arXiv:2606. 01294v1 Announce Type: cross Abstract: Linear attention reduces the quadratic cost of softmax attention by maintaining a recurrent fast-weight state, but it consistently lags on in-context retrieval and long-context tasks.
arXiv:2608. 09307v1 Announce Type: new Abstract: We present a linearized form of 2-simplicial attention by rewriting the trilinear score as an inner product between a composite query and a key, so that the sum over one token axis takes the same form as ordinary softmax attention.
arXiv:2606. 02332v1 Announce Type: new Abstract: Combining attention's global retrieval with the sequential importance signal of state space models (SSMs) is the open challenge of hybrid language modeling.
The paper introduces Interdomain Attention (IA), a hybrid architecture that blends softmax attention with recurrent memory within a single layer, leveraging a kernel regression perspective to compress keys and values into a state‑space model (SSM) memory. IA reconstructs the attention matrix implicitly from memory states, preserving the computational efficiency of SSMs while achieving performance gains. Experiments on models ranging from 125 M to 1.3 B parameters show IA surpasses an S4D baseline at matched memory size, outperforms standard Transformers on perplexity and common‑sense reasoning, and matches or exceeds SSM‑based models on genomic downstream tasks.
arXiv:2609.13141v1 Announce Type: new Abstract: Post-training attention sparsification reduces the quadratic cumulative attention cost of pretrained Transformers by selecting a small set of context u...
The paper investigates how quantizing the softmax operation in Transformer attention affects pretraining. It explores K‑interval attention, varying grid calibration, interpolation versus hard rounding, and surrogate placement relative to normalization, deriving backward rules for each case. Experiments on a 124M‑parameter model show that choices like fixed‑window calibration with a post‑normalization surrogate or a pre‑normalization surrogate can reduce validation loss gaps compared to standard softmax.
arXiv:2605. 18079v2 Announce Type: replace Abstract: Existing expressivity results for transformers typically rely on hardmax attention, high precision, and other architectural modifications that disconnect them from the models used in practice.
arXiv:2601. 12145v3 Announce Type: replace Abstract: Softmax attention struggles with long contexts due to structural limitations: the strict sum-to-one constraint forces attention sinks on irrelevant tokens, and probability mass disperses as sequence lengths increase.
arXiv:2605. 18848v3 Announce Type: replace Abstract: This paper introduces Exact Linear Attention (ELA), a mechanism that achieves linear computational complexity for Transformer attention by exploiting the exact decomposition property of kernel functions, thereby eliminating approximation error.
The paper introduces Stratified Sparse Attention (SSA), a framework that blends deterministic top‑k retrieval with score‑guided sampling to reduce the gap between sparse and dense attention in large language models. SSA attends to high‑scoring tokens deterministically while sampling the remaining attention mass using proxy scores, thereby recovering more of the attention distribution. Experiments on benchmarks such as RULER, LongBench, LOFT, and AIME show that SSA outperforms existing sparse methods across models ranging from 3B to 27B parameters, achieving significant accuracy gains and up to 2.96× decode speedup with optimized CUDA kernels.
The paper introduces Semantic Head Specialization (SHS), a phenomenon where Vision Transformer (ViT) attention heads specialize as either object- or background-focused, most evident under full attention. It proposes the SHS-Index to quantify this specialization, demonstrating its ability to distinguish full-attention from chunk-window ViTs and its strong correlation with downstream benchmark performance. Leveraging insights into window interaction, token serialization, and local softmax allocation, the authors design Ariadne Attention, a hybrid attention mechanism that matches full-attention performance on 22 image and video tasks while reducing attention compute by 6.5×.
arXiv:2508. 17821v3 Announce Type: replace-cross Abstract: This paper investigates the limitations of the normalization in attention mechanisms.