Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Efficient On-Device Machine Learning with a Biologically-Plausible Forward-Only Algorithm

Huang, Baichuan LU orcid and Aminifar, Amir LU orcid (2025) MLSys 2025
Abstract
The training of the state-of-the-art Deep Neural Networks (DNNs) consumes massive amounts of energy, while the human brain learns new tasks with remarkable efficiency. Currently, the training of DNNs relies almost exclusively on Backpropagation (BP). However, BP faces criticism due to its biologically implausible nature, underscoring the significant disparity in performance and energy efficiency between DNNs and the human brain. Forwardonly algorithms are proposed to be the biologically plausible alternatives to BP, to better mimic the learning process of the human brain and enhance energy efficiency. In this paper, we propose a biologically-plausible forward-only algorithm (Bio-FO), not only targeting the biological-implausibility issues... (More)
The training of the state-of-the-art Deep Neural Networks (DNNs) consumes massive amounts of energy, while the human brain learns new tasks with remarkable efficiency. Currently, the training of DNNs relies almost exclusively on Backpropagation (BP). However, BP faces criticism due to its biologically implausible nature, underscoring the significant disparity in performance and energy efficiency between DNNs and the human brain. Forwardonly algorithms are proposed to be the biologically plausible alternatives to BP, to better mimic the learning process of the human brain and enhance energy efficiency. In this paper, we propose a biologically-plausible forward-only algorithm (Bio-FO), not only targeting the biological-implausibility issues associated with BP, but also outperforming the state-of-the-art forward-only algorithms. We extensively evaluate our proposed Bio-FO against other forward-only algorithms and demonstrate its performance across diverse datasets, including two real-world medical applications on wearable devices with limited resources and relatively large-scale datasets such as mini-ImageNet. At the same time, we implement our proposed on-device learning algorithm on the NVIDIA Jetson Nano and demonstrate its efficiency compared to other state-of-the-art forward-only algorithms. The code is available at https://github.com/whubaichuan/Bio-FO (Less)
Please use this url to cite or link to this publication:
author
and
organization
publishing date
type
Contribution to conference
publication status
published
subject
pages
14 pages
conference name
MLSys 2025
conference location
Santa Clara, CA, United States
conference dates
2025-05-12 - 2026-04-15
project
Resource-Efficient Deep Learning
language
English
LU publication?
yes
id
38609041-478d-4af7-a8d0-dcd486455f9d
alternative location
https://proceedings.mlsys.org/paper_files/paper/2025/hash/b0131b6ee02a00b03fc3320176fec8f5-Abstract-Conference.html
date added to LUP
2026-03-11 18:28:12
date last changed
2026-04-02 11:28:53
@misc{38609041-478d-4af7-a8d0-dcd486455f9d,
  abstract     = {{The training of the state-of-the-art Deep Neural Networks (DNNs) consumes massive amounts of energy, while the human brain learns new tasks with remarkable efficiency. Currently, the training of DNNs relies almost exclusively on Backpropagation (BP). However, BP faces criticism due to its biologically implausible nature, underscoring the significant disparity in performance and energy efficiency between DNNs and the human brain. Forwardonly algorithms are proposed to be the biologically plausible alternatives to BP, to better mimic the learning process of the human brain and enhance energy efficiency. In this paper, we propose a biologically-plausible forward-only algorithm (Bio-FO), not only targeting the biological-implausibility issues associated with BP, but also outperforming the state-of-the-art forward-only algorithms. We extensively evaluate our proposed Bio-FO against other forward-only algorithms and demonstrate its performance across diverse datasets, including two real-world medical applications on wearable devices with limited resources and relatively large-scale datasets such as mini-ImageNet. At the same time, we implement our proposed on-device learning algorithm on the NVIDIA Jetson Nano and demonstrate its efficiency compared to other state-of-the-art forward-only algorithms. The code is available at https://github.com/whubaichuan/Bio-FO}},
  author       = {{Huang, Baichuan and Aminifar, Amir}},
  language     = {{eng}},
  title        = {{Efficient On-Device Machine Learning with a Biologically-Plausible Forward-Only Algorithm}},
  url          = {{https://proceedings.mlsys.org/paper_files/paper/2025/hash/b0131b6ee02a00b03fc3320176fec8f5-Abstract-Conference.html}},
  year         = {{2025}},
}