Interpretable Control

Interpretable control focuses on developing control systems whose decision-making processes are transparent and understandable, addressing the "black box" problem inherent in many machine learning approaches. Current research emphasizes the use of sparse, model-based reinforcement learning algorithms, differentiable decision trees, and techniques like sparse identification of nonlinear dynamics (SINDy) to create more interpretable policies, often alongside methods for disentangling latent spaces in generative models. This pursuit is crucial for building trust in AI systems, particularly in safety-critical applications, and for facilitating the design of more robust and reliable controllers across diverse fields like energy management and engineering.

Papers

October 7, 2024

Synthesizing Interpretable Control Policies through Large Language Model Guided Search
Carlo Bosio, Mark W. Mueller
Dynamical System Large AI Model Complex Behavior Interpretable Control

August 15, 2024

$EvoAl^{2048}$
Bernhard J. Berger (1 and 2), Christina Plump (3), Rolf Drechsler (4 and 3) ((1) University of Rostock, Software Engineering Chair Rostock, Germany, (2) Hamburg University of Technology, Institute of Embedded Systems, Germany, (3) DFKI - Cyber-Physical Systems Bremen, Germany, (4) University of Bremen, Departments of Mathematics and Computer Science)
AI System Policy Design Model Based Optimization Interpretable Policy Interpretable Control

March 22, 2024

Parametric PDE Control with Deep Reinforcement Learning and Differentiable L0-Sparse Polynomial Policies
Nicolò Botteghi, Urban Fasel
Deep Reinforcement Learning Optimal Control Linear Policy PDE Control Interpretable Control

March 18, 2024

Distill2Explain: Differentiable decision trees for explainable reinforcement learning in energy application controllers
Gargya Gokhale, Seyed Soroush Karimi Madahi, Bert Claessens, Chris Develder
Explainable Reinforcement Learning Data Driven Control Energy Flexibility Demand Side Interpretable Control Energy Efficient Control Differentiable Decision Tree

March 14, 2024

SINDy-RL: Interpretable and Efficient Model-Based Reinforcement Learning
Nicholas Zolman, Urban Fasel, J. Nathan Kutz, Steven L. Brunton
Deep Reinforcement Learning Nonlinear Dynamic Interpretable Way Neural Network Policy Weak Sindy Interpretable Control

October 31, 2023

EXTRACT: Explainable Transparent Control of Bias in Embeddings
Zhijin Guo, Zhaozhen Xu, Martha Lewis, Nello Cristianini
Knowledge Graph Jina Embeddings Absolute Stance Bias Knowledge Graph Embeddings Graph Embeddings KG Completion Interpretable Control

August 25, 2023

Pretty darn good control: when are approximate solutions better than approximate models
Felipe Montealegre-Mora, Marcus Lapeyrolerie, Melissa Chapman, Abigail G. Keller, Carl Boettiger
Deep Reinforcement Learning Optimal Control Policy Design Approximate Model Interpretable Control Approximate Solution

January 24, 2022

Which Style Makes Me Attractive? Interpretable Control Discovery and Counterfactual Explanation on StyleGAN
Bo Li, Qiulin Wang, Jiquan Pei, Yu Yang, Xiangyang Ji
Counterfactual Explanation StyleGAN Latent Style Consistency Latent Subspace Interpretable Control

Interpretable Control

Papers

Synthesizing Interpretable Control Policies through Large Language Model Guided Search

$EvoAl^{2048}$

Parametric PDE Control with Deep Reinforcement Learning and Differentiable L0-Sparse Polynomial Policies

Distill2Explain: Differentiable decision trees for explainable reinforcement learning in energy application controllers

SINDy-RL: Interpretable and Efficient Model-Based Reinforcement Learning

EXTRACT: Explainable Transparent Control of Bias in Embeddings

Pretty darn good control: when are approximate solutions better than approximate models

Which Style Makes Me Attractive? Interpretable Control Discovery and Counterfactual Explanation on StyleGAN