AI
ALICE: In-context, Zero-shot, Mutual Information Estimation
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 28/09/2026, 20:00:00 UTC
Briefing created: 01/10/2026, 02:00:37 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 15:18:24 UTC.
Briefings
ALICE is a foundation model designed to estimate mutual information from samples without per-distribution training, addressing limitations of current neural estimators. Trained on synthetic distributions, it serves as an in-context estimator for rectified-flow velocity fields, achieving competitive accuracy across biology, genetics, and neuroscience. The method uses a fixed identity integrating squared differences between joint and conditional fields, enabling zero-shot analysis across diverse scientific domains with varying data dimensions and sample sizes.
Key points
- ALICE removes per-distribution training by acting as an in-context estimator for rectified-flow velocity fields conditioned on unseen samples.
- The model achieves competitive estimation accuracy across biology, genetics, and neuroscience domains using a fixed identity integrating squared differences between joint and conditional fields.
Context and limitations
Validation is limited to a standard, challenging benchmark; results are from a single model trained on synthetic distributions, and specific performance metrics for each domain are not detailed.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 02:00:37 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeib3b5v45h747wpksuaepyly24zceluabyqhj4bcem3sqcit5sjw3e