Advances in Interpretable Machine Learning

The field of machine learning is moving towards increased interpretability, with a focus on developing methods that provide insights into model decisions. Recent research has introduced new techniques for comparing and visualizing learned representations, enabling more direct and interpretable model comparisons. There is also a growing interest in concept-based models, which aim to extract human-understandable concepts from data and provide transparent decision-making. These models have shown promising results in various applications, including image classification and medical diagnosis. Furthermore, researchers are exploring new approaches to improve the reliability and generalization of concept-based models, such as incorporating object-centric information and using generative models to learn latent cost variables. Noteworthy papers include: Representational Difference Explanations, which proposes a method for discovering and visualizing differences between learned representations. Object Centric Concept Bottlenecks, which introduces a framework that combines concept-based models with pre-trained object-centric foundation models. Through a Steerable Lens, which proposes a novel framework for visualizing the implicit path between classes in a neural network.

Sources

Representational Difference Explanations

Object Centric Concept Bottlenecks

Towards Better Generalization and Interpretability in Unsupervised Concept-Based Models

Through a Steerable Lens: Magnifying Neural Network Interpretability via Phase-Based Extrapolation

Learning Monotonic Probabilities with a Generative Cost Model

A Comprehensive Survey on the Risks and Limitations of Concept-based Models

Visualizing and Controlling Cortical Responses Using Voxel-Weighted Activation Maximization

Interpretable Few-Shot Image Classification via Prototypical Concept-Guided Mixture of LoRA Experts

Aligning Multimodal Representations through an Information Bottleneck

There Was Never a Bottleneck in Concept Bottleneck Models

Towards Reasonable Concept Bottleneck Models

Stable Vision Concept Transformers for Medical Diagnosis

Built with on top of