CRUMB: New Method Speeds Up Prior-Fitted Network Inference for Large Datasets
Researchers introduced CRUMB, a technique that makes prior-fitted networks (PFN) faster when working with large training datasets by selecting relevant subsets of training data rather than using all of it. Prior-fitted networks are foundation models for tabular data that perform in-context learning, but their quadratic computational scaling becomes impractical with large datasets. The method is significant because it enables PFNs to scale to real-world applications without requiring model retraining.
CRUMB (Clustered Retrieval Using Minimised-MMD Batching) is a three-stage inference wrapper designed to address computational bottlenecks in prior-fitted networks. The method works by clustering test queries, selecting a small distributionally matched subset of training data for each cluster using maximum mean discrepancy (MMD) minimization, and then running exact PFN inference on each reduced batch. The approach is architecture-agnostic and requires no model retraining, making it practical to apply to existing systems. Testing across 51 datasets and three different PFN architectures (TabPFNv2, TabICLv1, TabICLv2) showed CRUMB outperformed comparable context selection strategies. The method also demonstrated resilience to covariate drift, as the MMD-minimization step naturally aligns training context distributions with test batch distributions.
What's missing
The paper does not discuss computational overhead of the clustering and MMD-minimization stages themselves, or provide wall-clock time comparisons for the full three-stage pipeline versus inference time savings. Limitations regarding the method's performance on datasets with very high-dimensional features or extreme class imbalance are not addressed in the abstract.
What different sources said
- arXiv cs.AICenter
CRUMB: Efficient Prior Fitted Network Inference via Distributionally Matched Context Batching
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.