Put AI to work
June 26-27, 2017: Training
June 27-29, 2017: Tutorials & Conference
New York, NY
 
Grand Ballroom West
11:05am Teaching machines to reason and comprehend Russ Salakhutdinov (Carnegie Mellon University)
4:50pm Algorithms for hire Lindsey Zuloaga (HireVue)
Gramercy East/West
11:05am Cognitive mobile healthcare for the patient and physician Michael Nova (Pathway Genomics)
2:35pm Harnessing the power of artificial intelligence to diagnose diseases Kavya Kopparapu (GirlsComputingLeague)
4:00pm How AI powers the Comcast X1 voice interface Jan Neumann (Comcast), Ferhan Ture (Comcast), Shahin Sefati (Comcast)
4:50pm AI within O'Reilly Media Paco Nathan (derwen.ai)
Beekman
1:45pm Anaerobic AI: Developing in a data-starved environment Xiaofan Xu (Intel), Cormac Brick (Intel)
2:35pm AI for manufacturing: Today and tomorrow David Rogers (Sight Machine)
4:50pm Interpretable AI: Not just for regulators Patrick Hall (bnh.ai | H2O.ai), SriSatish Ambati (H2O.ai)
Sutton South/Regent Parlor
11:05am Building training data for autonomous driving Matt Shobe (Mighty AI)
11:55am Inverse reward design Anca Dragan (UC Berkeley)
1:45pm Scaling computer vision in the cloud Reza Zadeh (Matroid | Stanford)
2:35pm Integrating deep learning libraries with Apache Spark Joseph Bradley (Databricks), Xiangrui Meng (Databricks)
4:50pm Adding meaning to natural language processing Jonathan Mugan (DeepGrammar)
Murray Hill E/W
11:05am XPRIZE Workshop: Using AI for Impact amir banifatemi (Xprize), Balazs Kegl (CNRS)
11:55am Bayesian deep learning Yarin Gal (University of Cambridge)
1:45pm Learning to recreate our visual world Jun-Yan Zhu (Berkeley AI Research Lab)
4:00pm Deep learning in the fashion industry Pau Carre (Gilt)
Sutton Center/North
1:45pm Programming your way to explainable AI Mark Hammond (Microsoft)
2:35pm The AI-powered newsroom Codruta Gamulea (Bakken & Bæck)
4:00pm AI's legal history and some notions of the future Aileen Nielsen (Skillman Consulting)
4:50pm Rethinking design tools in the age of machine learning Patrick Hebron (New York University)
Grand Ballroom
9:00am Thursday opening remarks Ben Lorica (O'Reilly), Roger Chen (Computable)
9:05am Magenta: Machine learning and creativity Doug Eck (Google Brain)
9:25am Cars that coordinate with people Anca Dragan (UC Berkeley)
10:00am Evolve AI (sponsored by Intel Nervana) Naveen Rao (Intel)
10:30am Closing remarks Ben Lorica (O'Reilly), Roger Chen (Computable)
10:35am Morning Break | Room: Sponsor Pavilion
3:15pm Afternoon Break | Room: Sponsor Pavilion
12:35pm Lunch sponsored by Intel Nervana Thursday Industry Tables at Lunch | Room: Americas Hall
8:00am Wake up coffee | Room: 3rd Floor Promenade
8:15am Speed Networking | Room: 3rd Fl Promenade
11:05am-11:45am (40m)
Teaching machines to reason and comprehend
Russ Salakhutdinov (Carnegie Mellon University)
Russ Salakhutdinov discusses some of the key challenges to making machines more intelligent, focusing on the Gated-Attention (GA) Reader model, which integrates a multihop architecture with a novel attention mechanism, along with extensions that make use of external linguistic knowledge.
11:55am-12:35pm (40m) Implementing AI Deep Learning, Media
Dynamic deep learning: A paradigm shift in AI research and tools
Soumith Chintala (Facebook)
Soumith Chintala discusses paradigm shifts in cutting-edge AI research and applications such as self-driving cars, robots, and game playing.
1:45pm-2:25pm (40m) Implementing AI Deep Learning, Vision
Customizing state-of-the-art deep learning models for new computer vision solutions
Timothy Hazen (Microsoft)
Dramatic progress has been made in computer vision: deep neural networks (DNNs) trained on tens of millions of images can now recognize thousands of different object types. These DNNs can also be easily customized to new use cases. Timothy Hazen shares simple methods and tools that enable you to adapt Microsoft's state-of-the-art DNNs for use in your own computer vision solutions.
2:35pm-3:15pm (40m) Implementing AI Cloud, Deep Learning, Machine Learning, Vision
Risks, hidden costs, and how to escape the black hole of machine learning technical debt
Matt Zeiler (Clarifai)
AI-powered machine learning technologies bring a higher and more complex level of technical debt to applications. Matt Zeiler shares best practices for companies hoping to build AI into their businesses and explores how machine learning increases technical debt, the key contributors, and how to avoid or reduce technical debt related to machine learning.
4:00pm-4:40pm (40m) Implementing AI Cloud, Deep Learning, Vision
AI Vision: Enable deep learning-based visual analysis in edge and cloud environments
Yonghua Lin (IBM Research)
Yonghua Lin leads a deep dive into AI Vision, a deep learning system from IBM for image and video analysis in both edge and cloud environments, exploring its system design, performance optimization, and large-scale capability for training and inference.
4:50pm-5:30pm (40m) Implementing AI
Algorithms for hire
Lindsey Zuloaga (HireVue)
Lindsey Zuloaga explains how machine learning from video interviews is disrupting the human resources space, bringing top candidates to the attention of recruiters and drastically reducing the time and energy companies spend finding and assessing potential employees.
11:05am-11:45am (40m) Verticals and applications Health care, Machine Learning, Natural Language
Cognitive mobile healthcare for the patient and physician
Michael Nova (Pathway Genomics)
Precision medicine is largely a big data and systems problem, especially with many different types of "siloed" healthcare information, such as lab results, genetic tests, IoT and wearables data, and insurance information. Michael Nova explains why cognitive computing and artificial intelligence that can dynamically learn using any healthcare data will dramatically impact precision healthcare.
11:55am-12:35pm (40m)
AtomNet: A deep convolutional neural network for bioactivity prediction in structure-based drug discovery
Abe Heifets (Atomwise)
Abe Heifets offers an overview of AtomNet, a structure-based deep convolutional neural network designed to predict the bioactivity of small molecules for drug discovery applications. Abe discusses training AtomNet on millions of training examples derived from ChEMBL and the PDB and explains how autonomously discovered filters can outperform previous docking approaches and existing DNN techniques.
1:45pm-2:25pm (40m) Verticals and applications Health care
AI-assisted computational chemistry: Predicting chemical properties with minimal expert knowledge
Garrett Goh (Pacific Northwest National Lab)
Garrett Goh demonstrates how to use deep learning to construct computational chemistry models that compare favorably to existing state-of-the-art models developed by expert practitioners—with virtually no expert knowledge—proving the potential of AI assistance to accelerate the scientific discovery process from a typical span of years to a matter of months.
2:35pm-3:15pm (40m) Impact of AI on business and society Health care
Harnessing the power of artificial intelligence to diagnose diseases
Kavya Kopparapu (GirlsComputingLeague)
Artificial intelligence is revolutionizing medicine through computer-aided diagnostic systems. High school student Kavya Kopparapu presents the Eyeagnosis system, which utilizes artificial intelligence techniques and a smartphone camera to automatically screen for diabetic retinopathy, the leading cause of preventable blindness worldwide.
4:00pm-4:40pm (40m) Implementing AI Deep Learning, Media, Natural Language, Speech and Voice, User interface and experience
How AI powers the Comcast X1 voice interface
Jan Neumann (Comcast), Ferhan Ture (Comcast), Shahin Sefati (Comcast)
AI plays an essential role in creating the Comcast X1 entertainment experience and is how millions of its customers access their content on the TV. Jan Neumann, Ferhan Ture, and Shahin Sefati explain how AI enables Comcast to understand what you are looking for when you talk to the X1 voice remote and how Comcast scaled the voice interface to answer millions of voice queries every single night.
4:50pm-5:30pm (40m) Verticals and applications Machine Learning, Media, Natural Language
AI within O'Reilly Media
Paco Nathan (derwen.ai)
Paco Nathan explains how O'Reilly employs AI, from the obvious (chatbots, case studies about other firms) to the less so (using AI to show the structure of content in detail, enhance search and recommendations, and guide editors for gap analysis, assessment, pathing, etc.). Approaches include vector embedding search, summarization, TDA for content gap analysis, and speech-to-text to index video.
11:05am-11:45am (40m) Implementing AI
AI for smartphones: Running neural networks locally on phones for real-time use of the camera as a sensor
Alberto Rizzoli (Aipoly)
Alberto Rizzoli explains how Aipoly began running convolutional neural networks locally on smartphones, eventually reaching a level of performance that made it a better option than cloud services, in the process unlocking new possibilities for making phones contextually aware.
11:55am-12:35pm (40m) Implementing AI Deep Learning, Hardware, IoT and its applications
Software and hardware breakthroughs for deep neural networks at the edge
Michael B. Henry (Mythic)
Breakthroughs in deep learning and new analog-domain computation methods to deploy trained neural networks will deliver exciting new capabilities. Michael B. Henry explains why the combination of human-like levels of recognition and massive computation capabilities in a tiny package will enable products with true awareness and understanding of the user and environment.
1:45pm-2:25pm (40m) Implementing AI Hardware, IoT and its applications, Machine Learning
Anaerobic AI: Developing in a data-starved environment
Xiaofan Xu (Intel), Cormac Brick (Intel)
Data is the “oxygen” of the AI revolution, but access to data on a large scale remains a luxury of an elite group of tech companies, effectively creating a “data wall” blocking smaller companies. Cormac Brick and Xiaofan Xu explore the problem of the data wall and offer a solution: synthetic datasets.
2:35pm-3:15pm (40m) Verticals and applications
AI for manufacturing: Today and tomorrow
David Rogers (Sight Machine)
Join David Rogers to learn how AI can make your operations more efficient and profitable. David explains how existing technologies like the digital twin approach, advanced decision making, and downtime cause detection have primed manufacturing for a profitable and efficient future.
4:00pm-4:40pm (40m) Implementing AI
Rules of machine learning verification: From data-driven bugs to explainable AI
Suman Roy (Betaworks)
Machine learning is empowering, but a critical drawback in the current ecosystem is the lack of tactical verification tools that can guarantee its fidelity in real-world applications. Suman Roy explores the tools and best practices during training, implementation, and postdeployment that can help explain what exactly we are teaching these machines.
4:50pm-5:30pm (40m) Impact of AI on business and society
Interpretable AI: Not just for regulators
Patrick Hall (bnh.ai | H2O.ai), SriSatish Ambati (H2O.ai)
Interpreting deep learning and machine learning models is not just another regulatory burden to be overcome. People who use these technologies have the right to trust and understand AI. Patrick Hall and Sri Satish share techniques for interpreting deep learning and machine learning models and telling stories from their results.
11:05am-11:45am (40m) Implementing AI Machine Learning, Transportation and Logistics, Vision
Building training data for autonomous driving
Matt Shobe (Mighty AI)
Autonomous vehicles must recognize objects in context, no matter the weather, time of day, or season. What does a cat in the road look like on a sunny summer day? How about on a snow-covered road at night? Matt Shobe shares lessons Mighty AI has learned while creating a training dataset for autonomous driving, including workflow tips and guidance for engineers building computer vision models.
11:55am-12:35pm (40m)
Inverse reward design
Anca Dragan (UC Berkeley)
As AI agents become more capable of optimizing their objective functions, it's becoming increasingly important to make sure that we give them the right objectives in the first place. Anca Dragan explains why agents should have uncertainty about their objectives and use human input as valuable observations to improve their estimates.
1:45pm-2:25pm (40m) Implementing AI Cloud, Deep Learning, Vision
Scaling computer vision in the cloud
Reza Zadeh (Matroid | Stanford)
Providing customized computer vision solutions to a large number of users is a challenge. Matroid allows the creation and serving of computer vision models and algorithms, model sharing between users, and serving infrastructure at scale. Reza Zadeh offers an overview of Matroid's pipeline, which uses TensorFlow, Kubernetes, and Amazon Web Services.
2:35pm-3:15pm (40m) Implementing AI Cloud, Deep Learning
Integrating deep learning libraries with Apache Spark
Joseph Bradley (Databricks), Xiangrui Meng (Databricks)
Joseph Bradley and Xiangrui Meng share best practices for integrating popular deep learning libraries with Apache Spark, covering cluster setup, data ingest, configuring clusters, and monitoring jobs. Joseph and Xiangrui then demonstrate these techniques using Google’s TensorFlow library.
4:00pm-4:40pm (40m) Implementing AI
How Amy, an artificial intelligence capable of scheduling meetings, understands human intents
Rakesh Chada (x.ai)
Rakesh Chada introduces x.ai's Amy, an AI assistant that schedules meetings via email. Rakesh discusses Amy's architecture and the various challenges the team faced during its design and shares several machine learning approaches for intent classification. Rakesh concludes by exploring a novel method for error optimization in a conversational agent that exploits customer error tolerance.
4:50pm-5:30pm (40m) Implementing AI Machine Learning, Natural Language
Adding meaning to natural language processing
Jonathan Mugan (DeepGrammar)
Jonathan Mugan surveys the field of natural language processing (NLP), both from a symbolic and a subsymbolic perspective, arguing that the current limitations of NLP stem from computers having a lack of grounded understanding of our world. Jonathan then outlines ways that computers can achieve that understanding.
11:05am-11:45am (40m) Interacting with AI
XPRIZE Workshop: Using AI for Impact
amir banifatemi (Xprize), Balazs Kegl (CNRS)
Amir Banifatemi and Balazs Kegl discuss the XPRIZE Foundation, which has launched the largest global AI competition to use AI for impact, outlining the foundation's goals, what the prize entails, and what the 146 teams from 22 countries are working on. One of the teams will share its project and explore various methods and practical ways to interact with AI.
11:55am-12:35pm (40m) Verticals and applications
Bayesian deep learning
Yarin Gal (University of Cambridge)
Yarin Gal shares a new theory linking Bayesian modeling and deep learning and demonstrates the practical impact of the framework with a range of real-world applications. Yarin also explores open problems for future research—problems that stand at the forefront of this new and exciting field.
1:45pm-2:25pm (40m)
Learning to recreate our visual world
Jun-Yan Zhu (Berkeley AI Research Lab)
Jun-Yan Zhu explains how to learn natural image statistics directly from large-scale data and explores a class of image-generation and editing operations that constrain their output to look realistic according to the learned image statistics.
2:35pm-3:15pm (40m) Impact of AI on business and society Machine Learning, Media, Retail and e-commerce
Recommending products for 1.91 billion people on Facebook
Nikita Lytkin (Facebook)
Nikita Lytkin offers an overview of personalized digital advertising and explains how Facebook uses modern supervised machine learning methods, such as factorization machines and deep neural networks, to recommend ecommerce products to nearly two billion people.
4:00pm-4:40pm (40m) Implementing AI Deep Learning, Fashion, Retail and e-commerce, Vision
Deep learning in the fashion industry
Pau Carre (Gilt)
Pau Carré explains how Gilt is reshaping the fashion industry by leveraging the power of deep learning and GPUs to automatically detect similar products and identify facets in dresses.
4:50pm-5:30pm (40m) Interacting with AI Machine Learning, Retail and e-commerce, User interface and experience
Deep shopping bots: Building machines that think and sell like humans
Rupert Steffner (WUNDER)
70% of consumers do NOT feel that online offers resonate with their personal interests and needs. Rupert Steffner explains how cognitive AI can help create deep shopping bots based on true personal relevance. This shift in the shopping paradigm is built upon deep symbolic reinforcement learning, the psychometry of shopping, a new breed of playful UI, and cognified product metadata.
11:05am-11:45am (40m) Implementing AI Machine Learning
Human-assisted AI at B12: 10 lessons in giving humans superpowers
Adam Marcus (B12)
AI has a way to go before it replaces the jobs we know today. But long before AI automates away jobs, it will elevate expertise. B12 is building infrastructure that celebrates humans where they’re best while bringing machines in for the rest. Adam Marcus offers an overview of human-assisted AI and demonstrates how it is already changing creative (and fundamentally human) fields like design.
11:55am-12:35pm (40m) Implementing AI Machine Learning, Natural Language
Strategies for integrating people and machine learning in online systems
Jason Laska (Clara Labs)
Clara Labs is fusing machine learning (ML) with distributed human labor for natural language tasks. The result is a virtuous cycle: ML predictions improve workers’ efficiency, and workers help improve prediction models. Jason Laska explores the challenges of building a real-time(ish) knowledge workforce, how to integrate automation, and key strategies Clara Labs learned that enable scale.
1:45pm-2:25pm (40m) Implementing AI IoT and its applications, Machine Learning
Programming your way to explainable AI
Mark Hammond (Microsoft)
As interactive and autonomous systems make their way into nearly every aspect of our lives, it is crucial to gain more trust in intelligent systems. Mark Hammond explores the latest techniques and research in building explainable AI systems. Join in to learn approaches for building explainability into control and optimization tasks, including robotics, manufacturing, and logistics.
2:35pm-3:15pm (40m) Impact of AI on business and society Media, Natural Language
The AI-powered newsroom
Codruta Gamulea (Bakken & Bæck)
The promise of AI in the newsroom is contradictory: NLG revolutionizes news writing, but robot journalists threaten jobs; NLP improves fact-checking but requires investments that slimmed-down newsrooms cannot afford. Drawing on Norwegian AI startup Orbit’s experience, Codruta Gamulea explains how AI can help solve the industry resource constraints and improve the quality of journalism.
4:00pm-4:40pm (40m) Impact of AI on business and society Ethics, Governance, and Privacy
AI's legal history and some notions of the future
Aileen Nielsen (Skillman Consulting)
While the commercial use of AI in everything from hiring to medical diagnosis to work scheduling is exploding, legislation and case law alike have yet to make major statements about how AI will be treated by the American legal system. Aileen Nielsen offers a historical overview of how the law has dealt with decision-making technologies in the past and what this suggests about AI's legal future.
4:50pm-5:30pm (40m)
Rethinking design tools in the age of machine learning
Patrick Hebron (New York University)
Is it possible to simplify design tools without limiting their expressivity? Patrick Hebron investigates how recent advances in machine learning and artificial intelligence will enable a new generation of tools that help novice and expert designers alike develop deeply nuanced and original ideas without committing to a steep learning curve or ceding creative control to the machine.
9:00am-9:05am (5m)
Thursday opening remarks
Ben Lorica (O'Reilly), Roger Chen (Computable)
Program chairs Ben Lorica and Roger Chen open the second day of keynotes.
9:05am-9:20am (15m)
Magenta: Machine learning and creativity
Doug Eck (Google Brain)
Doug Eck offers an overview of Magenta, a Google Brain project to develop new generative machine learning models for art and sound creation, allowing us to better understand how machine learning can be used by artists and musicians to make something new. Doug provides demos and explains where this work fits in with other AI research being done at Google and elsewhere.
9:25am-9:40am (15m) Interacting with AI
Cars that coordinate with people
Anca Dragan (UC Berkeley)
Autonomous cars tend to treat people like obstacles whose motion needs to be anticipated so that the car can best stay out of their way, resulting in ultradefensive cars that can't coordinate with people. Anca Dragan demonstrates how learning and optimal control can be leveraged to generate car behavior that results in natural coordination strategies.
9:40am-9:45am (5m)
Machine learning on Google Cloud Platform (sponsored by Google)
Amy Unruh (Google)
Amy Unruh offers a quick overview of machine learning on Google Cloud Platform and demonstrates a couple of the Google Cloud ML APIs. She then briefly highlights a few OSS TensorFlow models and explains how to use transfer learning to fine-tune them with your own data.
9:45am-10:00am (15m)
Superhuman AI for strategic reasoning: Beating top pros in heads-up no-limit Texas hold’em
Tuomas Sandholm (Carnegie Mellon University)
Tuomas Sandholm offers an overview of Libratus—an AI that beat a team of four top specialist pros in heads-up no-limit Texas hold’em, which has 10^161 decision points—and explains how Strategic Machine is applying the domain-independent algorithms behind Libratus to a variety of imperfect-information games.
10:00am-10:10am (10m)
Evolve AI (sponsored by Intel Nervana)
Naveen Rao (Intel)
Naveen Rao explains how Intel Nervana is evolving the AI stack from silicon all the way to the cloud so that true AI transformation can happen across every experience and every vertical.
10:10am-10:30am (20m)
Artificial intelligence in the software engineering workflow
Peter Norvig (Google)
Artificial intelligence is playing an increasingly important role in new software products, but the workflow of an AI researcher is quite different from the workflow of the software developer. Peter Norvig explains how the two can come together.
10:30am-10:35am (5m)
Closing remarks
Ben Lorica (O'Reilly), Roger Chen (Computable)
Program chairs Ben Lorica and Roger Chen close the second day of keynotes.
10:35am-11:05am (30m)
Break: Morning Break
3:15pm-4:00pm (45m)
Break: Afternoon Break
12:35pm-1:45pm (1h 10m)
Thursday Industry Tables at Lunch
Industry Table discussions are a great way to informally network with people in similar industries or interested in the same topics.
8:00am-9:00am (1h)
Break: Wake up coffee
8:15am-8:45am (30m)
Speed Networking
Gather before keynotes on Wednesday and Thursday morning for a speed networking event. Enjoy casual conversation while meeting new attendees.