Salvato in:
Dettagli Bibliografici
Autori principali: Lygizou, Elpiniki Maria, Farsang, Mónika, Grosu, Radu
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2505.24054
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911174986039296
author Lygizou, Elpiniki Maria
Farsang, Mónika
Grosu, Radu
author_facet Lygizou, Elpiniki Maria
Farsang, Mónika
Grosu, Radu
contents Transformers excel across a large variety of tasks but remain susceptible to corrupted inputs, since standard self-attention treats all query-key interactions uniformly. Inspired by lateral inhibition in biological neural circuits and building on the recent use by the Differential Transformer's use of two parallel softmax subtraction for noise cancellation, we propose Multihead Differential Gated Self-Attention (M-DGSA) that learns per-head input-dependent gating to dynamically suppress attention noise. Each head splits into excitatory and inhibitory branches whose dual softmax maps are fused by a sigmoid gate predicted from the token embedding, yielding a context-aware contrast enhancement. M-DGSA integrates seamlessly into existing Transformer stacks with minimal computational overhead. We evaluate on both vision and language benchmarks, demonstrating consistent robustness gains over vanilla Transformer, Vision Transformer, and Differential Transformer baselines. Our contributions are (i) a novel input-dependent gating mechanism for self-attention grounded in lateral inhibition, (ii) a principled synthesis of biological contrast-enhancement and self-attention theory, and (iii) comprehensive experiments demonstrating noise resilience and cross-domain applicability.
format Preprint
id arxiv_https___arxiv_org_abs_2505_24054
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Differential Gated Self-Attention
Lygizou, Elpiniki Maria
Farsang, Mónika
Grosu, Radu
Machine Learning
Transformers excel across a large variety of tasks but remain susceptible to corrupted inputs, since standard self-attention treats all query-key interactions uniformly. Inspired by lateral inhibition in biological neural circuits and building on the recent use by the Differential Transformer's use of two parallel softmax subtraction for noise cancellation, we propose Multihead Differential Gated Self-Attention (M-DGSA) that learns per-head input-dependent gating to dynamically suppress attention noise. Each head splits into excitatory and inhibitory branches whose dual softmax maps are fused by a sigmoid gate predicted from the token embedding, yielding a context-aware contrast enhancement. M-DGSA integrates seamlessly into existing Transformer stacks with minimal computational overhead. We evaluate on both vision and language benchmarks, demonstrating consistent robustness gains over vanilla Transformer, Vision Transformer, and Differential Transformer baselines. Our contributions are (i) a novel input-dependent gating mechanism for self-attention grounded in lateral inhibition, (ii) a principled synthesis of biological contrast-enhancement and self-attention theory, and (iii) comprehensive experiments demonstrating noise resilience and cross-domain applicability.
title Differential Gated Self-Attention
topic Machine Learning
url https://arxiv.org/abs/2505.24054