or not) the current or future business use cases.</strong></p><h3>Implement quality processes and enrich data</h3><p>After the data quality evaluation, the issues are underlined and have to be resolved. The best way to solve these issues is to find the root causes.</p><p>Fortunately, the data life cycles of problematic data assets are known. Therefore, it is possible to locate the stage where the problem appears, its technical and/or business reason, and the people to talk with to fix it.</p><p>To conclude, a Data Audit is divided into three phases. The first is the identification of the data asset and the investigation on its life cycle. The second is the evaluation of the data quality. For this, it is necessary to measure the 6 main metrics and compare them to the defined business rules. The output of these two phases is a clear view of any problematic data. The third and last phase consists of treating the root causes of these issues. Using the data life cycle enables us to find why and where there is a problem and with whom to interact with for a sustainable data management solution.</p><p>Adapted from <a href="https://www.amazon.com/Executing-Data-Quality-Projects-Information/dp/0123743699">Executing Data Quality Projects: Ten Steps to Quality Data and Trusted Information,</a> Danette McGilvray</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=52fc1d4e2d6e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dain-studios/exploring-the-data-realm-running-a-data-audit-52fc1d4e2d6e">Exploring the Data Realm: Running a Data Audit</a> was originally published in <a href="https://medium.com/dain-studios">DAIN Studios</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>