Closed-Form Feedback-Free Learning with Forward Projection
Robert O’Shea, Bipin Rajendran · 2025
Read Closed-Form Feedback-Free Learning with Forward Projection by Robert O’Shea, Bipin Rajendran (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-16476.txt.
Description
Robert O'Shea and Bipin Rajendran propose a way to train neural networks with no feedback at all. Their 2025 paper works in the hardest corner of backpropagation-free learning, where a layer cannot hear anything back from the layers above it while it sets its own weights.
Their method, Forward Projection, needs a single pass through the data. Random projections of inputs and labels create targets for each layer's activity, and each layer fits those targets by closed-form regression. A bonus is clarity: each layer's activity reads as its own guess at the label. On biomedical data sets the method matched local learning rivals, learned faster, and in few-shot tests generalized better than backpropagation.
Catalog record
| CREATOR | Robert O’Shea, Bipin Rajendran |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 99 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-16476