Simulation is a crucial tool for accelerating the development of autonomous vehicles. Making simulation realistic requires models of the human road users who interact with such cars. Such models can be obtained by applying learning from demonstration (LfD) to trajectories observed by cars already on the road. However, existing LfD methods are typically insufficient, yielding policies that frequently collide or drive off the road. To address this problem, we propose Symphony, which greatly improves realism by combining conventional policies with a parallel beam search. The beam search refines these policies on the fly by pruning branches that are unfavourably evaluated by a discriminator. However, it can also harm diversity, i.e., how well the agents cover the entire distribution of realistic behaviour, as pruning can encourage mode collapse. Symphony addresses this issue with a hierarchical approach, factoring agent behaviour into goal generation and goal conditioning. The use of such goals ensures that agent diversity neither disappears during adversarial training nor is pruned away by the beam search. Experiments on both proprietary and open Waymo datasets confirm that Symphony agents learn more realistic and diverse behaviour than several baselines.

We demonstrate the first large-scale application of model-based generative adversarial imitation learning (MGAIL) to the task of dense urban self-driving. We augment standard MGAIL using a hierarchical model to enable generalization to arbitrary goal routes, and measure performance using a closed-loop evaluation framework with simulated interactive agents. We train policies from expert trajectories collected from real vehicles driving over 100,000 miles in San Francisco, and demonstrate a steerable policy that can navigate robustly even in a zero-shot setting, generalizing to synthetic scenarios with novel goals that never occurred in real-world driving. We also demonstrate the importance of mixing closed-loop MGAIL losses with open-loop behavior cloning losses, and show our best policy approaches the performance of the expert. We evaluate our imitative model in both average and challenging scenarios, and show how it can serve as a useful prior to plan successful trajectories.

We develop a cyclical blockwise coordinate descent algorithm for the multi-task Lasso that efficiently solves problems with thousands of features and tasks. The main result shows that a closed-form Winsorization operator can be obtained for the sup-norm penalized least squares regression. This allows the algorithm to find solutions to very large-scale problems far more efficiently than existing methods. This result complements the pioneering work of Friedman, et al. (2007) for the single-task Lasso. As a case study, we use the multi-task Lasso as a variable selector to discover a semantic basis for predicting human neural activation. The learned solution outperforms the standard basis for this task on the majority of test participants, while requiring far fewer assumptions about cognitive neuroscience. We demonstrate how this learned basis can yield insights into how the brain represents the meanings of words.

We consider the problem of zero-shot learning, where the goal is to learn a classifier $f: X \rightarrow Y$ that must predict novel values of $Y$ that were omitted from the training set. To achieve this, we define the notion of a semantic output code classifier (SOC) which utilizes a knowledge base of semantic properties of $Y$ to extrapolate to novel classes. We provide a formalism for this type of classifier and study its theoretical properties in a PAC framework, showing conditions under which the classifier can accurately predict novel classes. As a case study, we build a SOC classifier for a neural decoding task and show that it can often predict words that people are thinking about from functional magnetic resonance images (fMRI) of their neural activity, even without training examples for those words.

We provide a theoretical analysis of the chance accuracies of large collections of classifiers. We show that on problems with small numbers of examples, some classifier can perform well by random chance, and we derive a theorem to explicitly calculate this accuracy. We use this theorem to provide a principled feature selection criterion for sparse, high-dimensional problems. We evaluate this method on microarray and fMRI datasets and show that it performs very close to the optimal accuracy obtained from an oracle. We also show that on the fMRI dataset this technique chooses relevant features successfully while another state-of-the-art method, the False Discovery Rate (FDR), completely fails at standard significance levels.