DeepFocus-BP: Error-Aware Adaptive Backpropagation via Dynamic Alpha-Beta Routing (Achieving 66% FLOPs Reduction with Improved Accuracy) - SOTA NLP Confirmed v3. (Resnet FAIL)

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Fernandes, Cláudio
Format: Recurso digital
Sprache:Englisch
Veröffentlicht: Zenodo 2026
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866901555240763392
author Fernandes, Cláudio
author_facet Fernandes, Cláudio
contents <p>This technical note presents DeepFocus-BP, a novel adaptive backpropagation algorithm that dynamically routes computational resources based on real-time error magnitude. By employing stochastic probing and adaptive Alpha-Beta thresholding, the algorithm categorizes network blocks into Skip, Full Precision, and Low Precision regimes.</p> <p>KEY RESULTS (Seed 42, IMDB Dataset):<br>- Test Accuracy: 84.10% (Surpassing Dense Baseline by +3.3%)<br>- Computational Efficiency: ~66.2% reduction in total FLOPs.<br>- Mechanism: Selective gradient suppression acts as a powerful regularizer.</p> <p>COMMERCIAL POTENTIAL:<br>This technology offers significant potential for reducing cloud infrastructure costs and energy consumption in large-scale AI training. The author is actively seeking strategic partnerships, licensing agreements, or co-founding opportunities to commercialize this algorithm.</p> <p>COPYRIGHT:<br>© 2026 Cláudio Fernandes. All Rights Reserved. Unauthorized commercial use is prohibited without explicit written permission.</p> <p>Contact for partnerships: benficaizeda306@gmail.com</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19415887
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle DeepFocus-BP: Error-Aware Adaptive Backpropagation via Dynamic Alpha-Beta Routing (Achieving 66% FLOPs Reduction with Improved Accuracy) - SOTA NLP Confirmed v3. (Resnet FAIL)
Fernandes, Cláudio
<p>This technical note presents DeepFocus-BP, a novel adaptive backpropagation algorithm that dynamically routes computational resources based on real-time error magnitude. By employing stochastic probing and adaptive Alpha-Beta thresholding, the algorithm categorizes network blocks into Skip, Full Precision, and Low Precision regimes.</p> <p>KEY RESULTS (Seed 42, IMDB Dataset):<br>- Test Accuracy: 84.10% (Surpassing Dense Baseline by +3.3%)<br>- Computational Efficiency: ~66.2% reduction in total FLOPs.<br>- Mechanism: Selective gradient suppression acts as a powerful regularizer.</p> <p>COMMERCIAL POTENTIAL:<br>This technology offers significant potential for reducing cloud infrastructure costs and energy consumption in large-scale AI training. The author is actively seeking strategic partnerships, licensing agreements, or co-founding opportunities to commercialize this algorithm.</p> <p>COPYRIGHT:<br>© 2026 Cláudio Fernandes. All Rights Reserved. Unauthorized commercial use is prohibited without explicit written permission.</p> <p>Contact for partnerships: benficaizeda306@gmail.com</p>
title DeepFocus-BP: Error-Aware Adaptive Backpropagation via Dynamic Alpha-Beta Routing (Achieving 66% FLOPs Reduction with Improved Accuracy) - SOTA NLP Confirmed v3. (Resnet FAIL)
url https://doi.org/10.5281/zenodo.19415887