Generalization

Past work on curriculum generation in RL has focused on training a teacher agent to generate tasks for a student agent that accelerate student learning and improve generalization. In this work, we create a mathematical framework that formalizes these concepts and subsumes prior work, taking inspiration from the psychological concept of the Zone of Proximal Development. We propose two new techniques based on rejection sampling and maximizing the student’s gradient norm that improve curriculum learning.

We analyze and improve upon PAIRED in the case of learning to generate challenging compositional tasks. We apply our improved algorithm to the complex task of training RL agents to navigate websites, and find that it is able to generating a challenging curriculum of novel sites. We achieve a 4x improvement over the strongest web navigation baselines, and deploy our model to navigate real-world websites..

PsiPhi-Learning learns successor representations for the policies of other agents and the ego agent, using a shared underlying state representation. Learning from other agents helps the agent take better actions at inference time, and learning from RL experience improves modeling of other agents.

Model-free RL agents fail to learn from experts present in multi-agent environments. By adding a model-based auxiliary loss, we induce social learning, which allows agents to learn how to learn from experts. When deployed to novel environments with new experts, they use social learning to determine how to solve the task, and generalize better than agents trained alone with RL or imitation learning.

PAIRED trains an agent to generate environments that maximize regret between a pair of learning agents. This creates feasible yet challenging environments, which exploit weaknesses in the agents to make them more robust. PAIRED significantly improves generalization to novel tasks.

My PhD Thesis spans both Social Reinforcement Learning and Affective Computing, investigating how affective and social intelligence can enhance machine learning algorithms, and how machine learning can enhance our ability to predict and understand human affective and social phenomena.

Predicting signals like stress and health depends on collecting noisy data from a number of modalities, e.g. smartphone data, or physiological data from a wrist-worn sensor. Our method can continue making accurate predictions even when a modality goes missing; for example, if the person forgets to wear their sensor.

Traditional, one-size-fits-all machine learning models fail to account for individual differences in predicting wellbeing outcomes like stress, mood, and health. Instead, we personalize models to the individual using multi-task learning (MTL), employing hierarchical Bayes, kernel-based and deep neural network MTL models to improve prediction accuracy by 13-23%.

Modeling measures like mood, stress, and health using a monolithic machine learning model leads to low prediction accuracy. Instead, we develop personalized regression models using multi-task learning and Gaussian Processes, leading to dramatic improvements in next-day predictions.

To combine supervised learning on data with reinforcement learning, we pre-train a supervised data prior, and penalize KL-divergence from this model using RL training. This enables effective learning of complex sequence-modeling problems for which we wish to match the data while optimizing external metrics like drug effectiveness. The approach produces compelling results in the disparate domains of music generation and drug discovery.