Page 1
Data Science Weekly 15 In-Depth Interviews with Data Scientists Volume 2, December 2014 DataScienceWeekly.org 1
Contenu du document
Page 2
FOREWORD Given the success of our first Interview Series, we kept going! Over the past few months we have been lucky enough to conduct in- depth interviews with another 15 different Data Scientists. The 15 interviewees have varied roles and focus areas: from start-up founders to academics to those working at more established companies; working across content discovery, gaming, genomics, health and fitness, entertainment, mobile/communications and more… We wanted to make the interviews more easily accessible to the community, so have put them together in this pdf. We hope you enjoy the read and thanks again to all the interviewees! Hannah Brooks, Co-Editor DataScienceWeekly.org P.S. – If you enjoy these interviews, and want to learn more about what it takes to become a data scientist what skills do I need what type of work is currently being done in the field Check out “Data Scientists at Work” – a collection of 16 interviews with some the world's most influential and innovative data scientists, who each address all the above and more! :) [Full disclosure: author is my co-editor at DSW, Sebastian Gutierrez!] DataScienceWeekly.org 2
Page 3
CONTENTS Jarno Koponen: "Random" Predictive Content Discovery…………..Page 5 Co-Founder of Random Emmett Shear: Data Science at Twitch: CEO Perspective…………….Page 16 CEO of Twitch Dag Lohmann: Machine Learning in Catastrophe Modeling ………Page 23 Co-Founder of KatRisk Chul Lee: Data Science to improve Health & Fitness ………………….. Page 32 Director of Data Engineering & Science at MyFitnessPal Michael Watson: Applying & Teaching Data Science …………………. Page 42 Partner at Opex Analytics; Adjunct Professor at Northwestern University Chris Checco: Cloud-Based Analytics for Communications ………..Page 50 President and Chief Analytics Officer of Razorsight Reid Robison: Creating the "Dropbox of your Genome" ……………..Page 64 MD, MBA and CEO at Tute Genomics Kirk Borne: Data Mining at NASA to Teaching at GMU …............... Page 75 Data Scientist and Professor of Astrophysics & CS at GMU Dmytro Karamshuk: Optimal Retail Store Location…………………… Page 88 Researcher in Computer Science & Engineering at King's College London Nathan Kallus: Big Public Data to Predict Crowd Behavior…………Page 98 PhD Candidate at the Operations Research Center at MIT Joseph Misiti: Reducing Friendly Fire to Analyzing Social Data….Page 108 Co-founder of Math & Pencil, SocialQ and more DataScienceWeekly.org 3
Page 4
Jeroen Janssens: Data Science at the Command Line………………… Page 117 Author of Data Science at the Command Line Sudeep Das: Transforming OpenTable -> Local Dining Expert.... Page 128 Astrophysicist and Data Scientist at OpenTable Nick Elprin: Building a Data Science “Experiment Platform”…… Page 137 Founder of Domino Data Lab Szilard Pafka: Building the LA Data Science Community…………...Page 145 Chief Scientist at Epoch and Founder of Data Science LA DataScienceWeekly.org 4
Page 6
“Random” Predictive Content Discovery We recently caught up with Jarno Koponen, co-founder of Random. We were keen to learn more about his background, his perspective on predictive content discovery and what he is working on now at Random... Hi Jarno, firstly thank you for the interview. Let's start with your background and how you became interested in working with data... Q - What is your 30 second bio? A - I am the co-founder of Random, a designer and humanist passionate about augmenting our personal and collective future thinking. First step (with Random): a system that enables you to explore the unexpected and discover new interesting things… Q - How did you get interested in working with data? A - I believe we make sense of ourselves, other people and the world around us through stories that consist of bits and pieces of information... DataScienceWeekly.org 6
Page 7
that is, data. Data is everywhere. Yet data in itself is nothing. It needs to be processed and refined to become meaningful and valuable to anyone. Data needs a use case, it comes alive through a story or a product. I've always been passionate about personal stories and our digital existence. At some point I started thinking, how could personal data be used to open up an individual's future horizon. What kind of an application would help him / her to see the existence of different alternatives around themselves and to discover new interesting things. And that led me to start working around the topic of personal future simulations and predictive discovery. Q - What was the first data set you remember working with? What did you do with it? A - This is a bit more of an old-school thing, not a standard answer you'd get from a hard-core computer scientist... In a sense, we as humans are made of stories. I'm a deep-rooted humanist and I started working with history books. Going to the very source by looking at old texts and then starting to construct their meaning through related sources and relevant research literature. That to me was a unique data-set. Q - What makes you believe in the power of data? A - It's said that very little can be learned from history. This thought reflects a more fundamental belief: humanity itself doesn't change - basic things, like motivations and intentions affecting our behavior remain constant. However, I believe change can and does happen through individuals. If DataScienceWeekly.org 7
Page 8
learning and change can happen on a micro-scale, it could happen on a macro-scale too over time. If our personal data can be used to open up our personal future horizon, it might mean that such an application of data could potentially have more far-reaching consequences... even for humanity itself. In the end, everything depends on finding the use case and building products that matter. In a digital product both qualitative and quantitative data come together in a unique way. Both macro and micro events do count. Humanists, designers and data scientists are all needed when building adaptive and predictive systems - and together can unleash the power of data. Interesting to get such a different, humanist perspective - thanks for sharing! Let's talk in more detail about predictive content discovery… Q - What changes are we seeing in the world around us that are enabling Predictive Content Discovery? A - The amount of personal data has increased dramatically. Widely adopted use cases - such as Facebook or Netflix - make it possible to utilize data to understand an individual in a specific context. Every use case serves a certain purpose. Every user interface directly affects what kind of data can be meaningful and how the system can learn and adapt. Simultaneously, any specific use case does not provide a holistic understanding of an individual. New methods are needed to make it possible for an individual to benefit from his / her own data as much as DataScienceWeekly.org 8
Page 9
possible. The amount of information around us is increasing. Our current tools are inadequate for making sense of our needs and intention in relation to the information around us. New methods and tools are needed to access information that matters to us as unique individuals. We'll be moving from the age of search and social feed to a new territory, to a new paradigm in which the right information comes to us. We will go beyond linear to truly non-linear experiences, from feed scrolling or input-driven to something proactive, adaptive and personalized. Search and social will remain for the time being, yet a new user experience paradigm will be born around predictive content discovery. Q - What are the main types of problems now being addressed in the Predictive Content Discovery space? A - I don't think it's about problems, rather it's the new opportunities that are born with the evolving digital ecosystem. I think the opportunities can be put into two buckets: i) How can we get more of what we know we want (without asking for it) ii) How can we get (more) of what we don't know we want (without asking for it) For example, how to get better search results when you know what you want. Or how to get new movie recommendations when you don't quite know what you'd like to watch next. Both i) and ii) have content specific and more universal applications. It's said that there's more and more information and new content out there. However, we don't have proper tools to access them. To generalize, DataScienceWeekly.org 9
Page 10
with our current tools and services our everyday Internet has grown smaller instead of bigger. We end up in the same places far too often. The great opportunity lies in making tools and applications that widen our chances to find the things that matter to us personally. Q - Who are the big thought leaders? A - I think there are many interesting people exploring this field both in practice and theory. In this space, an interdisciplinary approach is vital. I would concentrate more on use cases and products rather than individuals. For example, Netflix is a great example of bringing together algorithms and human curation to power content discovery. Also, I've always been fascinated by Wolfram Alpha and the ideas and inspiration behind their work. I'm also following the development in the field of digital humanism with great interest. Pierre Lévy's (@plevy) work is full of inspiring insights. Alex Soojungkim Pang (@askpang) has explored how digital rhythms affect human behavior. We need a more human- centered approach when building digital products and especially systems that will augment integral parts of our existence. Q - What excites you most about bringing Machine Learning and Content Discovery together? A - To understand an individual we need to create a concrete connection between physical and digital worlds. We need to be able to bring the human mind closer to the digital realm - at some point even inside it - in a human-centered way. That's why the interaction between Machine Learning and human behavior is so exciting to me. To be more specific, to build a truly adaptive predictive system, both the DataScienceWeekly.org 10
Page 11
human and the system need to learn from each other continuously. To build such a system, both the user experience and system mechanics are equally important. Searching and discovering are different modes of thinking - we think and act differently when we explore to discover new things. And thus the system should act and learn differently too. In curiosity-driven exploration the unconscious and irrational parts of our thought-processes surface more explicitly. Using Machine Learning to capture our irrationality and associative thinking enables a loop to be created such that humans and machines are learning from each other continuously in a unique way. Q - What are the biggest areas of opportunity / questions you would like to tackle? A - It goes back to the previous themes we've discussed. How can we help an individual explore the unknown and unexpected for positive surprises? How can we better understand the needs and intentions of an individual? How can we augment our personal future thinking by creating a truly adaptive and predictive interface to the information around us? Really interesting questions - and fascinating to think about how the world of content discovery could evolve going forward. On that note, let's talk more about what you are building at Random... Q - Firstly, how did you come to found Futureful/Random? A - In a nutshell: a great bunch of talented people with an aligned vision came together and Futureful/Random was born. We used to be Futureful and now we're Random. The name change is related to the evolution of DataScienceWeekly.org 11
Page 12
our system and application as well as the world around us - the view of the future contained in Futureful was actualizing in the present so we updated our name. Q - What specific problem does Random solve? How would you describe it to someone not familiar with it? A - Random helps you to explore the unexpected and discover new interesting things. The app lets you go beyond what you know and find positive surprises in the Internet. Random is built to feed your curiosity: you can start with the topic "design" and end up learning new things about "algae". If it matches your personal interests that is. All this happens in a seamless personalized flow. Random never gives you the same content twice. Q - Why is it interesting to you? What is the most surprising insight you have found? A - Making an adaptive and predictive system that finds the right balance of serendipity and relevance is a great puzzle. And to do that one needs to understand both rational and irrational sides of a human being. Q - How do users find relevant content? How is this different from / better than traditional web-browsing? A - As I just mentioned, Random balances relevance and serendipity. It starts by suggesting topics that might be of interest to you. You then choose and get new content in front of you - be it a a blog post, video, photo or a news article. And then you choose again to find new things, at your own pace. You do not need to type in anything. You do not need to follow anyone. You do not need to sign in. The system learns from you DataScienceWeekly.org 12
Page 13
and also brings up stuff from the periphery of your interests. Everyone has a unique journey when exploring new things. Q - So, you don't have a sign-in process or link to social feeds to suggest content - could you tell us a more about the technology behind it, and how Machine Learning is helping...? A - From the very start, Random learns from you and lets you define what might be interesting for you. It learns from every interaction or non- action and starts to map the way you see the world. Everyone's reality consists of unique connections between things. To someone "art" is more strongly connected to "abstract art" than "photorealistic art". For someone else "art" refers most strongly to "oil paintings". Our system tries to capture and understand what kind of unique connections an individual has. Random tries to understand what "art" means to you personally and what kind of related things might be interesting to you. The app also allows you to connect things freely thus letting you express both your rational and irrational self. There're no universal categories or connections between different things - rather it's about an individual's own "ontology" that's created through usage. The "associative ontology" evolves continuously both through the actions of the individual and other people using the system. Random is out now in the App Store. Feel free to give it a spin and let me know what you think! DataScienceWeekly.org 13
Page 14
Q - That's great - look forward to trying it out :) ... Last thing on Random ... You mention on your website that you see Random becoming much bigger than an app - could you talk more about your future vision? A - Recommendation systems will play a key role in shaping our thinking / processes going forward. The technologies powering Random could be used to build an adaptive and predictive interface to all sorts of different kinds of information - hence driving much greater impact than just personal content discovery. For example, the next generation of operating systems will have predictive recommender systems built in their very core. A capability to learn, adapt and predict will not be a feature, but the core of the operating system itself. Jarno, thanks so much for all the insights and details behind Random - we wish you the best with the new product! Finally, it is time to look to the future and share some advice... Q - What does the future of Predictive Content Discovery look like? A - It will be more human(-centered). More mobile and ubiquitous. A new language - consisting of gestures, natural language and metadata - will be used to power the new interface to the information around us. Q - Any words of wisdom for Data Science / Machine Learning students or practitioners starting out? A - Follow the stories that matter to you personally. That's how it started with Random. Jarno - Thank you so much for your time! Really enjoyed learning more DataScienceWeekly.org 14
Page 15
about your background, your perspective on how predictive content discovery is evolving and what you are working on now at Random. Random can be found online at http://www.random.co/ and Jarno is on twitter @ilparone. DataScienceWeekly.org 15
Affichage des 15 premières pages. Téléchargez le document pour le voir en entier.
Commentaires (0)
Aucun commentaire pour le moment. Soyez le premier !