• 08:00

    REGISTRATION & LIGHT BREAKFAST

  • 09:00
    Anirudh Koul

    WELCOME

    Anirudh Koul - Head of AI & Research - Aira

    Down arrow blue

    Anirudh Koul is a noted AI expert, UN/TEDx speaker, author and a former senior scientist at Microsoft AI & Research, where he founded Seeing AI, often considered the most used technology among the blind community after the iPhone. Anirudh serves as the Head of AI & Research at Aira, recognized by Time Magazine as one of the best inventions of 2018. He is also the author of O’Reilly’s ‘Practical Deep Learning for Cloud, Mobile and Edge’. With features shipped to a billion users, he brings over a decade of production-oriented Applied Research experience on PetaByte scale datasets. He has been developing technologies using AI techniques for Augmented Reality, Robotics, Speech, Productivity as well as Accessibility. His work in the AI for Good field, which IEEE has called ‘life-changing’, has received awards from CES, FCC, MIT, Cannes Lions, American Council of the Blind, showcased at events by UN, World Economic Forum, White House, House of Lords, Netflix, National Geographic, and lauded by world leaders including Justin Trudeau and Theresa May.

    Twitter Linkedin
  • DEEP LEARNING LANDSCAPE

  • 09:15
    Anh Nguyen

    How to Explain Neural Network Decisions?

    Anh Nguyen - Assistant Professor - Auburn University

    Down arrow blue

    How to Explain Neural Network Decisions?

    Understanding a deep learning model's inner-workings and decisions is increasingly important, especially for life-critical applications e.g. in medical diagnosis or criminal justice. In this talk, I will discuss our recent findings of some interesting failures of state-of-the-art image classifiers. For example, simply randomly rotating and randomly placing a familiar, training-set object in front of the camera is sufficient to bring the classification accuracy from 77.5% down to 3%. Such notorious brittleness of neural networks, therefore, begs for better explanations of why a model makes a certain decision. In this quest, I will share some recent work showing that interpretability methods are unreliable, being sensitive to hyperparameters and how harnessing generative models to synthesize counterfactual intervention samples can improve the robustness and accuracy of the attribution methods.

    Anh completed his Ph.D. in 2017 at the University of Wyoming, working with Jeff Clune and Jason Yosinski. His current research focus is Deep Learning, specifically explainable artificial intelligence and generative models. He has also worked as an ML research intern at Apple and Geometric Intelligence (now Uber AI Labs), and Bosch. Anh’s research has won 3 Best Paper Awards at CVPR, GECCO, ICML Visualization workshop, respectively, and 2 Best Research Video Awards at IJCAI and AAAI, respectively.

    Twitter Linkedin
  • 09:35
    Ari Morcos

    Training BatchNorm and Only BatchNorm:On the Expressive Power of Random Features in CNNs

    Ari Morcos - Research Scientist - Facebook AI Research

    Down arrow blue

    Training BatchNorm and Only BatchNorm:On the Expressive Power of Random Features in CNNs

    Batch normalization has become an indispensable tool for training deep neural networks, yet it remains poorly understood. Although the normalization component of batch normalization has been most emphasized, batch normalization also adds two per-feature trainable affine parameters: a coefficient, gamma, and a bias, beta. However, the impact of these oft-ignored parameters relative to the normalization component remains unclear. In this talk, I will discuss recent work which aims to understand the role and expressive power of these affine parameters. To do so, we study the performance achieved when training only these parameters and freezing all others at their random initializations. Surprisingly, we found that training these parameters alone leads to high, though not state of the art, performance. For example, on a sufficiently deep ResNet, training only the affine batch normalization parameters reaches 83% accuracy on CIFAR-10. Interestingly, batch normalization achieves this performance in part by naturally learning to disable around a third of the random features without any changes to the training objective. In this way, this experiment can be viewed as characterizing the expressive power of neural networks constructed simply by shifting and rescaling random features, and highlight the under-appreciated role of the affine parameters in batch normalization.

    Ari Morcos is a Research Scientist at Facebook AI Research working on understanding the mechanisms underlying neural network computation and function, and using these insights to build machine learning systems more intelligently. In particular, Ari has worked on understanding the properties predictive of generalization, methods to compare representations across networks, the role of single units in computation, and on strategies to measure abstraction in neural network representations. Previously, he worked at DeepMind in London, and earned his PhD in Neurobiology at Harvard University, using machine learning to study the cortical dynamics underlying evidence accumulation for decision-making.

    Twitter Linkedin
  • 09:55
    Sara Hooker

    What Does a Pruned Deep Neural Network "Forget"?

    Sara Hooker - Research Scholar - Google Brain

    Down arrow blue

    What Does a Pruned Deep Neural Network "Forget"?

    Between infancy and adulthood, the number of synapses in our brain first multiply and then fall. Despite losing 50% of all synapses between age two and ten, the brain continues to function. The phrase “use it or lose it” is frequently used to describe the environmental influence of the learning process on synaptic pruning, however there is little scientific consensus on what exactly is lost.

    In this talk, we explore what is lost when we prune a deep neural network. State of art pruning methods remove the majority of the weights in deep neural networks with minimal degradation to top-1 accuracy. However, the ability to prune networks with seemingly so little degradation to generalization performance is puzzling. The cost to top-1 accuracy appears minimal if it is spread in a uniform manner across all classes, but what if the cost is concentrated in only a few classes? Are certain types of examples or classes disproportionately impacted by pruning? Our findings help provide intuition into why so much capacity is needed in the first place and has implications for other objectives we may care about such as fairness or AI safety.

    Sara Hooker is a research scholar at Google Brain doing deep learning research on reliable explanations of model predictions for black-box models. Her main research interests gravitate towards interpretability, predictive uncertainty, model compression and security. In 2014, she founded Delta Analytics, a non-profit dedicated to bringing technical capacity to help non-profits across the world use machine learning for good. She grew up in Africa, in Mozambique, Lesotho, Swaziland, South Africa, and Kenya. Her family now lives in Monrovia, Liberia.

    Twitter Linkedin
  • 10:20

    COFFEE

  • REINFORCEMENT LEARNING

  • 10:50
    Dawn Song

    AI and Security: Lessons, Challenges and Future Directions

    Dawn Song - Professor of Computer Science - UC Berkeley

    Down arrow blue

    AI and Security: Lessons, Challenges and Future Directions

    In this talk, I will talk about challenges and exciting new opportunities at the intersection of AI and Security,

    how AI and deep learning can enable better security, and how Security can enable better AI. In particular, I will talk about secure deep learning and challenges and approaches to ensure the integrity of decisions made by deep learning. I will also give an overview on challenges and new techniques to enable privacy-preserving machine learning. Finally, I will conclude with future directions at the intersection of AI and Security.

    Dawn Song is a Professor in the Department of Electrical Engineering and Computer Science at UC Berkeley. Her research interest lies in deep learning, security, and blockchain. She has studied diverse security and privacy issues in computer systems and networks, including areas ranging from software security, networking security, distributed systems security, applied cryptography, blockchain and smart contracts, to the intersection of machine learning and security. She is the recipient of various awards including the MacArthur Fellowship, the Guggenheim Fellowship, the NSF CAREER Award, the Alfred P. Sloan Research Fellowship, the MIT Technology Review TR-35 Award, the Faculty Research Award from IBM, Google and other major tech companies, and Best Paper Awards from top conferences in Computer Security and Deep Learning. She is an IEEE Fellow. She is ranked the most cited scholar in computer security (AMiner Award). She obtained her Ph.D. degree from UC Berkeley. Prior to joining UC Berkeley as a faculty, she was a faculty at Carnegie Mellon University from 2002 to 2007. She is also a serial entrepreneur.

    Twitter Linkedin
  • 11:10
    Jeff Clune

    ANML: Learning to Continually Learn

    Jeff Clune - Research Manager / Team Lead - OpenAI

    Down arrow blue

    ANML: Learning to Continually Learn

    Jeff Clune is the Research Manager / Team Lead at OpenAI. He focuses on robotics, reinforcement learning, and training neural networks either via deep learning or evolutionary algorithms. He has also researched open questions in evolutionary biology using computational models of evolution, including the evolutionary origins of modularity, hierarchy, and evolvability. Previously, Jeff was a Senior Research Manager and founding member at Uber AI Labs, and also the Loy and Edith Harris Associate Professor in Computer Science at the University of Wyoming. Prior to becoming a professor, he was a Research Scientist at Cornell University, received a PhD in computer science and an MA in philosophy from Michigan State University, and received a BA in philosophy from the University of Michigan.

    Twitter Linkedin
  • 11:30
    Ken Goldberg

    The New Wave in Robot Grasping

    Ken Goldberg - Professor - UC Berkeley

    Down arrow blue

    The New Wave in Robot Grasping

    Despite 50 years of research, robots remain remarkably clumsy, limiting their reliability for warehouse order fulfillment, robot-assisted surgery, and home decluttering. The First Wave of grasping research is purely analytical, applying variations of screw theory to exact knowledge of pose, shape, and contact mechanics. The Second Wave is purely empirical: end-to-end hyperparametric function approximation (aka Deep Learning) based on human demonstrations or time-consuming self-exploration. A "New Wave" of research considers hybrid methods that combine analytic models with stochastic sampling and Deep Learning models. I'll present this history with new results from our lab on grasping diverse and previously-unknown objects.

    Ken Goldberg is an artist, inventor, and roboticist. He is William S. Floyd Jr Distinguished Chair in Engineering at UC Berkeley and Chief Scientist at Ambidextrous Robotics. Ken is on the Editorial Board of the journal Science Robotics, served as Chair of the Industrial Engineering and Operations Research Department, and co-founded the IEEE Transactions on Automation Science and Engineering. Short documentary films he co-wrote were selected for Sundance and one was nominated for an Emmy Award. Ken and his students have published 300 peer-reviewed papers, 9 US patents, and created award-winning artworks featured in 70 exhibits worldwide.

    Twitter Linkedin
  • 11:50
    Ilya Sutskever

    The Power of Large scale RL and generative models

    Ilya Sutskever - Co-Founder & Chief Scientist - OpenAI

    Down arrow blue

    The Power of Large scale RL and generative models

    Ilya Sutskever received his PhD in 2012 from the University of Toronto working with Geoffrey Hinton. After completing his PhD, he cofounded DNNResearch with Geoffrey Hinton and Alex Krizhevsky which was acquired by Google. He is interested in all aspects of neural networks and their applications.

    Twitter Linkedin
  • 12:10
    Ilya Sutskever

    FIRESIDE CHAT: With Ilya Sutskever & Lex Fridman

    Ilya Sutskever - Co-Founder & Chief Scientist - OpenAI

    Down arrow blue

    The Power of Large scale RL and generative models

    Ilya Sutskever received his PhD in 2012 from the University of Toronto working with Geoffrey Hinton. After completing his PhD, he cofounded DNNResearch with Geoffrey Hinton and Alex Krizhevsky which was acquired by Google. He is interested in all aspects of neural networks and their applications.

    Twitter Linkedin
  • Lex Fridman

    Moderator

    Lex Fridman - Researcher - MIT

    Down arrow blue

    An Introduction to Reinforcement Learning

    Lex Fridman is a researcher at MIT, working on deep learning approaches in the context of semi-autonomous vehicles, human sensing, personal robotics, and more generally human-centered artificial intelligence systems. He is particularly interested in understanding human behavior in the context of human-robot collaboration, and engineering learning-based methods that enrich that collaboration. Before joining MIT, Lex was at Google working on machine learning for large-scale behavior-based authentication.

    Twitter Linkedin
  • 12:40

    LUNCH

  • 13:40
    Aravind Srinivas

    Massively Improving Data-efficiency of Supervised Learning Systems using Self-supervision from Unlabeled Data

    Aravind Srinivas - Ph.D. Student - UC Berkeley

    Down arrow blue

    Massively Improving Data-efficiency of Supervised Learning Systems using Self-supervision from Unlabeled Data

    I will talk about some recent work on improving the data efficiency of supervised learning models by learning rich representations of unlabeled data. Using self-supervised learning methods to predict missing information from a given context, useful features can be learned and used for downstream tasks with very few labels. It is possible to now match the performance of powerful image recognition systems such as AlexNet and VGG using as few as 2% and 10% of the labeled data respectively on the widely benchmarked ImageNet dataset. The benefits from pre-training also remain in the high data regime with the performance improving to 80% top-1 accuracy on ImageNet.

    Aravind is a Ph.D. student at UC Berkeley advised by Prof. Pieter Abbeel where he co-created and taught the first edition of the Deep Unsupervised Learning class. He has spent time at OpenAI and DeepMind and is broadly interested in unsupervised representation learning.

    Twitter Linkedin
  • 14:00
    Abhinav Gupta

    Towards Self-supervised Curious Robots

    Abhinav Gupta - Associate Professor and Research Manager - Carnegie Mellon University & Facebook AI Research (FAIR)

    Down arrow blue

    Towards Self-supervised Curious Robots

    In the last decade, we have made significant advances in the field of artificial intelligence thanks to supervised learning. But this passive supervision of our models has now become our biggest bottleneck. In this talk, I will discuss our efforts towards scaling up and empowering visual and robotic learning. First, I will show how amount of labeled data is crucial factor in learning. I will then describe how we can overcome the passive supervision bottleneck by self-supervised learning. Next, I will discuss how embodiment is crucial for learning -- our agents live in the physical world and need the ability to interact in the physical world. Towards this goal, I will finally present our efforts in large-scale learning of embodied agents in robotics. Finally, I will discuss how we can move from passive supervision to active exploration -- the ability of agents to create their own training data.

    Abhinav Gupta is an Associate Professor at the Robotics Institute, Carnegie Mellon University and Research Manager at Facebook AI Research (FAIR). Abhinav's research focuses on scaling up learning by building self-supervised, lifelong and interactive learning systems. Specifically, he is interested in how self-supervised systems can effectively use data to learn visual representation, common sense and representation for actions in robots. Abhinav is a recipient of several awards including ONR Young Investigator Award, PAMI Young Researcher Award, Sloan Research Fellowship, Okawa Foundation Grant, Bosch Young Faculty Fellowship, YPO Fellowship, IJCAI Early Career Spotlight, ICRA Best Student Paper award, and the ECCV Best Paper Runner-up Award. His research has also been featured in Newsweek, BBC, Wall Street Journal, Wired and Slashdot.

    Twitter
  • COMPUTER VISION

  • 14:20
    Georgia Gkioxari

    Visual Recognition beyond 2D

    Georgia Gkioxari - Research Scientist - Facebook AI Research (FAIR)

    Down arrow blue

    Visual Recognition beyond 2D

    Undoubtedly 2D visual recognition has seen unprecedented success, with the state of the art advancing every single conference cycle. But as we develop sophisticated machines that predict 2D object masks and object classes, we tend to ignore the fact that the world is not 2D and objects don't live in a 2D grid. On the other hand, the focus of 3D object recognition is dramatically different than its 2D counterpart, with benchmarks that lack in complexity compared to COCO or ImageNet and models that can not tackle diversity in appearance and shapes across object instances. In this talk, I will present some of our efforts to marry the advances in 2D recognition with 3D shape inference in the wild. I will also introduce our new library of 3D operators which builds on PyTorch and contains highly optimized 3D operations (including a differentiable renderer!) which are essential when designing and training deep learning models with 3D data sources. Lastly, I will present our new project on novel view synthesis for real complex scenes, in an effort to convince you that 3D representations can be quite impactful in a variety of tasks!

    Georgia Gkioxari is a research scientist at Facebook AI Research (FAIR). She received a PhD in computer science and electrical engineering from the University of California at Berkeley under the supervision of Jitendra Malik in 2016. Her research interests lie in computer vision, with a focus on object and person recognition from static images and videos. In 2017, Georgia received the Marr Prize at ICCV for "Mask R-CNN".

    Twitter Linkedin
  • 14:40
    Shalini Ghosh

    Multi-modal Video Content Analysis for Content Recommendation

    Shalini Ghosh - Principal Scientist (Global) and ML Research Leader, Visual Display Intelligence Lab (Smart TV Division) - Samsung Research America

    Down arrow blue

    Multi-modal Video Content Analysis for Content Recommendation

    Multimodal AI is useful for video content analysis in a variety of problems, e.g., Visual Dialog, Object Detection, Scene Understanding, 
Content Recommendation, etc. In this talk, we will focus on the problem of multimodal content recommendation, where an AI agent processes data in multiple modalities (e.g., video, images, audio, language) and learns how to recommend new multi-media content to an user. We show how the AI agent can process different characteristics of the current video as well as those in the user history, and infer the relevant matching characteristics based on which the AI agent will be able to recommend new videos of interest to the user. The particular characteristics we focus on are the fine-grained categories of the video. In this context, we have developed some key innovations: (a) a hierarchy of fine-grained categories, customized to the domain under consideration; (b) novel modeling approaches like temporal coherence-based regularization and hierarchical loss function, which improve the accuracy of the deep learning models in getting accurate predictions of fine-grained categories from videos; and (c) a novel multi-modal fusion architecture, which uses approaches like sparse fusion and gated mixture of experts to combine predictions from multiple modalities and get the final category prediction. We will discuss how these innovations come together in our proposed architecture for multimodal content recommendation, which out-performs current state-of-the-art models in performance.

    Dr. Shalini Ghosh is Principal Scientist and Leader of the Machine Learning Research at the Visual Display Intelligence Lab of Samsung Research America, where she leads a group working on Multi-modal AI (i.e., learning from vision, language, and speech). Before this she was the Director of AI Research at Samsung Research America. She has extensive experience and expertise in Machine Learning (ML), especially Deep Learning, and has worked on applications of ML to multiple domains. Before joining Samsung Research, Dr. Ghosh was a Principal Computer Scientist in the Computer Science Laboratory at SRI International, where she has been the Principal Investigator/Tech Lead of several impactful DARPA and NSF projects. She was also a Visiting Scientist at Google Research in 2014-2015, where she worked on applying deep learning (Google Brain) models to dialog systems and natural language applications. Dr. Ghosh has a Ph.D. in Computer Engineering from the University of Texas at Austin. She has won several grants and awards for her research, including a Best Paper award and a Best Student Paper Runner-up award for applications of ML to dependable computing. Dr. Ghosh is also an area chair of ICML and serves on the program committee of multiple impactful conferences and journals in ML and AI (e.g., NIPS, KDD, AAAI, IJCAI). She has served as invited panelist in multiple panels, and was invited to be a guest lecturer at UC Berkeley multiple times. Her work has been covered in an interview by the ReWork Women in AI program.

    Twitter Linkedin
  • 15:00
    Xinlei Chen

    Progress on Joint Vision and Language Understanding

    Xinlei Chen - Research Scientist - Facebook AI Research (FAIR)

    Down arrow blue

    Progress on Joint Vision and Language Understanding

    In this talk, I am going to introduce some of the recent efforts we have made on joint vision and language understanding at Facebook AI Research. First, we study the current dominating "bottom-up attention" features used by state-of-the-art VQA systems, and show that vanilla convolutional feature maps, or grid features, can perform similarly well while offering significant speed-ups to the pipeline. This finding not only gives us better understanding of the problem and the models, but also opens up new opportunities in vision and language research in general. Second, I am going to briefly mention some other new developments in this research direction, including 1) reasoning with both objects and texts in the scene; and 2) making the system more robust to question rephrasing through generative modeling and cycle-consistency training. Finally, I am also introducing the open sourced library from FAIR, Pythia, which helped us won two VQA challenges last year, and hope to further facilitate research for the community.

    I am a Research Scientist at Facebook AI Research, Menlo Park.

    I was a PhD student at Language Technology Institute, Carnegie Mellon University, from September 2012 to Feburary 2018, working mainly with Prof. Abhinav Gupta on computer vision, computational linguistics and the combination of both. Durthing the time at CMU, I had also worked with Prof. Tom Mitchell. In spring 2014, I did an internship in MSR with Prof. C. Lawrence Zitnick. Then in summer 2016 I did an internship in Prof. William T. Freeman's VisCAM group at Google. Before graduation, I also spent time at Google Cloud AI team working with Prof. Fei-Fei Li and Dr. Jia Li.

    I graduated with a bachelor's degree in computer science from Zhejiang University, China. During my undergraduate study, I was mainly under the supervision of Prof. Deng Cai in the State Key Laboratory of CAD & CG. I was a summer intern at UCLA in 2011, mainly work with Prof. Jenn Wortman Vaughan.

    Twitter Linkedin
  • 15:20

    COFFEE

  • EXPLAINABILITY

  • 16:00
    Sameer Singh

    Detecting Bugs and Explaining Predictions of Machine Learning Models

    Sameer Singh - Assistant Professor - University of California, Irvine

    Down arrow blue

    Debugging and Explaining Machine Learning Models

    Machine learning is at the forefront of many recent advances in science and technology, enabled in part by the sophisticated models and algorithms that have been recently introduced. However, as a consequence of this complexity, machine learning essentially acts as a black-box as far as users are concerned, making it incredibly difficult to understand, predict, or detect bugs in their behavior. For example, determining when a machine learning model is “good enough” is challenging since held-out accuracy metrics significantly overestimate real-world performance. I will describe our research on approaches that explain the predictions of any classifier in an interpretable and faithful manner, and automated techniques to detect bugs that can occur naturally when a model is deployed. I will cover various ways in which we summarize this relationship: as linear weights, as precise rules, and as counter-examples, and present examples that demonstrate their utility in understanding, and debugging, black-box machine learning algorithms.

    Dr. Sameer Singh is an Assistant Professor of Computer Science at the University of California, Irvine (UCI). He is working on robustness and interpretability of machine learning algorithms, along with models that reason with text and structure for natural language processing. Sameer was a postdoctoral researcher at the University of Washington and received his PhD from the University of Massachusetts, Amherst, during which he also worked at Microsoft Research, Google Research, and Yahoo! Labs. He was selected as a DARPA Riser, was awarded the Adobe Research Data Science Award, the grand prize in the Yelp dataset challenge, the Yahoo! Key Scientific Challenges award and the UMass Graduate School fellowship. His group has received funding from Allen Institute for AI, NSF, DARPA, Adobe Research, and FICO.

    Twitter Linkedin
  • 16:20

    PANEL: Explainable AI - Solving the Black Box Problem

  • Michela Paganini

    PANELIST

    Michela Paganini - Postdoctoral Researcher - Facebook AI Research

    Down arrow blue

    Michela Paganini is a postdoctoral researcher at Facebook AI Research in Menlo Park, and an affiliate at Lawrence Berkeley National Lab. She joined Facebook in 2018 after earning her PhD in particle physics from Yale University. Her work focuses on empirically characterizing neural network dynamics in the over-parametrized and under-parametrized regimes using tools from theoretical and experimental physics, and on making current AI models simpler and faster to compute on modern hardware architectures, while connecting emergent behavior in constrained networks to theoretical predictions. During her graduate studies, she worked on the design, development, and deployment of deep learning algorithms for the ATLAS experiment at CERN, with a focus on computer vision and generative modeling."

    Twitter Linkedin
  • Sameer Singh

    PANELIST

    Sameer Singh - Assistant Professor - University of California, Irvine

    Down arrow blue

    Debugging and Explaining Machine Learning Models

    Machine learning is at the forefront of many recent advances in science and technology, enabled in part by the sophisticated models and algorithms that have been recently introduced. However, as a consequence of this complexity, machine learning essentially acts as a black-box as far as users are concerned, making it incredibly difficult to understand, predict, or detect bugs in their behavior. For example, determining when a machine learning model is “good enough” is challenging since held-out accuracy metrics significantly overestimate real-world performance. I will describe our research on approaches that explain the predictions of any classifier in an interpretable and faithful manner, and automated techniques to detect bugs that can occur naturally when a model is deployed. I will cover various ways in which we summarize this relationship: as linear weights, as precise rules, and as counter-examples, and present examples that demonstrate their utility in understanding, and debugging, black-box machine learning algorithms.

    Dr. Sameer Singh is an Assistant Professor of Computer Science at the University of California, Irvine (UCI). He is working on robustness and interpretability of machine learning algorithms, along with models that reason with text and structure for natural language processing. Sameer was a postdoctoral researcher at the University of Washington and received his PhD from the University of Massachusetts, Amherst, during which he also worked at Microsoft Research, Google Research, and Yahoo! Labs. He was selected as a DARPA Riser, was awarded the Adobe Research Data Science Award, the grand prize in the Yelp dataset challenge, the Yahoo! Key Scientific Challenges award and the UMass Graduate School fellowship. His group has received funding from Allen Institute for AI, NSF, DARPA, Adobe Research, and FICO.

    Twitter Linkedin
  • Chandra Khatri

    PANELIST

    Chandra Khatri - Senior AI Scientist - Uber AI

    Down arrow blue

    Incorporating Common Sense and Semantic Understanding within the Assistants

    With advancements in Deep Learning and data collection techniques, we have built artificial agents which more functional such that they can perform significantly better than humans on well-defined and unambiguous tasks such as Atari games. However, they do poorly on tasks that are dynamic and seem straightforward to humans such as embodied navigation and performing open-ended conversations, even after training with millions of training samples. One key element which differentiates humans from artificial agents in performing various tasks is that humans have access to common sense and semantic understanding, which is learned from past experiences. In this talk, I will be presenting how incorporating common sense and semantic understanding significantly help the agents in performing a complex task such as house navigation. I will also showcase that the semantic embeddings learned by the agent mimic the structural and positional patterns of the environment.

    Chandra Khatri is a Senior AI Scientist at Uber AI driving Conversational AI efforts at Uber. Prior to Uber, he was the Lead AI Scientist at Alexa and was driving the Science for the Alexa Prize Competition, which is a $3.5 Million university competition for advancing the state of Conversational AI. Some of his recent work involves Open-domain Dialog Planning and Evaluation, Conversational Speech Recognition, Conversational Natural Language Understanding, and Sequential Modeling.

    Prior to Alexa, Chandra was a Research Scientist at eBay, wherein he led various Deep Learning and NLP initiatives such as Automatic Text Summarization and Automatic Content Generation within the eCommerce domain, which has lead to significant gains for eBay. He holds degrees in Machine Learning and Computational Science & Engineering from Georgia Tech and BITS Pilani.

    Twitter Linkedin
  • Krishna Sankar

    MODERATOR

    Krishna Sankar - Distinguished Engineer – AI - U.S. Bank

    Down arrow blue

    Krishna Sankar is a Distinguished Engineer – AI, at the Chief Digital Office of U.S. Bank, focusing on embedding intelligence – all aspects incl AI Ethics, Explainability and Conversational AI, in the financial sector. Earlier stints include Senior Data Scientist/Volvo, Chief Data Scientist/blackarrow.tv, Data Scientist/Tata America Intl, Director of Data Science/Bioinformatics startup & as a Distinguished Engineer/Cisco. His external work includes teaching, writing blogs and Lego Robotics. He has been speaking at various conferences incl Nvidia GTC2019 and GTC2020 [https://goo.gl/bxw6Qd], ML tutorials at Strata SJC & LONDON 2016, Spark Summit and others. His occasional blogs can be found at https://medium.com/@ksankar. They include Conversational AI [bit.ly/37PjPBh], Rebooting AI [bit.ly/2ststqi41], The Excessions of xAI [bit.ly/2LDXe2c], Robots Rules[goo.gl/5yyRv6], NeurIPS2018 [goo.gl/VgeyDT] and Garry Kasparov’s Deep Thinking [goo.gl/9qv671] He has done guest lecturing at the Naval Post Graduate School, Monterey; and you will find him at the FLL (Lego Robotics) World Competition as Robots Design Judge

    Twitter Linkedin
  • 17:00

    CONVERSATION & DRINKS

  • 08:00

    DOORS OPEN

  • 09:00
    Azalia Mirhoseini

    WELCOME

    Azalia Mirhoseini - Senior Research Scientist - Google Brain

    Down arrow blue

    My name is Azalia Mirhoseini. I am a Senior Research Scientist at Google Brain. I am the co-founder/tech-lead of the Machine Learning for Systems Moonshot at Brain where we focus on deep reinforcement learning based approaches to solve problems in computer systems and metalearning. I have a Ph.D. in Electrical and Computer Engineering from Rice University. I have received a number of awards, including the MIT Technology Review 35 under 35 award, the Best Ph.D. Thesis Award at Rice and a Gold Medal in the National Math Olympiad in Iran.

    Twitter Linkedin
  • IMPROVING MACHINE LEARNING PROJECTS

  • 09:15
    Lukas Biewald

    Why are Machine Learning Projects So Hard to Manage?

    Lukas Biewald - Co-Founder and CEO - Weights and Biases

    Down arrow blue

    Why are Machine Learning Projects So Hard to Manage?

    I’ve watched lots of companies attempt to deploy machine learning — some succeed wildly and some fail spectacularly. One constant is that machine learning teams have a hard time setting goals and setting expectations. This talk will give some examples of how teams fail and recommendations for everyone from executives to researchers to make their machine learning projects work better.

    Lukas Biewald is a co-founder and CEO of Weights and Biases which builds performance and visualization tools for machine learning teams and practitioners. Lukas also cofounded Figure Eight (formerly CrowdFlower) — a human in the loop platform that transforms unstructured text, image, audio, and video data into customized high quality training data. Prior to co-founding Weights and Biases and CrowdFlower, Biewald was a Senior Scientist and Manager within the Ranking and Management Team at Powerset, a natural language search technology company later acquired by Microsoft. From 2005 to 2006, Lukas also led the Search Relevance Team for Yahoo! Japan. Weights & Biases (wandb) helps you track your machine learning experiments. Easily add our package, wandb, to your model script to log hyperparameters and output metrics from your runs, explore model architectures, and compare results. Once you install our library we make it easy to share your results with colleagues and your future self.

    Twitter Linkedin
  • DEEP LEARNING FOR HEALTHCARE

  • 09:30
    Ivana Williams

    A Method for Automated Feed Generation Based on User’s Research Interests

    Ivana Williams - Staff Research Scientist - Chan Zuckerberg Initiative

    Down arrow blue

    A Method for Automated Feed Generation Based on User’s Research Interests

    Meta is a tool that helps scientists discover biomedical research. We organize and track over 67 million researchers, diseases, genes, proteins, pathways, and more — including full coverage of papers from PubMed and preprints from bioRxiv. In our platform currently, a feed comprises of a set of papers which are retrieved from our knowledge graph using a boolean query, composed from different entities from our knowledge graph (e.g. UMLS concepts, MeSH terms, journals, authors, etc.) and ranked based on their relevancy, publication date, and impact. In this paper, we describe a method used to automate and personalize a new users' onboarding experience, by building an automated feed (autofeed) based on minimal input from the user. Users' inputs, for example, may include keywords, free text, papers that are related to their research interests, etc. The goal of our method is to, given the user's inputs, compose an expanded query that retrieves papers that are highly specific and related to the user's research interests. This method, in turn, enables us to create a highly personalized and flexible user experience within our platform. We describe the following components of the autofeeds algorithm: data ingestion and pre-processing, document embedding which leverages the state of the art contextual embedding models, such as BioSentVec and BioBERT, hierarchical document clustering and the query composition approach. We conclude with a discussion of our qualitative and quantitative evaluation of independent components and the full onboarding process.

    Ivana Williams is a Staff Research Scientist at the Chan Zuckerberg Initiative, Meta team working on unlocking scientific insights from scientific publications. She is applying the state of the art machine learning and natural language processing methods to the generation of knowledge graphs and recommendation systems, research interest modeling, extraction and disambiguation of scientific elements and relationships as well as the development of novel predictive capabilities. Most recently she has been leading personalization efforts within Meta, including personalized user onboarding via personalized feed generation and recommendations. She holds a Master's degree in Mathematics and Statistics from Georgetown University.

    Twitter Linkedin
  • 09:55
    Daphne Koller

    Machine Learning: A New Approach to Drug Discovery

    Daphne Koller - CEO - insitro

    Down arrow blue

    Machine Learning: A New Approach to Drug Discovery

    Modern medicine has given us effective tools to treat some of the most significant and burdensome diseases. At the same time, it is becoming consistently more challenging to develop new therapeutics: clinical trial success rates hover around the mid-single-digit range; the pre-tax R&D cost to develop a new drug (once failures are incorporated) is estimated to be greater than $2.5B; and the rate of return on drug development investment has been decreasing linearly year by year, and some analyses estimate that it will hit 0% before 2020. A key contributor to this trend is that the drug development process involves multiple steps, each of which involves a complex and protracted experiment that often fails. We believe that, for many of these phases, it is possible to develop machine learning models to help predict the outcome of these experiments, and that those models, while inevitably imperfect, can outperform predictions based on traditional heuristics. The key will be to train powerful ML techniques on sufficient amounts of high-quality, relevant data. To achieve this goal, we are bringing together cutting edge methods in functional genomics and lab automation to build a bio-data factory that can produce relevant biological data at scale, allowing us to create large, high-quality datasets that enable the development of novel ML models. Our first goal is to engineer in vitro models of human disease that, via the use of appropriate ML models, are able to provide good predictions regarding the effect of interventions on human clinical phenotypes. Our ultimate goal is to develop a new approach to drug development that uses high-quality data and ML models to design novel, safe, and effective therapies that help more people, faster, and at a lower cost.

    Daphne Koller is the CEO and Founder of insitro, a startup company that aims to rethink drug development using machine learning. She is also the Co-Chair of the Board and Co-Founder of Coursera, the largest platform for massive open online courses (MOOCs). Daphne was the Rajeev Motwani Professor of Computer Science at Stanford University, where she served on the faculty for 18 years. She has also been the Chief Computing Officer of Calico, an Alphabet company in the healthcare space. She is the author of over 200 refereed publications appearing in venues such as Science, Cell, and Nature Genetics. Daphne was recognized as one of TIME Magazine’s 100 most influential people in 2012 and Newsweek’s 10 most important people in 2010. She has been honored with multiple awards and fellowships during her career including the Sloan Foundation Faculty Fellowship in 1996, the ONR Young Investigator Award in 1998, the Presidential Early Career Award for Scientists and Engineers (PECASE) in 1999, the IJCAI Computers and Thought Award in 2001, the MacArthur Foundation Fellowship in 2004, and the ACM Prize in Computing in 2008. Daphne was inducted into the National Academy of Engineering in 2011 and elected a fellow of the American Academy of Arts and Sciences in 2014 and of the International Society of Computational Biology in 2017. Her teaching was recognized via the Stanford Medal for Excellence in Fostering Undergraduate Research, and as a Bass University Fellow in Undergraduate Education.

    Twitter Linkedin
  • 10:20

    COFFEE

  • GENERATIVE MODELS

  • 11:00
    Dumitru Erhan

    Enabling World Models via Unsupervised Representation Learning of Environments

    Dumitru Erhan - Staff Research Scientist - Google Brain

    Down arrow blue

    Enabling World Models via Unsupervised Representation Learning of Environments

    Recent advances in deep neural networks have enabled impressive and often superhuman performance in tasks such as object recognition, object detection, segmentation, image description, visual question-answering and even medical image diagnosis. In many such scenarios, achieving state of the art performance requires collecting large amounts human-labeled data, which is expensive to acquire. In order to build intelligent agents that quickly adapt to new scenes, conditions, tasks, we need to develop techniques, algorithms and models that can operate on little data or that can generalize from training data that is not similar to the test data. World Models have long been hypothesized to be a key piece in the solution to this problem. In this talk I will describe our recent advances for modeling sequential observations. These approaches can help with building agents that interact with the environment and mitigate the sample complexity problems in reinforcement learning. They can also enable agents that generalize quicker to new scenarios, tasks, objects and situations and are thus more robust to environment changes.

    Dumitru Erhan is a Staff Research Scientist in the Google Brain team in San Francisco. He received a PhD from University of Montreal (MILA) in 2011 with Yoshua Bengio, where he worked on understanding deep networks. Afterwards, he has done research at the intersection of computer vision and deep learning, notably object detection (SSD), object recognition (GoogLeNet), image captioning (Show & Tell), visual question-answering, unsupervised domain adaptation (PixelDA), active perception and others. Recent work has focused on video prediction and generation, as well as its applicability to model-based reinforcement learning. He aims to build and understand agents that can learn as much as possible to self-supervised interaction with the environment, with applications to the fields of robotics and self-driving cars.

    Twitter Linkedin
  • 11:25
    Balaji Lakshminarayanan

    Do Deep Generative Models Know What They Don't Know

    Balaji Lakshminarayanan - Staff Research Scientist - DeepMind

    Down arrow blue

    Do Deep Generative Models Know What They Don't Know?

    A neural network deployed in the wild may be asked to make predictions for inputs that were drawn from a different distribution than that of the training data. Generative models are widely viewed to be a solution for detecting out-of-distribution (OOD) inputs and distributional skew, as they model the density of the input features p(x). We challenge this assumption by presenting several counter-examples. We find that deep generative models, such as flow-based models, VAEs and PixelCNN, which are trained on one dataset (e.g. CIFAR-10) can assign higher likelihood to OOD inputs from another dataset (e.g. SVHN). We further investigate some of these failure modes in detail, that help us better understand this surprising phenomenon, and potentially fix them.

    Balaji Lakshminarayanan is a staff research scientist at Google DeepMind. He is interested in scalable probabilistic machine learning and its applications. Most recently, his research has focused on probabilistic deep learning, specifically, deep generative models, uncertainty estimation and robustness to out-of-distribution inputs. He received his PhD from the Gatsby Unit, University College London where he worked with Yee Whye Teh.

    Twitter Linkedin
  • DEEP LEARNING FOR SOCIAL ROBOTS

  • 11:45
    Maja Matarić

    Socially Assistive Agents and Robots: What We Need to Learn About People

    Maja Matarić - Professor of Computer Science, Neuroscience, and Pediatrics - USC

    Down arrow blue

    Socially Assistive Agents and Robots: What We Need to Learn About People

    Digital assistants are becoming ubiquitous, and robots are following closely behind, both entering everyday lives and interacting with non-experts in unstructured settings. Digital assistants provide information and robots do physical work, but both can do much more to serve as important tools for goal-driven challenges in human behavior, such as health-related behavior change. Behavior change is difficult, and requires access to general data / knowledge, but also user data, and strategies for assisting effectively and adaptively. Embodied agents can leverage their expressivity to be more effective, but this involves challenges of embodied communication, social dynamics, and long-term adaptation and learning, bringing together the latest advances in AI, robotics, and human-machine interaction. This talk will discuss advances and opportunities in embodied assistive agents, including socially assistive robotics, and the often surprising and unintuitive role of machine learning, presenting the opportunity to drive the state of the art of those converging technologies with opportunities for impact on major societal challenges.

    Maja Matarić is Chan Soon-Shiong Professor of Computer Science, Neuroscience, and Pediatrics at the University of Southern California, founding director of the USC Robotics and Autonomous Systems Center, and Vice Dean for Research in the Viterbi School of Engineering. Her PhD and MS are from MIT, and BS from the University of Kansas. She is Fellow of AAAS, IEEE, and AAAI, and received the Presidential Award for Excellence in Science, Mathematics and Engineering Mentoring, Anita Borg Institute Women of Vision Award in Innovation, the Okawa Foundation, NSF Career, MIT TR35, and IEEE RAS Early Career Awards. A pioneer of socially assistive robotics, her research enables robots to help people through social interaction in therapy, rehabilitation, training, and education, developing robot-assisted therapies for autism, stroke, Alzheimer's and other special needs, as well as wellness interventions (http://robotics.usc.edu/interaction/).

    Linkedin
  • DEEP LEARNING FOR MEDIA & ENTERTAINMENT

  • 12:05
    Trung Nguyen

    Instant Recommendations

    Trung Nguyen - Senior Machine Learning Scientist - Netflix Research

    Down arrow blue

    Instant Recommendations

    The enormous scale of the intangible economy has created much better services and countless choices for Internet users. Online stores, without shelf space constraints, could offer their customers direct access to huge warehouses. Netflix could offer its members quality content to consume as much as they may have time to watch. A key point to a successful user experience is an effective way for users to explore and find what they want, which could be something that users do not know yet. I believe that recommendations systems that can instantly incorporate and response to subtle changes in a user action either implicit or explicit, would be critical. In this talk, I will discuss some challenges and approaches to designing these machine learning models.

    Trung is a Research Scientist at Netflix, where his main focus is to advance the search and recommendation systems serving hundreds of millions of users. He is interested in self-supervised learning methods, deep reinforcement learning, and imitation learning. Prior to Netflix, Trung was a lead researcher at Adobe working on display advertising and conversational bot. Trung holds a PhD from National University of Singapore where he researched and published work on model-based reinforcement learning.

    Twitter Linkedin
  • 12:25
    Oriol Nieto

    Recommending Music with Waveform-based Architectures

    Oriol Nieto - Senior Scientist - Pandora

    Down arrow blue

    Recommending Music with Waveform-based Architectures

    In this talk we discuss deep models that use waveform representations as input to estimate an embedded space where distances become meaningful when recommending music. Such space is initially produced by analyzing listener behavior, thus becoming a powerful tool to recommend the most popular content in a given music catalog but weak in terms of addressing the so-called "cold start problem," i.e., to recommend content that has spun infrequently or never at all and thus, has little listener behavior data associated with it. We show how, given enough data, deep waveform-based architectures [1] can estimate such spaces more accurately than spectrogram-based ones. Moreover, by using other sources of data (e.g., human labeled music attributes such as the ones in the Music Genome Project) with late-fusion multimodal networks [2], we achieve higher accuracy when predicting these embedded spaces. Finally, several musical examples are explored to further illustrate the recommendation results.

    [1] Pons, J., Nieto, O., Prockup, M., Schmidt, E., Ehmann, A., Serra, X., End-to-End Learning for Music Audio Tagging at Scale. Proc. of the 19th International Society for Music Information Retrieval Conference (ISMIR). Paris, France, 2018

    [2] Oramas, S., Barbieri, F., Nieto, O., Serra, X., Multimodal Deep Learning for Music Genre Classification. Transactions of the International Society for Music Information Retrieval (TISMIR). 2018

    Oriol Nieto is a Senior Scientist at Pandora, where he aims at improving the long tail music recommendations. Prior to that, he defended his Ph.D Dissertation in the Music and Audio Research Lab at NYU focusing on the automatic analysis of structure in music. He holds an M.A. in Music, Science, and Technology from the Center for Computer Research in Music and Acoustics at Stanford University, an M.S. in Information Theories from the Music Technology Group at Pompeu Fabra University, and a Bachelor’s degree in Computer Science from the Polytechnic University of Catalonia. His research focuses on music information retrieval, large scale recommendation systems, and machine learning with especial emphasis on deep architectures. Oriol plays guitar, violin, and sings (and screams) in his spare time.

    Twitter Linkedin
  • 12:45

    LUNCH

  • LARGE-SCALE ML DEVELOPMENT

  • 13:40
    Eero Laaksonen

    Asynchronous Model Training & Version Control for Jupyter Notebooks with Valohai

    Eero Laaksonen - CEO - Valohai

    Down arrow blue

    Asynchronous Model Training & Version Control for Jupyter Notebooks with Valohai

    Valohai is an end-to-end Machine Learning Platform that takes care of MLOps, automatic version control, and team collaboration. While notebooks are great for early experimentation with ML models, they are not ideal to facilitate company-wide, transparent and production-scale ML development. Jupyhai is Valohai’s Jupyter Notebook extension, letting data scientists run their notebook experiments asynchronously on the cloud while having automatic versioning for code, data, hyperparameters, and more. Time to take your Jupyter game to the next level!

    Eero is a San Francisco based startup founder and CEO. Their company Valohai focuses on large-scale, in-production machine learning and deep learning tooling. Eero is a frequent speaker in ML events around the world.

    Twitter Linkedin
  • DEEP LEARNING FOR GOOD

  • 13:55
    Audrey Boguchwal

    Fighting AI Bias: How To Obtain Secure and High-Quality Training Data

    Audrey Boguchwal - Senior Product Manager - Samasource

    Down arrow blue

    Audrey is passionate about the intersection of artificial intelligence, human behavior and ethics. She is currently a senior product manager at Samasource, a provider of ground truth training data for the world’s leading AI technologies. As a trained anthropologist, she is fascinated by how people interact with technology and how we can use technology for good. As an anthropologist, she also loves museums, public transit, and interactive art.

    Audrey is a graduate of Harvard College, where she studied history. She holds an MA in Anthropology from Columbia and received her MBA from the University of North Carolina at Chapel Hill, Kenan-Flagler Business School. Previous work includes The Wall Street Journal, Huge Design and NBC Universal.

    Linkedin
  • 14:15

    PANEL: What Can Industry Do to Improve AI Computing & Energy Efficiency?

  • Rick Calle

    MODERATOR

    Rick Calle - Head of BD for AI Research - M12

    Down arrow blue

    Rick works at the intersection of emerging AI algorithms, hardware efficiency, and novel AI use cases. He leads AI business development for M12, Microsoft’s corporate venture group. Prior to Microsoft he worked at Qualcomm Research with the team who developed the Snapdragon AI Engine, and developed emerging business around always-on edge AI, Autonomous driving sensors, AI based cyber security and high performance computing at low energy. Having studied Electrical and Computer Engineering at Cornell, Rick passionately opines that deep learning is more than just Hot Dog Detectors – it’s really the next wave of Digital Signal Processing algorithms and Optimal Control. In his spare time with family Rick tracks global snowfalls, hoping for an epic snowboarding season!

    Twitter Linkedin
  • Abigail Wen

    PANELIST

    Abigail Wen - Sr. Director of Emerging AI Tech - Intel Corporation

    Down arrow blue

    Abigail Hing Wen is Sr. Director of Emerging AI Tech, Office of the CTO, Artificial Intelligence Products Group, Intel Corporation, and an Author. She serves as a business strategist and thought leader for next-gen AI emerging technologies and products, engaging with key AI customers and the startup ecosystem. She speaks frequently on AI and venture capital investing in national and international forums, most recently in Beijing, Berlin and Brussels. Her writings include pieces on AI and Privacy (Fortune), AI and Bias (Forbes) and Recent Breakthroughs in AI (Forbes). Abigail is also the New York Times best selling author of the novel Loveboat, Taipei, and has been featured in print and television on Bloomberg, Entertainment Weekly, the World Journal and South China Morning Post, among others. Her debut novel, a romantic comedy addressing issues of identity, immigration and leadership, sold in a six-house auction. Abigail is the Co-Chair of the Partnership on AI’s Expert Group for Fairness, Transparency and Accountability and sits on their Transparency Steering Committee. Previously, Abigail partnered closely with Silicon Valley investors as legal lead for Intel Capital’s AI investments and strategic transactions. She has worked with more than a hundred startups from incorporation to IPO or acquisition. Exemplary transactions include Intel’s $4.1B investment in ASML and $740M in Cloudera. She serves as board observer for Two Bit Circus, a virtual reality entertainment company based in Los Angeles. Prior to joining Intel in 2012, Abigail advised clients on Wall Street and in DC with the corporate group of Sullivan & Cromwell LLP, clerked for the US Court of Appeals for the DC Circuit and worked on tech and innovation policy for the Senate Judiciary Committee, Subcommittee on Antitrust, Business Rights and Competition. Abigail holds a BA from Harvard in Government and International Relations and JD from Columbia Law School.

    Twitter Linkedin
  • Kushagra Vaid

    PANELIST

    Kushagra Vaid - GM and Distinguished Engineer - Microsoft

    Down arrow blue

    Kushagra Vaid is GM and Distinguished Engineer in Microsoft’s Azure Division, and is responsible for the architecture and design for cloud datacenter infrastructure hosting Azure's global scale services. Kushagra has been instrumental in driving Microsoft’s success as a leading Hyperscale public cloud operator and is a recognized industry leader on infrastructure innovation and Open Source hardware. Prior to joining Microsoft in 2007, Kushagra was a Principal Engineer at Intel where his responsibilities included technology strategy, architecture and design for Intel’s Enterprise and Cloud Platforms. Kushagra has published over 25+ research papers in international conferences, and is also the holder of 30+ patents in computer architecture and systems design. Kushagra is a strategic advisor to the Global Semiconductor Alliance (GSA) and several infrastructure startups, and is also a sought after keynote speaker at top tier conferences and industry events.

    Twitter Linkedin
  • Andrew Feldman

    PANELIST

    Andrew Feldman - Founder & CEO - Cerebras

    Down arrow blue

    Andrew Feldman is founder and CEO of Cerebras Systems, a startup dedicated to accelerating Artificial intelligence (AI) compute. Cerebras is a team of pioneering computer architects, computer scientists, deep learning researchers, and engineers of all types who have come together to build a new class of computer optimized for AI work. Prior to Cerebras, Andrew was founder and CEO of SeaMicro. SeaMicro, acquired by AMD for $355 million, was the pioneer in energy efficient computation. SeaMicro invented the microserver category and changed the trajectory of the server industry by creating a new class of the high density, energy efficient servers. Prior to co-founding SeaMicro, Andrew was Vice President of Marketing and Product Management at Force10 Networks (acquired by Dell for $800 Million) and before that was Vice President of Corporate Marketing and Corporate Development for Riverstone Networks (NASDAQ: RSTN) from inception through IPO.

    Twitter Linkedin
  • 15:00

    END OF SUMMIT

  • 15:00

    FAREWELL NETWORKING MIXER

  • Day 1 10:25

    Introduction to Reinforcement Learning

    Lex Fridman - Researcher - MIT

    Down arrow blue

    An Introduction to Reinforcement Learning

    Lex Fridman is a researcher at MIT, working on deep learning approaches in the context of semi-autonomous vehicles, human sensing, personal robotics, and more generally human-centered artificial intelligence systems. He is particularly interested in understanding human behavior in the context of human-robot collaboration, and engineering learning-based methods that enrich that collaboration. Before joining MIT, Lex was at Google working on machine learning for large-scale behavior-based authentication.

    Twitter Linkedin
  • Day 1 11:10

    Muppets and Transformers: The New Stars of NLP

    Joel Grus - Principal Engineer - Capital Group

    Down arrow blue

    Muppets and Transformers: The New Stars of NLP

    The last few years have seen huge progress in NLP. Transformers have become a fundamental building block for impressive new NLP models. ELMo, BERT, and their descendants have achieved new state-of-the-art results on a wide variety of tasks. In this talk I'll give some history of these "new stars" of NLP, explain how they work, compare them to their predecessors, and discuss how you can apply them to your own problems.

    Joel Grus is Principal Engineer at Capital Group, where he oversees the development and deployment of machine learning systems. Previously he was a research engineer at the Allen Institute for Artificial Intelligence, where he helped develop AllenNLP, a deep learning library for NLP researchers. Before that he worked as a software engineer at Google and a data scientist at a variety of startups. He is the author of the beloved book Data Science from Scratch: First Principles with Python, the beloved blog post "Fizz Buzz in Tensorflow", and the polarizing JupyterCon talk "I Don't Like Notebooks". You can find him on Twitter @joelgrus

    Twitter Linkedin
  • Day 1 12:45

    Lunch & Learn

    Join the Speakers for Lunch - - Roundtable Discussions during Lunch

  • Day 1 14:25

    How Can AI Aid Digital Transformation – Mesh Twin Learning?

    Maciej Mazur - Chief Data Scientist - PGS Software

    Down arrow blue

    Fraud Detection in 2020 - Bad Guys Perspective

    AI is evolving rapidly these days, and together with it are our fraud detection systems. I want to show you what is the current state of the art approach to fraud detection, how are such systems implemented, and what are key differentiators to look at when choosing a solution for your business (AML, credit card frauds and insurance). Next we will focus on credit card frauds, but not from a payment provider or a bank perspective but a criminal. Learn more on what are the newest technology trends for card frauds, how bad guys build their infrastructure and how they cheat and manipulate your million dollars black boxes that are supposed to keep you safe.

    As Chief Data Scientist at PGS Software, Maciej is the technical lead of the data team and implements ML-based solutions for clients around the globe. In his 10 years of IT-experience, he’s worked for major players like Nokia and HPE, developing complex optimisation algorithms even before the term Data Science was coined.

    Twitter Linkedin
  • Day 1 16:00

    Hands-on Workshop: BERT based Conversational Q&A Platform for Querying a complex RDBMS with No Code

    Peter Relan - Chairman and CEO - Got-it.ai

    Down arrow blue

    Hands-on Workshop: BERT based Conversational Q&A Platform for Querying a complex RDBMS with No Code

    Most business and operations people in organizations want to ask questions of databases regularly. But they are limited by minimal schema understanding and SQL skills. In the field of AI, conversational agents like Rasa, Dialogflow, Lex, Watson, Luis are emerging as NLU-based dialog agents that hook into actions or custom fulfillment logic. Got It is unveiling the first AI product that creates a conversational interface to any custom database schema on MySQL or Google Big Query, using Rasa or Dialog Flow. Got It’s No Code approach automates the discovery and addition of new intents/slots and actions, based on incoming user questions and knowledge of the database schema. Thus, the end-end system adapts itself to an evolving schema and user questions until it can answer virtually any question. Got It supports full sentence NLP for chat based UIs, and search keyword NLP for Analytics UIs to dynamically query a database, without custom fulfillment logic, by utilizing a proprietary DNN.

    This workshop provides a hands-on session demonstrating how quick the set up is for the product to start retrieving data from a sophisticated retail industry database schema, for both business analytics as well as for customer service use cases.

    Peter Relan is the founding investor and chairman of breakthrough companies, including Discord (300M users), Epic! (95% of US elementary schools) and Got-it.ai (AI+Human Intelligence for Saas and Paas products). Formerly a Hewlett Packard Resident Fellow at Stanford University, and a senior Oracle executive, Peter is working with the Got It team on driving user and business productivity higher by 10X, applying Google BERT and transfer learning to real business databases with minimal training data sets, that allow users to program queries and analytics tools with no technical skills.

    Twitter Linkedin
  • Day 2 10:30

    Panel & Networking

    Investing in Startups: Hear from the Investors - - Panel & Connect

    Down arrow blue

    Session takeaways: 1) What are the short, medium and long-term challenges in investing in AI to solve challenges in business & society? 2) What are the main success factors for AI startups? 3) What are the challenges from a VC perspective?

  • Day 2 11:20

    Ludwig, a Code-Free Deep Learning Toolbox

    Piero Molino, Uber AI - Sr. Research Scientist & Co-Founder - Uber AI

    Down arrow blue

    Ludwig, a Code-Free Deep Learning Toolbox

    The talk will introduce Ludwig, a deep learning toolbox that allows to train models and to use them for prediction without the need to write code. It is unique in its ability to help make deep learning easier to understand for non-experts and enable faster model improvement iteration cycles for experienced machine learning developers and researchers alike. By using Ludwig, experts and researchers can simplify the prototyping process and streamline data processing so that they can focus on developing deep learning architectures.

    Piero Molino is a Senior Research Scientist at Uber AI with focus on machine learning for language and dialogue. Piero completed a PhD on Question Answering at the University of Bari, Italy. Founded QuestionCube, a startup that built a framework for semantic search and QA. Worked for Yahoo Labs in Barcelona on learning to rank, IBM Watson in New York on natural language processing with deep learning and then joined Geometric Intelligence, where he worked on grounded language understanding. After Uber acquired Geometric Intelligence, he became one of the founding members of Uber AI Labs. He currently leads the development of Ludwig, a code-free deep learning framework.

    Twitter Linkedin
  • Day 2 11:50

    Building a Conversational Experience in Minutes with Samsung’s Bixby

    Adam Cheyer - Co-Founder and VP Engineering/VP of R&D - Viv Labs/Samsung

    Down arrow blue

    Building a Conversational Experience in Minutes with Samsung’s Bixby

    For decades, the relationship between developer and computer was simple: the human told the machine what to do. Next came machine learning systems, where the machine was in charge of computing the functional logic behind developer-supplied examples, typically in a form that humans couldn't even understand. Now we are entering a new age of software development, where humans and machines work collaboratively together, each doing what they do best. The Developer describes the "what" -- objects, actions, goals -- and the machine produces the "how", writing the code that satisfied each user's request by interweaving developer-provided components. The result is a system that is easier to create and maintain, while providing an end-user experience that is more intelligent and adaptable to users' individual needs. In this talk, we will show concrete examples of this software trend using a next-generation conversational assistant named Bixby. We will supply you with a freely downloadable development environment so that you can give this a try yourself, and teach you how to build a conversational experience in minutes, to start monetizing your content and services through a new channel that will be backed by more than a billion devices in just a few years.

    Adam Cheyer is co-Founder and VP Engineering of Viv Labs, and after acquisition in 2016, a VP of R&D at Samsung. Previously, Mr. Cheyer was co-Founder and VP Engineering at Siri, Inc. In 2010, Siri was acquired by Apple, where he became a Director of Engineering in the iPhone/iOS group. Adam is also a Founding Member and Advisor to Change.org, the premier social network for positive social change, and a co-Founder of Sentient Technologies. Mr. Cheyer is an author of more than 60 publications and 27 issued patents.

    Linkedin
  • Day 2 14:00

    Panel & Q&A

    Ethics in AI: Panel, Q&A & Drop-In - - Hear from Experts in Ethics and Ask your Questions

This website uses cookies to ensure you get the best experience. Learn more