ZeroS: Zero-Sum Linear Attention for Efficient Transformers
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917249289289728 |
|---|---|
| author | Lu, Jiecheng Han, Xu Sun, Yan Pati, Viresh Kim, Yubin Somani, Siddhartha Yang, Shihao |
| author_facet | Lu, Jiecheng Han, Xu Sun, Yan Pati, Viresh Kim, Yubin Somani, Siddhartha Yang, Shihao |
| contents | Linear attention methods offer Transformers $O(N)$ complexity but typically underperform standard softmax attention. We identify two fundamental limitations affecting these approaches: the restriction to convex combinations that only permits additive information blending, and uniform accumulated weight bias that dilutes attention in long contexts. We propose Zero-Sum Linear Attention (ZeroS), which addresses these limitations by removing the constant zero-order term $1/t$ and reweighting the remaining zero-sum softmax residuals. This modification creates mathematically stable weights, enabling both positive and negative values and allowing a single attention layer to perform contrastive operations. While maintaining $O(N)$ complexity, ZeroS theoretically expands the set of representable functions compared to convex combinations. Empirically, it matches or exceeds standard softmax attention across various sequence modeling benchmarks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_05230 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | ZeroS: Zero-Sum Linear Attention for Efficient Transformers Lu, Jiecheng Han, Xu Sun, Yan Pati, Viresh Kim, Yubin Somani, Siddhartha Yang, Shihao Machine Learning Artificial Intelligence Linear attention methods offer Transformers $O(N)$ complexity but typically underperform standard softmax attention. We identify two fundamental limitations affecting these approaches: the restriction to convex combinations that only permits additive information blending, and uniform accumulated weight bias that dilutes attention in long contexts. We propose Zero-Sum Linear Attention (ZeroS), which addresses these limitations by removing the constant zero-order term $1/t$ and reweighting the remaining zero-sum softmax residuals. This modification creates mathematically stable weights, enabling both positive and negative values and allowing a single attention layer to perform contrastive operations. While maintaining $O(N)$ complexity, ZeroS theoretically expands the set of representable functions compared to convex combinations. Empirically, it matches or exceeds standard softmax attention across various sequence modeling benchmarks. |
| title | ZeroS: Zero-Sum Linear Attention for Efficient Transformers |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2602.05230 |