Zero-shot Concept Bottleneck Models

📰 ArXiv cs.AI

Zero-shot concept bottleneck models enable interpretable neural networks without requiring target task training

advanced Published 6 Apr 2026
Action Steps
  1. Identify high-level semantic concepts relevant to the task
  2. Use pre-trained language models or knowledge graphs to learn input-to-concept mappings
  3. Learn concept-to-label mappings using zero-shot learning techniques
  4. Evaluate the model's performance on the target task without requiring target task training
Who Needs to Know This

ML researchers and engineers working on interpretable models can benefit from this approach as it reduces the need for extensive training data and resources, while also providing insights into the decision-making process of the model

Key Insight

💡 Zero-shot concept bottleneck models can learn to predict labels without requiring target task training, making them more efficient and interpretable

Share This
💡 Zero-shot concept bottleneck models enable interpretable neural networks without target task training!

Key Takeaways

Zero-shot concept bottleneck models enable interpretable neural networks without requiring target task training

Full Article

Title: Zero-shot Concept Bottleneck Models

Abstract:
arXiv:2502.09018v2 Announce Type: replace-cross Abstract: Concept bottleneck models (CBMs) are inherently interpretable and intervenable neural network models, which explain their final label prediction by the intermediate prediction of high-level semantic concepts. However, they require target task training to learn input-to-concept and concept-to-label mappings, incurring target dataset collections and training resources. In this paper, we present zero-shot concept bottleneck models (Z-CBMs),
Read full paper → ← Back to Reads