# Free-Lunch Saliency via Attention in Atari Agents

arXiv: https://arxiv.org/abs/1908.02511 · HTML: https://mlsanity.com/paper/1908.02511/

## Key facts

- Cited by 12 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 7 Aug 2019.

## Built on by (most cited first, 12 of 12)

- [Explainable Reinforcement Learning via a Causal World Model](https://mlsanity.com/paper/2305.02749/) (2023)
- [Selective Eye-gaze Augmentation To Enhance Imitation Learning In Atari Games](https://mlsanity.com/paper/2012.03145/) (2020)
- [Temporal-Spatial Causal Interpretations for Vision-Based Reinforcement Learning](https://mlsanity.com/paper/2112.03020/) (2021)
- [Leveraging Reward Consistency for Interpretable Feature Discovery in Reinforcement Learning](https://mlsanity.com/paper/2309.01458/) (2023)
- [AEGIS: Human Attention-based Explainable Guidance for Intelligent Vehicle Systems](https://mlsanity.com/paper/2504.05950/) (2025)
- [An Approach to Partial Observability in Games: Learning to Both Act and Observe](https://mlsanity.com/paper/2108.05701/) (2021)
- [Revealing Human Attention Patterns from Gameplay Analysis for Reinforcement Learning](https://mlsanity.com/paper/2504.11118/) (2025)
- [The Atari Data Scraper](https://mlsanity.com/paper/2104.04893/) (2021)
- [MSVIPER: Improved Policy Distillation for Reinforcement-Learning-Based Robot Navigation](https://mlsanity.com/paper/2209.09079/) (2022)
- [Compositional Concept-Based Neuron-Level Interpretability for Deep Reinforcement Learning](https://mlsanity.com/paper/2502.00684/) (2025)
- [Attention Trajectories as a Diagnostic Axis for Deep Reinforcement Learning](https://mlsanity.com/paper/2511.20591/) (2025)
- [A Differentiable Atari VCS:A Complex, Fully Known Ground Truth for Explainable AI](https://mlsanity.com/paper/2606.22447/) (2026)

## Abstract (arXiv)

We propose a new approach to visualize saliency maps for deep neural network models and apply it to deep reinforcement learning agents trained on Atari environments. Our method adds an attention module that we call FLS (Free Lunch Saliency) to the feature extractor from an established baseline (Mnih et al., 2015). This addition results in a trainable model that can produce saliency maps, i.e., visualizations of the importance of different parts of the input for the agent's current decision making. We show experimentally that a network with an FLS module exhibits performance similar to the baseline (i.e., it is "free", with no performance cost) and can be used as a drop-in replacement for reinforcement learning agents. We also design another feature extractor that scores slightly lower but provides higher-fidelity visualizations. In addition to attained scores, we report saliency metrics evaluated on the Atari-HEAD dataset of human gameplay.