model interpretability
The degree to which a human can understand the cause of a decision made by an AI model. Interpretability is crucial for trust, accountability, and debugging of AI systems.
- DATE-LM: Benchmarking Data Attribution Evaluation for Large Language Models
- Dataset Distillation for Pre-Trained Self-Supervised Vision Models
- Defining and Discovering Hyper-meta-paths for Heterogeneous Hypergraphs
- Demystifying Reasoning Dynamics with Mutual Information: Thinking Tokens are Information Peaks in LLM Reasoning
- Fairness-aware Anomaly Detection via Fair Projection
- Geometry of Decision Making in Language Models
- How to Auto-optimize Prompts for Domain Tasks? Adaptive Prompting and Reasoning through Evolutionary Domain Knowledge Adaptation
- One-Step is Enough: Sparse Autoencoders for Text-to-Image Diffusion Models
- Parameterized Synthetic Text Generation with SimpleStories
- Register and [CLS] tokens induce a decoupling of local and global features in large ViTs
- Self-Assembling Graph Perceptrons
- Sound Logical Explanations for Mean Aggregation Graph Neural Networks
- The Fragile Truth of Saliency: Improving LLM Input Attribution via Attention Bias Optimization