skip to main content

Title: Explain Your Move: Understanding Agent Actions Using Specific and Relevant Feature Attribution
As deep reinforcement learning (RL) is applied to more tasks, there is a need to visualize and understand the behavior of learned agents. Saliency maps explain agent behavior by highlighting the features of the input state that are most relevant for the agent in taking an action. Existing perturbation-based approaches to compute saliency often highlight regions of the input that are not relevant to the action taken by the agent. Our proposed approach, SARFA (Specific and Relevant Feature Attribution), generates more focused saliency maps by balancing two aspects (specificity and relevance) that capture different desiderata of saliency. The first captures the impact of perturbation on the relative expected reward of the action to be explained. The second downweighs irrelevant features that alter the relative expected rewards of actions other than the action to be explained. We compare SARFA with existing approaches on agents trained to play board games (Chess and Go) and Atari games (Breakout, Pong and Space Invaders). We show through illustrative examples (Chess, Atari, Go), human studies (Chess), and automated evaluation methods (Chess) that SARFA generates saliency maps that are more interpretable for humans than existing approaches.
Authors:
; ; ; ; ; ;
Award ID(s):
1756023
Publication Date:
NSF-PAR ID:
10166401
Journal Name:
International Conference on Learning Representations (ICLR)
Sponsoring Org:
National Science Foundation
More Like this
  1. Recent reinforcement learning (RL) approaches have shown strong performance in complex domains such as Atari games, but are often highly sample inefficient. A common approach to reduce interaction time with the environment is to use reward shaping, which involves carefully designing reward functions that provide the agent intermediate rewards for progress towards the goal. However, designing appropriate shaping rewards is known to be difficult as well as time-consuming. In this work, we address this problem by using natural language instructions to perform reward shaping. We propose the LanguagE-Action Reward Network (LEARN), a framework that maps free-form natural language instructions to intermediate rewards based on actions taken by the agent. These intermediate language-based rewards can seamlessly be integrated into any standard reinforcement learning algorithm. We experiment with Montezuma’s Revenge from the Atari Learning Environment, a popular benchmark in RL. Our experiments on a diverse set of 15 tasks demonstrate that, for the same number of interactions with the environment, language-based rewards lead to successful completion of the task 60 % more often on average, compared to learning without language.
  2. In imperfect-information games, subgame solving is significantly more challenging than in perfect-information games, but in the last few years, such techniques have been developed. They were the key ingredient to the milestone of superhuman play in no-limit Texas hold'em poker. Current subgame-solving techniques analyze the entire common-knowledge closure of the player's current information set, that is, the smallest set of nodes within which it is common knowledge that the current node lies. However, this set is too large to handle in many games. We introduce an approach that overcomes this obstacle, by instead working with only low-order knowledge. Our approach allows an agent, upon arriving at an infoset, to basically prune any node that is no longer reachable, thereby massively reducing the game tree size relative to the common-knowledge subgame. We prove that, as is, our approach can increase exploitability compared to the blueprint strategy. However, we develop three avenues by which safety can be guaranteed. First, safety is guaranteed if the results of subgame solves are incorporated back into the blueprint. Second, we provide a method where safety is achieved by limiting the infosets at which subgame solving is performed. Third, we prove that our approach, when applied atmore »every infoset reached during play, achieves a weaker notion of equilibrium, which we coin affine equilibrium, and which may be of independent interest. We show that affine equilibria cannot be exploited by any Nash strategy of the opponent, so an opponent who wishes to exploit must open herself to counter-exploitation. Even without the safety-guaranteeing additions, experiments on medium-sized games show that our approach always reduced exploitability even when applied at every infoset, and a depth-limited version of it led to--to our knowledge--the first strong AI for the massive challenge problem dark chess.« less
  3. As machine learning methods see greater adoption and implementation in high-stakes applications such as medical image diagnosis, the need for model interpretability and explanation has become more critical. Classical approaches that assess feature importance (e.g., saliency maps) do not explain how and why a particular region of an image is relevant to the prediction. We propose a method that explains the outcome of a classification black-box by gradually exaggerating the semantic effect of a given class. Given a query input to a classifier, our method produces a progressive set of plausible variations of that query, which gradually changes the posterior probability from its original class to its negation. These counter-factually generated samples preserve features unrelated to the classification decision, such that a user can employ our method as a “tuning knob” to traverse a data manifold while crossing the decision boundary. Our method is model agnostic and only requires the output value and gradient of the predictor with respect to its input.
  4. The conventional machine learning (ML) and deep learning (DL) methods use large amount of data to construct desirable prediction models in a central fusion center for recognizing human activities. However, such model training encounters high communication costs and leads to privacy infringement. To address the issues of high communication overhead and privacy leakage, we employed a widely popular distributed ML technique called Federated Learning (FL) that generates a global model for predicting human activities by combining participated agents’ local knowledge. The state-of-the-art FL model fails to maintain acceptable accuracy when there is a large number of unreliable agents who can infuse false model, or, resource-constrained agents that fails to perform an assigned computational task within a given time window. We developed an FL model for predicting human activities by monitoring agent’s contributions towards model convergence and avoiding the unreliable and resource-constrained agents from training. We assign a score to each client when it joins in a network and the score is updated based on the agent’s activities during training. We consider three mobile robots as FL clients that are heterogeneous in terms of their resources such as processing capability, memory, bandwidth, battery-life and data volume. We consider heterogeneous mobile robotsmore »for understanding the effects of real-world FL setting in presence of resource-constrained agents. We consider an agent unreliable if it repeatedly gives slow response or infuses incorrect models during training. By disregarding the unreliable and weak agents, we carry-out the local training of the FL process on selected agents. If somehow, a weak agent is selected and started showing straggler issues, we leverage asynchronous FL mechanism that aggregate the local models whenever it receives a model update from the agents. Asynchronous FL eliminates the issue of waiting for a long time to receive model updates from the weak agents. To the end, we simulate how we can track the behavior of the agents through a reward-punishment scheme and present the influence of unreliable and resource-constrained agents in the FL process. We found that FL performs slightly worse than centralized models, if there is no unreliable and resource-constrained agent. However, as the number of malicious and straggler clients increases, our proposed model performs more effectively by identifying and avoiding those agents while recognizing human activities as compared to the stateof-the-art FL and ML approaches.« less
  5. Recent advances in off-policy deep reinforcement learning (RL) have led to impressive success in complex tasks from visual observations. Experience replay improves sample-efficiency by reusing experiences from the past, and convolutional neural networks (CNNs) process high-dimensional inputs effectively. However, such techniques demand high memory and computational bandwidth. In this paper, we present Stored Embeddings for Efficient Reinforcement Learning (SEER), a simple modification of existing off-policy RL methods, to address these computational and memory requirements. To reduce the computational overhead of gradient updates in CNNs, we freeze the lower layers of CNN encoders early in training due to early convergence of their parameters. Additionally, we reduce memory requirements by storing the low-dimensional latent vectors for experience replay instead of high-dimensional images, enabling an adaptive increase in the replay buffer capacity, a useful technique in constrained-memory settings. In our experiments, we show that SEER does not degrade the performance of RL agents while significantly saving computation and memory across a diverse set of DeepMind Control environments and Atari games. Finally, we show that SEER is useful for computation-efficient transfer learning in RL because lower layers of CNNs extract generalizable features, which can be used for different tasks and domains.