New method uses Probabilistic Circuits for improved out-of-distribution detection

Researchers have developed a new method for detecting out-of-distribution (OOD) data using Probabilistic Circuits (PCs). This approach, termed Hierarchical Likelihood Vector (HLV) and Hierarchical Likelihood Distance (HLD), leverages the internal hierarchical structure of PCs rather than just the root likelihood. The HLD metric compares probability distributions by examining the expectations of their HLVs, offering an integral probability metric. This method allows the trained PC itself to represent the in-distribution data, eliminating the need for held-out data during deployment and enabling exact computation of OOD detection metrics. AI

IMPACT Enhances the ability of AI models to identify unfamiliar or anomalous data, crucial for robust deployment in real-world scenarios.

RANK_REASON Academic paper detailing a new method for out-of-distribution detection. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Hierarchical Likelihood Distance
• Hierarchical Likelihood Vector
• MNIST database
• Probabilistic Circuits for Autonomous Learning: A Simulation Study

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →