Bayesian Concept Bottleneck Models with LLM Priors: A Novel Approach for Interpretable and Accurate Machine Learning
This paper introduces BC-LLM, a novel method for learning Concept Bottleneck Models (CBMs) that leverages Large Language Models (LLMs) within a Bayesian framework to iteratively discover and refine interpretable concepts from data, achieving both high accuracy and uncertainty quantification.
Stochastic Concept Bottleneck Models for Enhanced Interpretability and Intervention Effectiveness in Machine Learning
Stochastic Concept Bottleneck Models (SCBMs) improve the interpretability and intervention effectiveness of traditional Concept Bottleneck Models (CBMs) by explicitly modeling concept dependencies, leading to more accurate and efficient user interventions for enhanced model performance.
Inspecting and Controlling Information Leakage in Concept Bottleneck Models Using Decision Trees
This paper introduces a novel method using decision trees to inspect and control information leakage in Concept Bottleneck Models (CBMs), improving interpretability and reliability, especially when dealing with incomplete concept sets.
Editable Concept Bottleneck Models for Efficient Data and Concept Editing
This paper introduces Editable Concept Bottleneck Models (ECBMs), a novel approach to efficiently edit pre-trained Concept Bottleneck Models (CBMs) without expensive retraining, addressing challenges related to data privacy, mislabeling, and concept updates.
Improving Concept Bottleneck Model Efficacy through Concept Realignment
Concept Bottleneck Models can be significantly improved by leveraging concept relations to realign concept assignments after human interventions.
© 2024 by Linnk AI