Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization

Sascha Marton, Tim Grams, Florian Vogt, Stefan Ludtke, Christian Bartelt +1 more
2/12/2026

No official code found. Best implementation: s-marton/SYMPOL (31 stars, MIT, last commit 2024-09-05).

Abstract

Reinforcement learning (RL) has seen significant success across various domains, but its adoption is often limited by the black-box nature of neural network policies, making them difficult to interpret. In contrast, symbolic policies allow representing decision-making strategies in a compact and interpretable way. However, learning symbolic policies directly within on-policy methods remains challenging. In this paper, we introduce SYMPOL, a novel method for SYMbolic tree-based on-POLicy RL. SYMPOL employs a tree-based model integrated with a policy gradient method, enabling the agent to learn and adapt its actions while maintaining a high level of interpretability. We evaluate SYMPOL on a set of benchmark RL tasks, demonstrating its superiority over alternative tree-based RL approaches in terms of performance and interpretability. Unlike existing methods, it enables gradient-based, end-to-end learning of interpretable, axis-aligned decision trees within standard on-policy RL algorithms. Therefore, SYMPOL can become the foundation for a new class of interpretable RL based on decision trees. Our implementation is available under: https://github.com/s-marton/sympol

View on arXivView PDF

Code Implementations(1)

s-marton/SYMPOLCommunity74%
MIT

(ICLR 2025) Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization

312Shell, PythonAug 16, 20242 years agoMIT
explainabilityexplainable-aiexplainable-mlinterpretabilityinterpretable-reinforcement-learning+3 more

Similar Papers

Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization

100% match1 repo312024

S-REINFORCE: A Neuro-Symbolic Policy Gradient Approach for Interpretable Reinforcement Learning

87% match2023

Optimizing Interpretable Decision Tree Policies for Reinforcement Learning

87% match1 repo102024

Efficient Symbolic Policy Learning with Differentiable Symbolic Expression

84% match1 repo32026

Interpretable and Editable Programmatic Tree Policies for Reinforcement Learning

83% match1 repo152024

Synthesizing Programmatic Policies with Actor-Critic Algorithms and ReLU Networks

83% match2023

Cite this paper

@article{marton2026mitigating,
  title  = {Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization},
  author = {Sascha Marton and Tim Grams and Florian Vogt and Stefan Ludtke and Christian Bartelt and Heiner Stuckenschmidt},
  year   = {2026},
  eprint = {2408.08761},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/2408.08761},
  journal = {ICLR 2024 2024}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2408.08761.svg)](https://codeofpaper.com/paper/2408.08761)

Discussion