Abstract:Despite advances in language modelling, distributional methods that build semantic representations from co-occurrences fail to discriminate between plausible and implausible events. In this work, we investigate how plausibility prediction can be improved by injecting latent knowledge prompted from large language models using parameter-efficient fine-tuning. We train 12 task adapters to learn various physical properties and association measures and perform adapter fusion to compose latent semantic knowledge from each task on top of pre-trained AlBERT embeddings. We automate auxiliary task data generation, which enables us to scale our approach and fine-tune our learned representations across two plausibility datasets. Our code is available at https://github.com/Jacob-Chmura/plausibility-vaccine.
Abstract:Sparse reward environments are known to be challenging for reinforcement learning agents. In such environments, efficient and scalable exploration is crucial. Exploration is a means by which an agent gains information about the environment. We expand on this topic and propose a new intrinsic reward that systemically quantifies exploratory behavior and promotes state coverage by maximizing the information content of a trajectory taken by an agent. We compare our method to alternative exploration based intrinsic reward techniques, namely Curiosity Driven Learning and Random Network Distillation. We show that our information theoretic reward induces efficient exploration and outperforms in various games, including Montezuma Revenge, a known difficult task for reinforcement learning. Finally, we propose an extension that maximizes information content in a discretely compressed latent space which boosts sample efficiency and generalizes to continuous state spaces.