Search for: All records

Although deep Reinforcement Learning (RL) has proven successful in a wide range of tasks, one challenge it faces is interpretability when applied to real-world problems. Saliency maps are frequently used to provide interpretability for deep neural networks. However, in the RL domain, existing saliency map approaches are either computationally expensive and thus cannot satisfy the real-time requirement of real-world scenarios or cannot produce interpretable saliency maps for RL policies. In this work, we propose an approach of Distillation with selective Input Gradient Regularization (DIGR) which uses policy distillation and input gradient regularization to produce new policies that achieve both high interpretability and computation efficiency in generating saliency maps. Our approach is also found to improve the robustness of RL policies to multiple adversarial attacks. We conduct experiments on three tasks, MiniGrid (Fetch Object), Atari (Breakout) and CARLA Autonomous Driving, to demonstrate the importance and effectiveness of our approach. 

Humans and other animals have a remarkable capacity to translate their position from one spatial frame of reference to another. The ability to seamlessly move between top-down and first-person views is important for navigation, memory formation, and other cognitive tasks. Evidence suggests that the medial temporal lobe and other cortical regions contribute to this function. To understand how a neural system might carry out these computations, we used variational autoencoders (VAEs) to reconstruct the first-person view from the top-down view of a robot simulation, and vice versa. Many latent variables in the VAEs had similar responses to those seen in neuron recordings, including location-specific activity, head direction tuning, and encoding of distance to local objects. Place-specific responses were prominent when reconstructing a first-person view from a top-down view, but head direction–specific responses were prominent when reconstructing a top-down view from a first-person view. In both cases, the model could recover from perturbations without retraining, but rather through remapping. These results could advance our understanding of how brain regions support viewpoint linkages and transformations.