Neural Networks Maintain Robustness When Learning From Heavily Corrupted Input Data
Researchers found that neural networks can learn effectively from datasets where over 90% of input data is corrupted, far exceeding human recognition limits. The study analyzed both additive and replacement noise corruption models using multi-layer perceptrons and infinite-width network theory. This finding has implications for understanding how machine learning systems can remain robust in real-world scenarios with imperfect data.
A new study on arXiv examines how neural networks maintain learning capability when trained on heavily corrupted input data while keeping labels intact—a scenario less studied than label noise. Researchers tested multi-layer perceptrons on corrupted classification datasets and discovered networks retained well-above-chance accuracy even with >90% input corruption. Through theoretical analysis of infinite-width networks using mean-field approaches, they identified that networks implement a prototype rule, assigning test points to the class whose training-set average they most closely resemble. This centroid-based decision mechanism proved universal across various MLP architectures, depths, activation functions, and noise distributions. The findings provide both practical insights into neural network robustness and theoretical understanding of why learning succeeds despite minimal signal in individual corrupted examples.
What's missing
The study's limitations regarding finite-width networks in practical settings, scalability to larger modern architectures (transformers, CNNs), and applicability beyond classification tasks are not detailed in the abstract.
What different sources said
- arXiv cs.LGCenter
Learning from almost nothing: How neural networks survive heavy input corruption
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.