Joelle Pineau
Facebook & McGill University

Title: Data-Driven Dialogue Systems: Models, Algorithms, Evaluation, and Ethical Challenges

The use of dialogue systems as a medium for human-machine interaction is an increasingly prevalent paradigm. A growing number of dialogue systems use conversation strategies that are learned from large datasets. In this talk I will review several recent models and algorithms based on both discriminative and generative models, and discuss new results on the proper performance measures for such systems. Finally, I will highlight potential ethical issues that arise in dialogue systems research, including: biases, adversarial examples, privacy, and safety concerns.

Peter Bartlett
CS and Statistics, UC Berkeley 

Title: Some representation, optimization and generalization properties of deep networks

Deep neural networks have improved the state-of-the-art performance for prediction problems across an impressive range of application areas.  This talk describes some recent results in three directions.  First, we investigate the impact of depth on representational properties of deep residual networks, which compute near-identity maps at each layer, showing how their representational power improves with depth and that the functional optimization landscape has the desirable property that stationary points are optimal.  Second, we study the implications for optimization in deep linear networks, showing how the success of a family of gradient descent algorithms that regularize towards the identity function depends on a positivity condition of the regression function.  Third, we consider how the performance of deep networks on training data compares to their predictive accuracy, we demonstrate deviation bounds that scale with a certain "spectral complexity," and we compare the behavior of these bounds with the observed performance of these networks in practical problems.

[Joint work with Steve Evans, Dylan Foster, Dave Helmbold, Phil Long, and Matus Telgarsky.]

  Jennifer Listgarten
Center for Computational Biology & Berkeley AI Research, UC Berkeley

Title: Statistical and machine learning challenges from genetics to CRISPR gene editing 

Molecular biology, healthcare and medicine have been slowly morphing into large-scale, data driven sciences dependent on machine learning and applied statistics. Many of the same challenges from other domains are applicable here: causality vs association; covariate shift; hidden confounders; heterogenous target space; model validation; (multiple) hypothesis testing; feature engineering (owing to relatively small data sets). In this talk, I will go over domain-specific instantiations of some of these problems, along with proposed solutions. In particular, I will start by presenting modelling challenges in finding the genetic underpinnings of disease, which is important for screening, treatment, drug development. Assuming that we have uncovered genetic causes, genome editing—which is about deleting or changing parts of the genetic code—will one day let us fix the genome in a bespoke manner. Editing will also help researchers understand mechanisms of disease, enable precision medicine and drug development, to name just a few more important applications. I will close this talk by discussing how we have advanced CRISPR gene editing with machine learning. 

Sergey Levine
Berkeley AI Research Lab, UC Berkeley and Google

Title: What's Wrong with Meta-Learning (and how we can fix it)

Deep learning enables machines to perceive the world, interpret speech and text, recognize activities in video, and perform dozens of other open-world recognition tasks. However, artificial intelligence systems must be able not only to perceive and recognize, but also to act. To bring the power of deep learning into decision making and control, we must combine it with reinforcement learning or optimal control, which provide a mathematical framework for decision making. This combination, often termed deep reinforcement learning in the literature, has been applied to tasks from robotic control to game playing, but questions remain about its applicability to real-world problems. A major strength of deep learning is its scalability and practicality: the same algorithm that can be trained to recognize MNIST digits and recognize one of a thousand different object categories in ImageNet, with real images, and with minimal human effort. Can the same kind of real-world impact be attained with deep reinforcement learning? The usual concerns surround efficiency (will it ever be practical to train agents with deep RL in the real world?) and generalization (doesn't deep RL overfit catastrophically to the task at hand?). Some have proposed that model-based algorithms, that first learn to predict the future and then use these predictions to act, can mitigate some of these concerns, but these methods give rise to additional questions: can model-based RL ever attain the performance of model-free methods? Can model-based RL algorithms scale to high-dimensional (e.g., raw image) observations? And do model-based RL methods even have anything in common with model-free methods, or will we have to abandon everything we know about those and start over? In this talk, I will address these questions, and show that, not only can model-free RL algorithms already scale to complex real-world settings and achieve excellent generalization, but that model-based RL algorithms can also achieve comparable performance and scale to high-dimensional (e.g., raw image) observations, and conclude with a discussion of how model-free and model-based reinforcement learning algorithms might have more in common than we might at first think.