aiqbench analysis
LocalProp introduces a neuro-inspired training method that restricts weight updates to localized blocks, significantly reducing GPU memory usage while maintaining competitive performance on standard benchmarks like ImageNet. By integrating pruning and a specific fine-tuning phase, this approach offers a practical solution for scaling deep learning models in memory-constrained environments without sacrificing accuracy.
LocalProp: Neuro-Localized Memory-Efficient Backpropagation
arXiv:2609.32517v1 Announce Type: cross Abstract: The current deep learning training paradigm employs end-to-end backpropagation, regardless of the training stage, i.e. pre-training or fine-tuning. However, backpropagating through the entire model is neither biologically plausible nor memory efficient, since learning inside the brain is highly localized. Therefore, we propose LocalProp, a training procedure that locally updates the weights of a model. Our neuro-localized weight updates follow the "pre-training then fine-tuning" paradigm, where the pre-training is based on I-JEPA. After locally updating the weights, a pruning operation is performed, followed by a short final fine-tuning phase. Pruning helps by sending the learning signal from higher blocks to lower blocks. We perform experiments on several datasets, including large-scale benchmarks such as ImageNet, and empirically show that LocalProp reaches good performance at a fraction of GPU peak memory. By varying the number of jointly optimized blocks, we identify gradient-propagation span as a practical control over the accuracy-memory trade-off.
Abstract:The current deep learning training paradigm employs end-to-end backpropagation, regardless of the training stage, i.e. pre-training or fine-tuning. However, backpropagating through the entire model is neither biologically plausible nor memory efficient, since learning inside the brain is highly localized. Therefore, we propose LocalProp, a training procedure that locally updates the weights of a model. Our neuro-localized weight updates follow the "pre-training then fine-tuning" paradigm, where the pre-training is based on I-JEPA. After locally updating the weights, a pruning operation is performed, followed by a short final fine-tuning phase. Pruning helps by sending the learning signal from higher blocks to lower blocks. We perform experiments on several datasets, including large-scale benchmarks such as ImageNet, and empirically show that LocalProp reaches good performance at a fraction of GPU peak memory. By varying the number of jointly optimized blocks, we identify gradient-propagation span as a practical control over the accuracy-memory trade-off.
Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Optimization and Control (math.OC)
arXiv:2609.32517 [cs.AI]
(or arXiv:2609.32517v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.32517 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 255 words
Related coverage
Neuromorphic Reinforcement Learning for Quadruped Locomotion Control on Uneven Terrain
arXiv · Jun 10, 2026
Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
AI Alignment Forum · Sep 23, 2026
Energy-Efficient Deep Learning Without Backpropagation: A Rigorous Hardware-Validated Benchmarking Study of Forward-Only Algorithms
arXiv · Oct 15, 2026
Academia is for Ambition — Alex Zhang, MIT
Latent Space · Oct 2, 2026