New Training Method Improves Deep Spiking Neural Networks with Adaptive Asymmetric Surrogate Gradients
Researchers propose A2SG, a new framework using adaptive and asymmetric surrogate gradients to improve training of deep spiking neural networks (SNNs), which are brain-inspired neural networks that process information more efficiently than traditional models. The method addresses key challenges in SNN training by reducing gradient variation and better reflecting how neurons actually work. The approach shows consistent improvements in accuracy and energy efficiency across multiple tasks including image classification and segmentation.
A team of researchers has introduced A2SG, a unified framework designed to overcome training difficulties in deep spiking neural networks. SNNs are computationally efficient neural networks inspired by biological neurons, but they are challenging to train due to sharp loss landscapes and temporal inconsistency caused by surrogate gradients—mathematical approximations used during training. The proposed method uses adaptive gradients that adjust an effective window for spatio-temporal adaptation to reduce spatial gradient variation, and asymmetric gradients that assign larger values to neurons with higher membrane potentials, better reflecting actual neuronal dynamics. The authors provide theoretical analysis showing that asymmetric surrogates yield lower variation than symmetric ones and establish a connection between local gradient variation and loss landscape curvature. Extensive experiments on CNN-based and Transformer-based SNNs across image classification (both static and neuromorphic datasets) and segmentation tasks demonstrate consistent improvements in both accuracy and energy efficiency.
What different sources said
- arXiv cs.LGCenter
A2SG:Adaptive and Asymmetric Surrogate Gradients for Training Deep Spiking Neural Networks
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.