Addressing Concept Mislabeling in Concept Bottleneck Models Through Preference Optimization
This addresses a practical issue for improving the trustworthiness of AI systems by handling noisy concept labels, though it is incremental as it builds on existing methods like Direct Preference Optimization.
The paper tackles the problem of concept mislabeling in Concept Bottleneck Models, which can degrade performance by up to 25%, by introducing Concept Preference Optimization (CPO) to mitigate this impact and showing it consistently outperforms Binary Cross Entropy on real-world datasets.
Concept Bottleneck Models (CBMs) propose to enhance the trustworthiness of AI systems by constraining their decisions on a set of human-understandable concepts. However, CBMs typically assume that datasets contain accurate concept labels-an assumption often violated in practice, which we show can significantly degrade performance (by 25% in some cases). To address this, we introduce the Concept Preference Optimization (CPO) objective, a new loss function based on Direct Preference Optimization, which effectively mitigates the negative impact of concept mislabeling on CBM performance. We provide an analysis of key properties of the CPO objective, showing it directly optimizes for the concept's posterior distribution, and contrast it against Binary Cross Entropy (BCE), demonstrating that CPO is inherently less sensitive to concept noise. We empirically confirm our analysis by finding that CPO consistently outperforms BCE on three real-world datasets, both with and without added label noise. We make our code available on Github.