Bergson: Open Source Library for Data Attribution in Machine Learning Released
Researchers have released Bergson, an open-source library designed to scale data attribution techniques to large language models and pre-training datasets. Data attribution is an interpretability field that explains model behavior by measuring the influence of training data, with applications in debugging models and dataset curation. The library addresses a gap in accessible tooling for data attribution research and includes the first open-source implementations of three leading methods: MAGIC, SOURCE, and TrackStar.
Bergson is a new open-source library that provides scalable implementations of data attribution techniques for machine learning models. Data attribution aims to explain how training data influences model behavior, enabling researchers to debug undesirable outputs and improve dataset quality. The library natively supports on-disk gradient storage and multi-node distributed training, making it feasible to apply these techniques to very large language models and pre-training datasets. A key contribution is the first open-source implementations of three leading data attribution methods: MAGIC, SOURCE, and TrackStar. The release addresses a significant engineering bottleneck in the field, as many cutting-edge data attribution techniques previously lacked accessible tooling and support for large-scale applications.
What different sources said
- arXiv cs.LGCenter
Bergson: An Open Source Library for Data Attribution
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.