Invited speakers
- Thorsten Joachims (Cornell University)
Structured Prediction with Logged Bandit Feedback
Conventional supervised learning algorithms require training data that includes "optimal" labels. Unfortunately, such optimal labels may be difficult to annotate or even define for many constructive ML tasks. For example, what is the optimal layout of a personalized newspaper for a particular user on a given day? While the optimal layout may be unattainable as training data, it may be easy to infer the quality of a particular layout that was presented to the user (e.g., from behavioral signals). This means that we may easily get bandit feedback for learning, but not full-information feedback. In fact, such bandit-style log data is one of the most ubiquitous forms of data available, as it can be recorded from a variety of systems (e.g., search engines, recommender systems, ad placement) at little cost.
In this talk, I will explore approaches and methods for batch learning from logged bandit feedback (BLBF). Unlike the well-explored problem of online learning with bandit feedback, batch learning with bandit feedback does not require interactive experimental control of the underlying system, but merely exploits log data collected in the past. The talk explores how Empirical Risk Minimization can be used for BLBF, the suitability of various counterfactual risk estimators in this context, and a new learning method for structured output prediction in the BLBF setting. From this, I will draw connections to methods for causal inference in Statistics and Economics.
Thorsten Joachims is a Professor in the Department of Computer Science and the Department of Information Science at Cornell University. His research interests center on a synthesis of theory and system building in machine learning, with applications in information access, language technology, and recommendation. His past research focused on support vector machines, text classification, structured output prediction, convex optimization, learning to rank, learning with preferences, and learning from implicit feedback. In 2001, he finished his dissertation advised by Prof. Katharina Morik at the University of Dortmund. From 1994 to 1996 he was a visiting scholar with Prof. Tom Mitchell at Carnegie Mellon University. He is an ACM Fellow, AAAI Fellow, and Humboldt Fellow.
- Florian Pinel (The Weather Company - IBM)
Chef Watson: Computational Creativity Applied To Recipes
Can computers be creative? Meet Chef Watson. Aimed at adventurous cooks, Chef Watson is a cognitive computing application revolutionizing how people combine ingredients to create unique dishes with novel flavors. Compared to artifacts in expressive or performance domains, work products resulting from scientific creativity (including culinary recipes) seem much more conducive to data-driven assessment. One can apply computationally intensive techniques to generate many possible combinations and use automated assessors to evaluate each of them. Assembly work plans for the selected novel products can subsequently be inferred from existing records. Chef Watson applies this approach to the culinary world. After gathering data and creating a knowledge base of recipes, ingredients, and flavor compounds, the system generates ingredient combinations that satisfy user inputs such as the choice of a key ingredient, desired dish, and dietary constraints. Once a combination has been selected with the help of novelty and quality evaluators, the system further generates ingredient proportions and recipe steps. Using several variations of this approach, the system can generate new wildly creative recipes, or merely adapt existing recipes to personal preferences.
Florian Pinel is a Senior Technical Staff Member and Master Inventor at The Weather Company, an IBM Business. He received a M.S. in Computer Science and Engineering from Ecole Centrale de Paris in France, and a Culinary Arts diploma from the Institute of Culinary Education in New York. Before joining The Weather Company, Florian worked at the IBM Watson Research Center for fifteen years, focusing on Business Process Management, IT Services Management, and Software as a Service, then in the Watson Group. He's currently the lead engineer for IBM Chef Watson (www.ibmchefwatson.com), an application that uses machine learning and natural language processing to demonstrate computational creativity and suggest original recipe ideas. He also writes a blog about Eastern European cuisine, Food Perestroika (www.foodperestroika.com).
- Ruslan Salakhutdinov (University of Toronto)
Multiplicative and Fine-grained Gating for Reading Comprehension
In this talk, we will tackle the problem of reading comprehension. First, we will introduce the Gated-Attention (GA) Reader model, that integrates a multi-hop architecture with a novel attention mechanism, which is based on multiplicative interactions between the query embedding and the intermediate states of a recurrent neural network document reader. This enables the reader to build query-specific representations of tokens in the document for accurate answer selection. Second, we will develop a fine-grained gating mechanism to dynamically combine word-level and character-level representations based on properties of the words. We show that the proposed models improve over the current state-of-the-art on several benchmark datasets, including the CNN and Daily Mail news stories, the Who Did What and Children's Book Test datasets. (Joint work with Bhuwan Dhingra, Zhilin Yang, Ye Yuan, Junjie Hu, Hanxiao Liu, and William Cohen)
Ruslan Salakhutdinov received his PhD in computer science from the University of Toronto in 2009. After spending two post-doctoral years at the Massachusetts Institute of Technology Artificial Intelligence Lab, he joined the University of Toronto as an Assistant Professor in the Departments of Statistics and Computer Science. In 2016 he joined the Machine Learning Department at Carnegie Mellon University as an Associate Professor. Ruslan's primary interests lie in deep learning, machine learning, and large-scale optimization. He is an action editor of the Journal of Machine Learning Research and served on the senior programme committee of several learning conferences including NIPS and ICML. He is an Alfred P. Sloan Research Fellow, Microsoft Research Faculty Fellow, Canada Research Chair in Statistical Machine Learning, a recipient of the Early Researcher Award, Connaught New Researcher Award, Google Faculty Award, Nvidia's Pioneers of AI award, and is a Senior Fellow of the Canadian Institute for Advanced Research.
- Gisbert Schneider (ETH Zürich),
Artificially-intelligent drug design
Future success in pharmaceutical research will fundamentally rely on the combination of advanced synthetic and analytical technologies that are embedded in a theoretical framework that provides a rationale for the interplay between chemical structure and biological effect. A driving role in this setting falls on leading edge concepts in computer-assisted molecular design and machine learning, by providing access to a virtually infinite source of novel tool compounds and lead structures, and guiding experimental screening campaigns. We will discuss representations of molecular structure, predictive models of structure-activity relationships using constructive machine learning, automated molecular de novo design, and showcase prospective applications. Emphasis will be put on the automated construction of potent and selective new chemical entities. As we are currently witnessing strong renewed interest in bioactive natural products we will present applications of this approach to natural-product inspired molecular design.
Selected references:
- Schneider, P., Schneider, G. (2016) De novo design at the edge of chaos. J. Med. Chem. 59, 4077-4086
- Rodrigues, T., Reker, D., Schneider, P., Schneider, G. (2016) Counting on natural products for drug design. Nature Chem. 8, 531-542
- Schneider, P., Röthlisberger, M., Reker, D. and Schneider, G. (2016) Spotting and designing promiscuous ligands for drug discovery. Chem. Commun. 52, 1135-1138.
- Gawehn, E., Hiss, J. A., Schneider, G. (2016) Deep learning in drug discovery. Mol. Inf., 35, 3-14
- Reker, D., Schneider, G. (2015) Active learning strategies in computer-assisted drug discovery. Drug Discovery Today 20, 458-465.
Gisbert Schneider has been Full Professor of Computer-Assisted Drug Design at ETH Zürich since 2010. He received a doctorate in biochemistry from the Freie Universität Berlin, Germany, before working as a postdoc in various laboratories around the world. He then spent five years in industrial research at Hoffmann – La Roche Ltd. in Basel, Switzerland. In 2000, he received the venia legendi for biochemistry and bioinformatics from the University of Freiburg, Germany. From 2002 to 2009, he worked as a full professor at the Goethe University in Frankfurt, Germany (Beilstein Endowed Chair for Chem- and Bioinformatics). In 2015, he was distinguished as Fellow of The University of Tokyo, Japan. His research is centered on the development and practical application of new computational approaches for drug design, with a focus on adaptive and constructive machine learning methods. He authored more than 350 peer-reviewed papers and coined the terms "scaffold hopping" and "frequent hitter" in drug discovery.
- Simon Colton (Goldsmiths University of London),
Computational Creativity
In Computational Creativity research, we study how to engineer software which can take on creative responsibilities in arts and science projects. At the heart of most creative systems is a generative engine, and constructive machine learning has the potential to drive forward Computational Creativity research with new generative processes and the production of new cultural artifacts such as paintings and musical compositions. In an effort to help the emerging field of constructive machine learning to fast-track to having cultural (as well as scientific) impact, in the talk, I’ll describe some of the practical projects I’ve been involved with and what lessons I’ve learned about the value of creative software in society at large. I’ll describe some foundational philosophical issues that have arise in the field over recent years, and discuss how we’ve addressed these issues to make scientific progress, but also to lay the groundwork for creative software to have an important and lasting impact in certain cultural spheres.
Simon Colton holds an ERA Chair in Digital Games Technology at Falmouth University. He's also a Professor of Computational Creativity in the Department of Computing of Goldsmiths College, University of London, and he holds an EPSRC Leadership Fellowship. Previously, he was a Reader in Computational Creativity in the Department of Computing at Imperial College, London. Simon Colton is an Artificial Intelligence researcher, specialising in questions of Computational Creativity. In particular, he leads the Computational Creativity Group. They develop and investigate novel AI techniques and apply them to creative tasks in domains such as pure mathematics, graphic design, video game design, creative language and the visual arts. By taking an overview of creativity in such domains, they also add to the philosophical discussion of creativity, by addressing issues raised by the idea of autonomously creative software. This has enabled them to drive forward various formalism projects aimed at bringing more rigour to the assessment of creativity in software.
- Douglas Eck (Google Brain)
Magenta
I'll give an update on Magenta (magenta.tensorflow.org), a Google Brain project with the goal to generate media with deep learning and reinforcement learning using TensorFlow. I'll describe recent work on using RL to fine-tune the output of gradient descent-trained sequence generators. In this framework, it's possible to provide tutorial guidance to a sequence generator via reward functions that embed desired music theory characteristics. I'll relate this to curriculum learning. I'll also show how to work with Magenta and TensorFlow in real-time using common music creation software like Ableton and Logic. Finally I'll describe our progress on building an open-source platform for training and evaluating sequence generators using real-world feedback from musicians and listeners.
Douglas Eck is a Research Scientist at Google working in the areas of music and machine learning. Currently he is leading the Magenta Project, a Google Brain effort to generate music, video, images and text using deep learning and reinforcement learning. One of the primary goals of Magenta is to better understand how machine learning algorithms can learn to produce more compelling media based on feedback from artists, musicians and consumers. Doug led the Search, Recommendations and Discovery team for Play Music from the product's inception as Music Beta by Google through its launch as a subscription service. Before joining Google in 2010, Doug was an Associate Professor in Computer Science at University of Montreal (MILA lab) where he worked on rhythm and meter perception, machine learning models of music performance, and automatic annotation of large audio data sets.
- Ross Goodwin (NYU ITP)
Narrated Reality
Can machine intelligence enable new forms and interfaces for written language, or does it merely reveal an "uncanny valley" of text? Join Ross Goodwin as he discusses his work with neural networks for creative applications, including expressive image captioning, narration devices for your home and car, and a film (Sunspring) created from a computer generated screenplay.
Ross Goodwin is a creative technologist, artist, hacker, data scientist, and former White House ghostwriter. He employs machine learning, natural language processing, and other computational tools to realize new forms and interfaces for written language. His work has been discussed in the New York Times, the Chicago Tribune, CBS News, the Financial Times, the Guardian, the Globe and Mail, Ars Technica, VICE Motherboard, Gizmoto, Engadget, TechCrunch, CNET, Forbes, Slate, Fast Company, the Huffington Post, Mashable, Fusion, Quartz, PetaPixel, and other publications. He has exhibitied or spoken at the International Documentary Film Festival (IDFA) DocLab in Amsterdam, the TriBeCa Film Festival Interactive Showcase in New York, the International Center of Photography (ICP) in New York, the Phi Center in Montreal, Gray Area in San Francisco, the MIT Media Lab, Maker Faire, GitHub Universe, Molasses Books in Bushwick, and other venues. Ross earned his undergraduate degree in Economics from MIT in 2009, and his graduate degree from NYU ITP in May 2016.