Self-Attention

Follow this topic Ask about this topic

Momentum

9 papers in the last four weeks, up 50% on the four weeks before. 0.1% of all new papers.

Jul 13Week of Sep 28

Latest papers 195

May 31, 2026cs.CV

Interpolation between Convolution and Attention via K-Nearest Neighbors

The shift from Convolutional Neural Networks to Transformers has reshaped computer vision, yet these two architectural families are typically viewed as fundamentally distinct. Convolutional Neural Networks are defined by spatially local convolution operations, while Transformers rely on global self-attention. We argue that convolution and self-attention, despite their apparent differences, can be unified within a single k-nearest neighbor aggregation framework. The critical insight is that both operations are special cases of neighbor selection and weighted aggregation. Convolution selects neighbors by spatial proximity while self-attention selects by feature similarity, revealing that they lie on a continuous spectrum rather than representing categorically different computations. We introduce Convolutional Nearest Neighbors (ConvNN), a unified framework that formalizes this connection. ConvNN exactly recovers standard and depthwise convolution by restricting neighbor selection to normalized spatial coordinates, and exactly recovers self-attention and its sparse variants, including KVT-attention, by replacing spatial proximity with scaled dot-product similarity. Beyond these special cases, ConvNN serves as a drop-in replacement for both convolution and attention layers, enabling systematic exploration of the intermediate spectrum between local and global aggregation through configurable similarity functions, neighbor selection strategies, positional encodings, and aggregation kernels.

Mingi Kang

Bowdoin College

Self-Attention Convolutional Neural Networks k-Nearest Neighbors

May 29, 2026cs.LG

Contribution Weights: A Geometrical Analysis of Self-Attention Transformers

Analyzing attention weights has become a standard approach for interpreting the information flow of Large Language Models (LLMs). However, this approach has significant limitations as it neglects the geometric properties of the value vectors being aggregated. To address this gap, we introduce \emph{Contribution Weights}, a projection-based metric that quantifies a token's influence by accounting for it's attention weight, value magnitude, and directional alignment with the layer output. We demonstrate that contribution weights provide a more faithful measure of token importance, consistently outperforming attention-based metrics in identifying semantically critical tokens across different decoder-only models, tasks, and datasets. Further, our metric enables novel mechanistic analysis of \emph{attention sinks}. While previous work characterized sinks as passive repositories for excess attention, we reveal they serve an active functional role, suppressing information through a convex relationship between sink rate and output norm, stabilizing representations by opposing the semantic drift of low-confidence tokens.

Harry Jake Cunningham, Nicola Muca Cirone

Department of Computer Science, University College London, London, UK. · Cartesia.AI, San Francisco, US.

Transformer Interpretability Feature Attribution Self-Attention Attention Sinks

May 28, 2026cs.CV

VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion

Long-rollout causal video diffusion has converged on a fixed-size sliding-window KV cache, with recent progress innovating within this layout by changing which tokens occupy the window or how their positions are encoded. The per-head KV layout itself, a dominant contributor to streaming memory and latency, has been mostly left unchanged. In this paper, we present the first study of Multi-Head Latent Attention (MLA) in video diffusion. VideoMLA replaces per-head keys and values with a shared low-rank content latent and a shared decoupled 3D-RoPE positional key, reducing per-token KV memory by 92.7% at every cached layer. We further investigate why MLA succeeds in video diffusion even though the spectral assumption often used to motivate it in language models does not hold: pretrained video attention is not low-rank, with 99%-energy effective rank far above any practical latent dimension. VideoMLA retains quality at compression ratios where direct spectral approximation would predict large reconstruction error. We show that the MLA bottleneck, rather than the pretrained spectrum, determines the effective rank: both spectral and random initialization occupy nearly the full rank budget from initialization, and training preserves this budget while adapting within it. On VBench, VideoMLA matches short-horizon streaming video diffusion baselines, achieves the best overall score at long horizons among evaluated methods, and improves throughput by 1.23x on a single B200.

Hidir Yesiltepe, Jiazhen Hu, Tuna Han Salih Meral +4

1Virginia Tech · 2fal

Video Diffusion Models Self-Attention Long-Horizon Video Generation Diffusion Model Caching Attention Mechanisms Autoregressive Video Generation+2

May 28, 2026cs.LG

Anti Mode-Collapse in Mean-Field Transformer via Auxiliary Variables

We use a mean-field-based transformer model to theoretically investigate how auxiliary variables, such as positional encoding, prevent mode collapse of self-attention mechanisms. The use of mean-field transformers to analyze the properties of self-attention mechanisms has garnered significant attention in recent years due to their ability to comprehensively analyze token interactions. However, analysis of this simple model suggests that mode collapse, where token distributions degenerate to a single point, occurs during long inferences (i.e., many layers), indicating a discrepancy with reality. This study investigates this mean-field transformer model and demonstrates that the introduction of auxiliary variables, such as positional encoding, acts as a counterforce against theoretical mode collapse. Specifically, we show that in the theoretical scheme, the energy-maximizing distribution does not degenerate to a single point; instead, it is characterized by a pushforward of the auxiliary variable distribution, thereby avoiding concentration in the Dirac measure. Our main examples are the positional encoding and the fixed prompt insertion treated as a parallel auxiliary-variable mechanism. Furthermore, we demonstrate that positional encoding and prompt insertion possess universality of representation in the limit, meaning that the limit distribution of inference can exactly represent a wide class of distributions. We also analyze several key properties of positional encoding and metastability, and validate our theoretical results through mathematical experiments.

Masaaki Imaizumi, Masanori Koyama, Noboru Isobe +1

The University of Tokyo, Tokyo, Japan · RIKEN Advanced Intelligene Project, Tokyo, Japan · Kyoto University, Kyoto, Japan

Transformer Self-Attention Model Collapse Positional Encoding

May 28, 2026cs.LG

Attention as In-Context Empirical Bayes: A Two-Stage View via Particle Dynamics

We study minimal attention-only transformers under all-token corruption and show they admit a two-stage empirical Bayes interpretation. A single attention step computes a kernel-weighted posterior mean with respect to the empirical distribution defined by the context. Depth refines this distribution through particle dynamics (Stage 1), while a long-range skip-connection carries the noisy input as a query for posterior inference (Stage 2), revealing distinct statistical roles for depth and attention residuals. The framework isolates a minimal setting in which the context itself induces a depth-dependent energy landscape governing in-context inference. We show that effective denoising can emerge without an explicit noise schedule: a fixed kernel bandwidth and finite integration horizon suffice, yielding a principled depth-noise relationship. We further establish a posterior-mean recovery guarantee for a class of well-behaved priors, where the empirical estimator converges to the Bayes-optimal predictor under asymptotic conditions. Connecting these dynamics to reverse-diffusion limits, our results provide a statistical interpretation of attention as in-context inference via sample-based posterior estimation, without explicit density modeling.

Matthew Smart, Soumya Ganguly, Nilava Metya +2

Lewis-Sigler Institute for Integrative Genomics, Princeton University, Princeton, NJ, USA · Department of Mathematics, Rutgers University, Piscataway, NJ, USA · Department of Physics and Astronomy, Rutgers University, Piscataway, NJ, USA +1

Self-Attention Empirical Bayes Bayesian Inference Transformer Attention In-Context Learning

May 27, 2026cs.LG

Parallax: Parameterized Local Linear Attention for Language Modeling

Large Language Models (LLMs) have become the central paradigm in artificial intelligence, yet the core computational primitive of attention has remained structurally unchanged. Local Linear Attention (LLA) is an attention mechanism derived from nonparametric statistics in the test-time regression framework. In contrast to prior research on efficient attention variants, LLA upgrades the local constant estimate in softmax attention to a local linear estimate, yielding provably superior bias-variance tradeoffs for associative memory. However, LLA has not been scaled in LLM pretraining due to computational and numerical stability concerns. We introduce Parallax, a parameterized Local Linear Attention that is scalable for LLMs. Parallax eliminates the numerical solver in LLA and learns an extra query-like projector that probes the KV covariance. We place Parallax within a family of attention mechanisms connected by the bandwidth, the probe construction and the affine structure. We propose a hardware-aware algorithm that increases the arithmetic intensity over FlashAttention, shifting attention into a more compute bound regime. Our prototype decode kernel matches or outperforms FlashAttention 2/3 across diverse batch sizes and context lengths. We pretrain Parallax at 0.6B and 1.7B scales and find consistent perplexity improvements throughout pretraining with gains that transfer to downstream benchmarks. The advantage persists under both parameter-matched and compute-matched controls, demonstrating a Pareto improvement. We perform careful pretraining ablations and identify a novel phenomenon whereby Muon unlocks the capacity of Parallax. To our knowledge, this is the first empirical demonstration of strong architecture-optimizer codesign for attention mechanisms in the architecture research literature.

Yifei Zuo, Dhruv Pai, Zhichen Zeng +3

Northwestern University · Tilde Research · University of Washington

Language Model Pretraining Self-Attention LLM Inference Acceleration Linear Attention

May 27, 2026cs.LG

Augmenting Attention with Exponentially Decaying Memory Improves Query-Aware KV Sparsity

Efficient inference is critical for long-context language models, where attention computation and KV-cache access dominate the cost. Recent work RAT+, introduces a recurrence-augmented attention backbone that enables flexible dilated attention at inference time. In this paper, we investigate whether this exponentially decaying memory can also improve existing query-aware sparse inference methods. Using representative methods including Quest, MoBA, and SnapKV, we show that RAT+ consistently improves accuracy over standard attention across sparse budgets on eight needle-in-a-haystack tasks. We validate these gains both on the released checkpoints from the RAT+ paper and on OLMo2-7B, which we continue pretraining with the added memory module for 10B tokens. Finally, we propose two hypotheses explaining why this memory module benefits query-aware sparse inference and design targeted experiments to support them.

Xiuying Wei, Caglar Gulcehre

EPFL, CLAIRE Lausanne, Switzerland

Memory-Augmented Language Models Self-Attention KV Caching Long-Context Language Model Inference LLM Inference Acceleration Sparse Attention

May 27, 2026cs.LG

PLS in the Mirror of Self-Attention

This note provides an interesting observation on casting partial least square (PLS) as a linearized self-attention so that PLS may be studied within the neural network paradigm. On the other hand, the dimensionality reduction and selection of predictors in PLS may indicate that self-attention includes certain degree of dimensionality normalization toward improved learning.

Jiangsheng, You

Jason

Recursive Least Squares Self-Attention

May 27, 2026cs.LG

Meta-Attention: Bayesian Per-Token Routing for Efficient Transformer Inference

Standard transformer architectures apply a single attention mechanism uniformly across all tokens and sequence positions, irrespective of local context or computational budget. We propose Meta-Attention, a framework that dynamically routes each token to the most appropriate attention strategy -- full softmax attention, linear (kernel) attention, or sliding-window local attention -- via a Bayesian Meta-Controller. Unlike prior routing approaches that use deterministic or prior-free learned routing, the Meta-Controller treats per-token mechanism selection as posterior inference under a compute-aware Dirichlet prior: routing weights are the output of an amortised variational posterior q(alpha | x_t; phi) trained with an Evidence Lower Bound (ELBO) objective that jointly encodes task performance and attention-mechanism cost. This design produces principled routing uncertainty estimates that govern the soft-to-hard routing transition, mitigates routing collapse without ad hoc load-balancing losses, and yields better compute-performance trade-offs than deterministic or prior-free learned routing at negligible overhead. Phase 1 empirical results on a Tiny LM benchmark confirm core predictions: the Bayesian controller's learned routing distribution implies a projected normalised FLOP cost of 25.1% under hard routing, vs. 59.3% for the prior-free baseline (-34.2 pp), and reduces routing entropy from 55.8% to 43.3% (-12.5 pp), demonstrating that the Dirichlet prior prevents routing collapse while the non-Bayesian model defaults to full attention. We present the Bayesian architecture, ELBO training objective, and a Phase 1 PyTorch prototype validating forward-pass correctness, posterior diversity, and a controlled ablation against a prior-free baseline. Code available at: https://github.com/KFEAL/meta-attention

Alan Ferrari

Knowledge Lab AG Zürich, Switzerland

Efficient Transformer Inference Self-Attention Bayesian Inference Transformer Attention

May 26, 2026cs.CL

UNIQUE: Universal Top-k Sparse Attention for Training-free Inference and Sparsity-aware Training

Long-context inference in large language models (LLMs) is bottlenecked by the linear growth of the self-attention key-value (KV) cache. Top-k sparse attention alleviates this by loading only a small fraction of the KV cache, but accurately and cheaply estimating cache importance, for both training-free use and sparsity-aware training, remains challenging. This paper proposes UNIQUE, a universal top-k sparse attention framework that addresses both requirements and stays consistently effective across LLM modalities. UNIQUE operates at the granularity of KV pages and estimates per-page importance with a simple yet accurate score combining the mean of the page's keys as a representative vector with their standard deviation as an offset term. To further close the train-inference gap, this paper introduces a soft-mask sparsity-aware training scheme that uses the top-k score boundary as a per-query threshold and a sigmoid soft mask around it, requiring neither auxiliary losses nor architectural changes. Experiments on text and speech LLMs show that UNIQUE preserves task performance on long-context benchmarks such as LongBench Pro and on long-form speech recognition, while delivering up to 11.4x attention-kernel speedup over FlashInfer dense attention and at least 5.3x end-to-end decoding speedup over a vLLM-based dense model.

Keqi Deng, Shaoshi Ling, Ruchao Fan +1

Microsoft, USA

Self-Attention Long-Context Language Model Inference LLM Inference Acceleration Sparse Attention KV-Cache Compression

May 26, 2026cs.LG

Balancing Fidelity and Diversity in Diffusion Models via Symmetric Attention Decomposition: Hopfield Perspective

We characterize the pre-softmax attention matrix

\mathbf{QK^\top}

in transformers as an associative memory matrix encoding pairwise associations between input features. By decomposing this matrix into its symmetric and skew-symmetric parts, we interpret the symmetric component as governing the structure of the energy landscape, and the skew-symmetric component as driving circulation on that landscape. Leveraging the energy formulation induced by the symmetric component, we derive Hopfield-style stability measures that quantify the stability of retrieved features. We observe meaningful correlations between Hopfield-style stability measures and the fidelity-diversity trade-offs in generation. Finally, we propose a controllable knob to modulate this trade-off by modifying the circulation of the underlying dynamics. Code is available at our GitHub (https://github.com/hyeon-cho/Attention-Symmetric-Decomposition).

Hyunmin Cho, Woo Kyoung Han, Kyong Hwan Jin

Department of Electrical Engineering, Korea University, Seoul, South Korea.

Self-Attention Transformer Attention Generative Modeling Diffusion Models

May 26, 2026cs.CV

AdaMerge: Salience-Aware Adaptive Token Merging for Training-Free Acceleration of Vision Transformers

The quadratic cost of self-attention in Vision Transformers (ViTs) constitutes a fundamental bottleneck for practical deployment, motivating a vibrant line of research on token reduction. Among existing approaches, token merging (ToMe) has emerged as an elegant training-free solution; yet its design rests on an unspoken premise of token equality, which contravenes the well-documented non-uniformity of self-attention and leads to information loss in high-salience tokens under aggressive compression. We address this limitation with AdaMerge, a token-merging framework based on two complementary mechanisms. First, salience-weighted similarity leverages column-wise feature-affinity centrality as a token-importance proxy and incorporates the resulting salience scores into the bipartite matching score, ensuring that pivotal tokens contribute more strongly to the merged representation. Second, adaptive merging intensity uses pre-computed layer-wise similarity statistics to dynamically modulate the per-layer reduction count in accordance with input-specific redundancy. On ImageNet-1k with ViT-B/16, AdaMerge consistently outperforms ToMe, PiToMe, and DSM across all FLOPs-matched regimes. The accuracy gap widens monotonically with compression: at the 13.4G FLOPs operating point, AdaMerge sustains a Top-1 degradation of only -1.06%, compared to -1.45% for PiToMe and -4.62% for DSM. To our knowledge, AdaMerge is the first to combine salience-weighted similarity and adaptive per-layer reduction into a single training-free token merging framework, advancing the accuracy-FLOPs Pareto frontier of ViT acceleration.

Semi Lee, Hyejin Go, Hyesong Choi

Electronic Engineering · Soongsil University · Seoul, South Korea

Token Merging Self-Attention Vision Transformer Efficient ViTs Visual Token Merging Visual Token Compression

May 25, 2026cs.LG

Energy-Gated Attention and Wavelet Positional Encoding: Complementary Inductive Biases for Transformer Attention

Standard transformer attention computes pairwise token similarity but treats all tokens as equally salient and all positions as equally local, regardless of the informational structure of the input. We identify two complementary inductive biases that standard attention lacks: energy salience (which tokens concentrate informational energy, learned end-to-end without explicit frequency decomposition) and scale-selective locality (how far positional influence extends at each frequency, implemented via Morlet wavelet encoding). We address both with two simple components. Energy-Gated Attention (EGA) gates value aggregation by a learned energy estimate of key token embeddings, computed via a single linear projection; it selects what to attend to. Morlet Positional Encoding (MoPE) replaces fixed sinusoidal encodings with learned Gaussian-windowed wavelets that adapt the joint position-frequency localization to the corpus; it specifies where attention operates at each scale. On TinyShakespeare, EGA alone achieves +0.092 validation loss improvement over standard attention (+0.103 over Phase 1-3 baseline); MoPE alone is -0.032 (below baseline as a standalone encoding); but their combination achieves +0.119 -- more than the sum of parts. This superadditivity, observed across two independent training runs, is the central empirical finding: salience and locality are complementary inductive biases, each addressing a gap the other cannot fill alone. Ablations confirm that structured spectral priors (Morlet wavelet gates, scale-initialized heads, fixed sinusoidal PE) consistently underperform their unconstrained learned counterparts, while complementary learned components interact superadditively. All experiments are at small scale (<=6M parameters, character-level benchmarks, single seed); larger-scale multi-seed validation is the most important direction for future work.

Athanasios Zeris

Independent Researcher, Athens, Greece.

Self-Attention Gated Attention Transformer Attention Positional Encoding Attention Mechanisms

May 25, 2026cs.LG

Quantized Keys Steal Attention: Bias Correction for KV-Cache Compression in Video Diffusion

Chunk-wise autoregressive video diffusion models rely on a KV cache of previously generated chunks to avoid redundant computation, but this cache quickly becomes a memory bottleneck as videos grow longer. Methods that quantize the KV cache to low bitwidths reduce memory pressure but degrade video quality. We show that a key driver of this degradation is a systematic bias in attention weights: due to the convexity of the exponential in softmax attention, quantization noise inflates the contribution of cached keys, a phenomenon we call the Jensen bias. This effect causes quantized keys to steal attention mass from the unquantized current chunk. We derive a per-attention-score correction that removes this bias in expectation, computed on the fly from the quantization step sizes of the cached keys and the query norm. Using a second-order Taylor approximation, the additional computational overhead is negligible, and no additional memory is needed alongside the cache. Evaluated on MAGI-1, SkyReels-V2, and HY-WorldPlay at INT2 quantization, our correction recovers most of the quality lost to aggressive quantization, reaching near-BF16 video quality, and can outperform INT4 quantization while using 50% less memory.

Tuna Tuncer, Felix Becker, Thomas Pfeil

1Technical University of Munich · 2Tensordyne

Video Diffusion Models Self-Attention Diffusion Model Quantization KV Caching Attention Quantization Autoregressive Video Generation+2

May 25, 2026cs.CL

IndexMem: Learned KV-Cache Eviction with Latent Memory for Long-Context LLM Inference

Large Language Models (LLMs) are increasingly expected to operate over long contexts, yet standard softmax attention incurs a KV cache that grows linearly with sequence length, quickly becoming the bottleneck for long context inference. A practical remedy is to evict less important KV entries; however, existing eviction policies are largely heuristic and struggle to capture the rich, input-dependent distribution of token importance. In this work, we introduce a learnable indexer that predicts KV importance, enabling more accurate retention of critical tokens. Meanwhile, naively evicting tokens permanently discards their information, leading to irreversible forgetting and degraded retrieval over long ranges. To address this, we propose a lightweight latent memory module that compresses evicted tokens into a compact, online-updated state and provides residual readouts to compensate for the attention contributions lost through KV eviction. Collectively, our method enables accurate long-context inference under a bounded KV budget, delivering consistent improvements on RULER (4K/16K) across Qwen, Mistral, and Llama models (up to 25 points under aggressive eviction), markedly more stable Needle-in-a-Haystack retrieval, and superior LongBench scores and compression curves compared to existing eviction policies.

Xintong Yang, Hao Gu, Binxing Xu +6

The Hong Kong University of Science and Technology · Zhejiang University

Memory-Augmented Language Models Self-Attention KV Caching Long-Context Language Model Inference KV-Cache Eviction LLM Inference Acceleration

May 24, 2026cs.LG

Quaternion Self-Attention with Shared Scores

Quaternion neural networks are parameter-efficient and model multidimensional dependencies by representing four related features as a single entity. However, existing quaternion self-attention computes component-wise scores and applies independent softmax operations to each component, which increases the computational cost and allows attention distributions to diverge across components. We propose a shared-score quaternion self-attention mechanism that computes a single real-valued score using the quaternion inner product and applies a shared attention distribution across all components. This reduces score-computation multiplications by 75% and the number of softmax operations from four to one. We prove that, when queries and keys are produced by quaternion linear projections that induce component pre-mixing, the component-wise and shared scores lie in the same interaction subspace, indicating that independent component-wise attention primarily re-parameterizes the same interactions rather than expanding the feature interaction space. In speech enhancement, our method reduces inference time by up to 44.3% on a GPU and 58.1% on a CPU while maintaining quality, with consistent trends across vision and natural language processing.

Shogo Yamauchi, Tohru Nitta, Hideaki Tamori

The Asahi Shimbun Company, Tokyo, Japan · Tokyo Woman’s Christian University, Tokyo, Japan.

Self-Attention Attention Mechanisms

May 23, 2026cs.LG

Interdomain Attention: Beyond Token-Level Key-Value Memory

Transformers and deep state space models (SSMs) sit at opposite ends of a basic design choice: attention routes each query through a growing key-value (KV) cache by content-based matching at quadratic cost, while deep SSMs compress context into a fixed-size recurrent state that is not directly addressed by query-key matching. We propose Interdomain Attention, which integrates an SSM into an attention module through kernel methods: an attention kernel is approximated by a finite feature map, the resulting key features and values are projected onto a shared set of basis functions maintained by a single SSM recurrence, and each query attends to the compressed coefficients through its own feature map, recovering query-conditioned attention over a fixed-size state. The scalable layer is a learned relaxation of this derivation, and we validate its components through ablations. In a 125M to 1.3B autoregressive language-modeling study on FineWeb-Edu at matched recurrent-state budget, Interdomain Attention improves on an SSM token mixer at every scale, surpasses a same-recipe softmax baseline at 1.3B on validation perplexity and on the eight-task commonsense suite, and inherits the length-flat behavior of its fixed-state core out to 3.5x the training context. Ablations indicate that the query-conditioned projection is the main source of the gain.

Naoki Kiyohara, Harrison Bo Hua Zhu, Riccardo El Hassanin +4

Self-Attention Long-Context Language Modeling Efficient Attention Linear Attention State Space Models

May 22, 2026cs.LG

Approaching I/O-optimality for Approximate Attention

We revisit the I/O complexity of attention in large language models. Given query-key-value matrices

Q,K,V\in\mathbb{R}^{n\times d}

, and a machine with fast memory size

M

, the goal is to compute the "attention matrix"

A=\text{softmax}(Q K ^{\top}/\sqrt{d}) V

with the minimal number of data transfers between fast and slow memory. Existing methods in the literature, most notably FlashAttention and its variants, incur an I/O cost that depends quadratically on

n

, while a trivial lower bound only requires

Ω(nd)

I/O's to read the inputs and write the output. In this work, we present a technique for computing attention where the I/O cost only depends almost-linearly on

n

in most parameter regimes. This is achieved by developing I/O-efficient algorithms inspired by the recent approximate attention framework of Alman and Song. We also prove corresponding lower bounds in each parameter regime to show that our algorithms are indeed close to I/O-optimal.

Pál András Papp, Aleksandros Sobczyk, Anastasios Zouzias

Computing Systems Lab, Huawei Technologies, Zurich, Switzerland

Self-Attention Efficient Attention

May 22, 2026cs.CV

Improved Belief-Attention in Vision Task

Recently, Belief-Attention \cite{Guoqiang25BeliefAttention} has been proposed by first performing an orthogonal projection of the softmax-based weighted summation of

V

vectors with respect to the original

V

vectors and then taking the perpendicular component as the residual signal in Transformer for performance improvement. In this paper, we first conduct an ablation study showing the projected component also carries information about the token correlation, which should not be ignored. We then propose to extend Belief-Attention by making use of both the perpendicular and projected components. In particular, the projected component goes through certain activation function and then a linear mapping before merging with the considered token. Conceptually speaking, the neural block for the projected component can be viewed as a two-layer feedforward network (FFN) within the new attention block. It is also noted that standard attention captures the token correlation via the inner-product matrix

QK^T

. We propose to introduce an additional inner-product matrix

ZZ^T

to

QK^T

to capture richer token correlation. We refer to the new module as Belief2-Attention. It can be easily shown that Belief2-Attention is more expressive than standard Attention. We then verify the effectiveness of Belief2-Attention for vision tasks of image classification and segmentation.

Guoqiang Zhang

University of Exeter

Visual Attention Self-Attention Vision Transformer Transformer Attention

May 22, 2026cs.LG

Spectral Probe-Circuits: A Three-Step Recipe for Identifying Attention-Head Circuits in Pretrained Transformers

We present a three-step recipe for identifying attention-head circuits in pretrained transformers. A per-head spectral signal -- the time-integrated participation ratio of each head's attention output -- ranks heads doing sustained content-dependent computation without labels or attribution gradients. A task-pattern screen filters this general indicator into a task-specific candidate circuit, and group ablation against a matched-random control completes the causal claim. We validate across an 8x parameter range (51M to 1B-active / 7B-total), two architecture families (dense, mixture-of-experts), and four pretraining pipelines. The recipe ports: a 2-6 head induction circuit is causally necessary in every model tested, with a 94-100% drop in synthetic-induction top-1 after ablation. The spectral signal is predictive without supervision: on six independent seeds of a 51M-parameter probe model, the same computation identifies the seed-specific circuit on each seed. The fraction of heads doing identifiable specialized computation is conserved at 17-19% across the Pythia family (124M to 410M), while specific induction circuits stay 3-11 heads -- sublinear in total head count. This paper is the methodology anchor of a three-paper program; companion papers extend the recipe to developmental trajectories during pretraining and to composed-task circuits where pattern selectivity decouples from task-causal structure.

Yongzhong Xu

Transformer Interpretability Circuit Discovery Attention Head Analysis Self-Attention Mechanistic Interpretability Induction Heads+2

May 21, 2026cs.LG

ThriftAttention: Selective Mixed Precision for Long-Context FP4 Attention

Efficient attention algorithms are critical to mitigate the quadratic cost of attention in long-context workloads. Prior work utilises block-scaled quantisation techniques on Blackwell GPUs to move attention computation to 4-bit precision to accelerate inference. However, these techniques result in significant quality degradation in long-context settings. We show that the output impact of quantisation error is highly non-uniform and increases with the importance of each query-key interaction, concentrating functionally relevant error in a small number of attention blocks that contain the most important tokens. We propose ThriftAttention, a low-bit attention variant that delivers near-FP16 long-context quality at FP4 inference efficiency. This approach proceeds in two stages. First, a heuristic rapidly selects a small number of important query-key block pairs for FP16 precision. Second, the selected blocks are computed in FP16 and the remaining blocks in FP4, with both paths merged via online softmax into a single output. We demonstrate across long-context benchmarks and model families that by computing only 5% of query-key blocks in FP16, ThriftAttention recovers on average 89.1% of the FP4-to-FP16 performance gap. We show ThriftAttention's advantage grows with sequence length, mitigating the systematic FP4 quality degradation observed at longer contexts. The code is available at https://github.com/joesharratt1229/ThriftAttention.

Joe Sharratt

Mixed-Precision Quantization Self-Attention Long-Context Language Model Inference Efficient Attention Attention Quantization Low-Bit Quantization

May 21, 2026cs.LG

Structured-Sparse Attention for Entity Tracking with Subquadratic Sequence Complexity

Entity tracking requires maintaining and updating latent states for entities and attributes over long sequences. Recent task-specific attention operators can compress deep Transformer stacks into a few layers by performing multi-hop state propagation within a single layer, but their dense evaluation remains expensive. We show that in this setting, learned attention is strongly structured: most mass concentrates in local block-diagonal neighborhoods with a light cross-block residue. Exploiting this, we derive a blockwise evaluation of a resolvent-style operator that keeps within-block interactions exact and routes cross-block interactions through a reduced system. The resulting evaluation is subquadratic in sequence length

O(n^{4/3}d)

(and

O(n^{7/3})

when

d\approx n

). On controlled tracking benchmarks, our method matches the dense operator's accuracy while reducing wall-clock time by

12-29\%

under a standardized measurement protocol, and is up to

2.4 \times

faster than a compact dense Transformer at comparable exact-match accuracy. We further provide ablations over block size and model capacity, and identify a limitation: performance collapses when the number of simultaneously evolving properties exceeds the number of attention heads.

Hangyue Zhao, Paul Caillon, Erwan Fagnou +1

ESPCI PSL, Paris, France · LAMSADE, Université Paris Dauphine - PSL, Paris, France

Self-Attention Structured Sparsity Transformer Attention Efficient Attention Sparse Attention State Tracking

May 21, 2026cs.LG

ASAP: Attention Sink Anchored Pruning

Vision Transformers (ViTs) face severe computational bottlenecks due to the quadratic complexity of self-attention at high resolutions. Existing token reduction methods rely on local metrics - such as single-layer attention scores - that are inherently vulnerable to the attention sink phenomenon, where uninformative tokens are paradoxically preserved over salient foreground objects. We propose ASAP (Attention Sink Anchored Pruning), a training-free framework that recasts this sink as a feature. Modeling ViT information flow as a Lazy Random Walk, ASAP identifies the sink as a dominant accumulator of probability mass. By computing the diffusion distance to the sink within the cumulative transition matrix, ASAP partitions tokens via Radial Diffusion Clustering and compresses background redundancy through Transition Weight Pooling in a single shot. Extensive experiments across image, video, and vision-language tasks demonstrate ASAP outperforms state-of-the-art methods, accelerating throughput by up to 48% while maintaining - or even exceeding - baseline accuracy.

Jaehyuk Lee, Hanyoung Kim, Yanggee Kim +1

Department of Mathematics Korea University Seoul, Republic of Korea · Korea University Seoul, Republic of Korea

Self-Attention Efficient ViTs Attention Sinks Visual Token Pruning Visual Token Compression

May 21, 2026cs.LG

Tensor Cache: Eviction-conditioned Associative Memory for Transformers

Autoregressive Transformer KV caches grow linearly with context length; sliding-window caching bounds memory but discards evicted tokens entirely, so relevant evidence outside the window becomes inaccessible. We introduce \emph{Tensor Cache}, a two-level cache that pairs sliding-window softmax attention as a first-level cache (L1) with a fixed-size outer-product fast-weight memory as a second-level cache (L2) fed by KV pairs evicted from the window. Recent tokens remain in exact local attention; evicted pairs are compressed into a per-layer matrix

A

and read by future queries through a single matrix multiplication, exploiting the linear-attention identity

q_t(k_i \otimes v_i)=\langle q_t,k_i\rangle v_i

. A learned scalar gate fuses the L1 and L2 outputs, and per-head decay and write-rate parameters are trained end-to-end. The outer-product memory and the read identity are well-known; our contribution is their use as an L2 cache fed exclusively by sliding-window evictions, plus identifying that the common chunked-mean training shortcut

A\!\leftarrow\!λA\!+\!η(\bar k\!\otimes\!\bar v)

silently introduces

C^2{-}C

spurious cross-token outer products per chunk, and closing the gap with a parallel weighted-sum scan equivalent to per-token writes within float32 epsilon. Across systems scaling, controlled associative recall, long-context language modeling, and memory-capacity diagnostics, Tensor Cache improves the memory--quality frontier over bounded-state baselines.

Kabir Swain, Sijie Han, Daniel Karl I. Weidele +2

Massachusetts Institute of Technology, Cambridge, MA, USA · University of Toronto, Toronto, Canada · IBM Research, Cambridge, MA, USA

Memory-Augmented Neural Networks Self-Attention KV Caching KV-Cache Eviction KV-Cache Management

May 21, 2026cs.LG

Energy-Gated Attention: Spectral Salience as an Inductive Bias for Transformer Attention

Standard transformer attention computes pairwise similarity between queries and keys, treating all tokens as equally salient regardless of their intrinsic informational content. In turbulent fluid dynamics, coherent structures -- the energetically dominant, spatially organized patterns that persist amid background chaos -- carry a disproportionate fraction of total energy and govern all transport. We propose that tokens play an analogous role in transformer attention: informationally dense positions (morphological boundaries, syntactic heads, discourse markers) concentrate spectral energy and should attract proportionally more attention than background tokens (function words, repeated patterns, low-information filler). We propose Energy-Gated Attention (EGA): a simple modification that gates value aggregation by the spectral energy of key token embeddings, computed by a single learned linear projection that discovers the dominant spectral mode of the embedding field. On TinyShakespeare, EGA achieves +0.103 validation loss improvement with only 12,480 additional parameters (<0.26% overhead) and no measurable computational cost. The result is consistent on Penn Treebank (+0.101), demonstrating dataset independence. A systematic ablation across three wavelet families (fixed Morlet, Daubechies db2/db4, and a parametric Morlet) establishes that fixed structured bases are suboptimal -- the optimal energy direction is data-adaptive and non-sinusoidal -- while identifying learned wavelet packets as a promising open direction. The learned energy threshold converges to tau ~= 0.35 independently of initialization, corresponding to the fraction (~36%) of tokens carrying above-average spectral energy in English text, a stable linguistic property consistent with the fraction of content words in running English text.

Athanasios Zeris

Independent Researcher, Athens, Greece.

Self-Attention Gated Attention Transformer Attention Language Modeling Attention Mechanisms

May 20, 2026cs.LG

Manifold-Guided Attention Steering

Large language models frequently produce errors in reasoning tasks despite possessing the underlying knowledge required for correct reasoning. One possible approach to improve reasoning consistency is through activation steering. However, existing activation steering approaches apply fixed, pre-computed correction vectors, ignoring where the model currently sits along its generation trajectory; the result is indiscriminate perturbation that disrupts already-correct steps as freely as erroneous ones. We propose Manifold-Guided Attention Steering (MAGS), a trajectory-aware inference-time intervention grounded in a geometric observation: the output activations of specific attention heads diverge from a low-dimensional correctness manifold at the point of error, and this deviation compounds through subsequent steps. For each identified attention head, we learn a low-dimensional subspace from contrastive pairs of correct and incorrect traces that capture the directions along which error behavior deviates from correct behavior. During inference, we monitor each head's proximity to this manifold and apply a targeted projection correction when deviation exceeds a learned threshold, steering the attention output back toward the correct subspace before the error propagates. MAGS consistently outperforms both unsteered baselines and static steering approaches across benchmarks spanning mathematical reasoning (MATH-500, GSM8K), code generation (HumanEval, MBPP), and molecular generation (SMILES), suggesting that correctness manifolds are a general feature of LLM attention geometry.

Ian Li, Kapilesh Guruprasad, Raunak Sengupta +3

University of California, San Diego

Language Model Steering Self-Attention Activation Steering LLM Reasoning

May 20, 2026cs.LG

EntmaxKV: Support-Aware Decoding for Entmax Attention

Long-context decoding is increasingly limited by KV-cache memory traffic since each generated token attends over a cache whose size grows linearly with context length. Existing sparse decoding methods reduce this cost by selecting subsets of tokens or pages, but are designed for softmax attention, whose dense tails make any truncation discard nonzero probability mass. In contrast,

α

-entmax produces exact zeros, turning sparse decoding from dense-tail approximation into support recovery: if the selected candidates contain the entmax support, sparse decoding remains exact. While recent entmax kernels enable efficient training, they do not address the autoregressive decoding bottleneck, where dense inference still streams the full KV cache before sparsity is known. In this work, we introduce EntmaxKV, an entmax-native sparse decoding framework that exploits sparsity before KV pages are loaded. EntmaxKV combines query-aware page scoring, support-aware candidate selection, and sparse entmax attention. We analyze truncation error through the dropped probability mass

δ

, showing that output error is controlled by

δ

and vanishes when the entmax support is recovered. We further introduce a Gaussian-aware entmax selector that estimates the entmax threshold from lightweight page statistics, adapting the selected budget to the score distribution. Empirically, EntmaxKV drops less probability mass, retains more support tokens, and achieves lower output error than softmax-based sparse decoding at matched KV budgets. On long-context and language modeling benchmarks, it closely matches full-cache entmax while using a small fraction of the KV cache, achieving up to

3.36\times

(softmax) and

5.43\times

(entmax) speedup over full attention baselines at 1M context length. Code available at: https://github.com/deep-spin/entmaxkv.

Gonçalo Duarte, Miguel Couceiro, Marcos V. Treviso

Instituto Superior Técnico, Universidade de Lisboa. · ELLIS Unit Lisbon. · INESC-ID. +1

Self-Attention KV Caching Long-Context Language Model Inference LLM Inference Acceleration Sparse Attention KV-Cache Compression

May 20, 2026cs.LG

Towards Understanding Self-Pretraining for Sequence Classification

Amos et al. (2024) showed that the accuracy of Transformer models in sequence classification can be significantly improved by first pretraining with a masked token prediction objective without external data or augmentation, a procedure referred to as self-pretraining (SPT). While the primary objective of Amos et al. (2024) was to showcase that Transformers can achieve strong performance on the Long-Range Arena (LRA), their pipeline raises more fundamental questions: How does SPT drive optimization to better solutions? Why can standard supervised training fail in Transformers? To better understand this, we replicate and systematically ablate the findings of Amos et al. (2024). Our ablations suggest that a central bottleneck in the studied settings is not depth or generalization alone, but the ability of label supervision to learn useful query-key Attention patterns from random initialization. With a minimal setup, we identify learning proximity interactions - turning absolute positional encodings into proximity-biased Attention scores - as a key source of the improvements brought by SPT. Finally, in a simplified theoretical setup, we show that label supervision can be locally blind to certain Attention-score directions that are instead detectable through masked reconstruction.

Omar Coser, Loredana Zollo, Paolo Soda +1

Self-Attention Masked Language Modeling Self-Supervised Pre-Training Transformer Attention Positional Encoding Sequence Modeling

May 20, 2026cs.CL

PulseCol: Periodically Refreshed Column-Sparse Attention for Accelerating Diffusion Language Models

Inference in diffusion large language models (dLLMs) is computationally expensive, as full self-attention must be repeatedly executed at each step of the denoising process without KV cache. Recent sparse attention methods for dLLMs mitigate this cost via block-sparse computation, which is applied only in later iterations when model performance is less sensitive to coarse-grained sparse approximation, but yields limited improvements in computational efficiency and acceleration. This motivates a finer-grained sparsification strategy that can be applied from earlier iterations and leverages reusable sparsity patterns, enabling further efficiency gains. In this work, we introduce PulseCol, a periodically refreshed column-sparse attention method for accelerating diffusion language models. PulseCol replaces coarse block-level sparsity with a finer-grained column-sparse structure, allowing important attention interactions to be retained more precisely while exposing greater sparsity. Built on this column-level formulation, PulseCol further identifies sparse patterns at the early denoising step and reuses them across subsequent iterations, refreshing them only at a small number of intermediate steps to track the evolution of sparse attention patterns during denoising. Experiments show that PulseCol achieves higher sparsity and greater practical speedup than prior sparse attention methods for dLLMs, while maintaining model quality. Enabled by optimized GPU kernels for column-sparse attention, PulseCol delivers up to 1.95

\times

end-to-end speedup over FlashAttention across several context lengths.

Yanyi Lyu, Letian Chen, Futing Sun +3

Harbin Institute of Technology (Shenzhen)

Self-Attention Diffusion Language Model Inference Sparse Attention Diffusion Model Inference Acceleration

May 19, 2026cs.CV

Efficient Long-Context Modeling in Diffusion Language Models via Block Approximate Sparse Attention

Diffusion Language Models (DLMs) enable globally coherent, bidirectional, and controllable text generation, offering advantages over traditional autoregressive LLMs, while scaling to ultra-long sequences remains costly. Many existing block-sparse attention methods select blocks by fixed sampling patterns over the high-resolution attention space, such as tail regions or anti-diagonal stripes. Such prior-driven sampling can miss salient tokens and introduce instability under distribution shifts. In this paper, we propose the Block Approximate Sparse Attention framework (BA-Att) with block-wise pre-downsampled operation, which identifies informative regions within a compact downsampled space, avoiding reliance on brittle positional priors. To analyze its theoretical behavior, we define an oracle post-downsample attention map and formalize the approximation error between pre- and post-downsample schemes. Based on this insight, we introduce a lightweight norm-sorting module and a covariance-compensated correction that approximates full covariance using diagonal QK variances, reducing computational complexity. Extensive experiments show that our operator achieves up to 6.95x acceleration over FlashAttention in attention computation, and maintains near full-attention performance at 50% sparsity across language models, multimodal language models, and video generation models, demonstrating strong efficiency and generalization.

Wenhu Zhang, Yiming Wu, Huanyu Wang +6

The Hong Kong University of Science and Technology · The University of Hong Kong · Zhejiang University +1

Self-Attention Long-Context Language Modeling Sparse Attention Diffusion Language Models Diffusion Model Inference Acceleration Block-Sparse Attention