Researchers have developed a new method to audit neural network decisions by decomposing action scores into a weighted sum of training case returns. This approach, grounded in Case-Based Decision Theory (CBDT), allows for tracing scores back to specific training data and measuring the coherence of actions. The method was tested on synthetic CBDT, PJM, Adult Income, and Default Credit tasks, demonstrating its ability to recover case-level preference structures and provide robust audit signals. AI
IMPACT Enables more transparent and auditable AI systems in high-stakes applications.
RANK_REASON The cluster contains a research paper detailing a new theoretical framework for auditing neural networks.
- Adult Income
- alphaXiv
- arXiv
- Case-Based Decision Theory
- CatalyzeX
- DagsHub
- Default Credit
- Gotit.pub
- Gram geometry
- Hugging Face
- Ordinary Least Squares
- Neural Networks
- PJM Interconnection
- ScienceCast
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
