Subquadratic Algorithms and Hardness for Attention with Any Temperature

• URL: http://arxiv.org/abs/2505.14840v1
• Date: Tue, 20 May 2025 19:12:43 GMT
• Title: Subquadratic Algorithms and Hardness for Attention with Any Temperature
• Authors: Shreya Gupta, Boyang Huang, Barna Saha, Yinzhan Xu, Christopher Ye,
• Abstract summary: We characterize and characterize when fast Attention for arbitrary temperatures is possible.<n>We show that any substantial improvement on our algorithm is unlikely.<n>In particular, we show that even when $d = 2Theta(log* n)$, Attention requires $n2 - o(1)$ time under $mathsfSETH$.
• Score: 4.1906341910583045
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Despite the popularity of the Transformer architecture, the standard algorithm for computing Attention suffers from quadratic time complexity in context length $n$. Alman and Song [NeurIPS 2023] showed that when the head dimension $d = \Theta(\log n)$, subquadratic Attention is possible if and only if the inputs have small entries bounded by $B = o(\sqrt{\log n})$ in absolute values, under the Strong Exponential Time Hypothesis ($\mathsf{SETH}$). Equivalently, subquadratic Attention is possible if and only if the softmax is applied with high temperature for $d=\Theta(\log n)$. Running times of these algorithms depend exponentially on $B$ and thus they do not lead to even a polynomial-time algorithm outside the specific range of $B$. This naturally leads to the question: when can Attention be computed efficiently without strong assumptions on temperature? Are there fast attention algorithms that scale polylogarithmically with entry size $B$? In this work, we resolve this question and characterize when fast Attention for arbitrary temperatures is possible. First, for all constant $d = O(1)$, we give the first subquadratic $\tilde{O}(n^{2 - 1/d} \cdot \mathrm{polylog}(B))$ time algorithm for Attention with large $B$. Our result holds even for matrices with large head dimension if they have low rank. In this regime, we also give a similar running time for Attention gradient computation, and therefore for the full LLM training process. Furthermore, we show that any substantial improvement on our algorithm is unlikely. In particular, we show that even when $d = 2^{\Theta(\log^* n)}$, Attention requires $n^{2 - o(1)}$ time under $\mathsf{SETH}$. Finally, in the regime where $d = \mathrm{poly}(n)$, we show that the standard algorithm is optimal under popular fine-grained complexity assumptions.

Related papers

• Private Continual Counting of Unbounded Streams [11.941250828872189]
  We study the problem of differentially private continual counting in the unbounded setting where the input size $n$ is not known in advance.<n>Using the common doubling trick' avoids knowledge of $n$ but leads to suboptimal and non-smooth error.<n>We introduce novel matrix factorizations based on logarithmic perturbations of the function $frac1sqrt1-z$ studied in prior works.
  arXiv  Detail & Related papers  (2025-06-17T23:09:53Z)
• Improved Robust Estimation for Erdős-Rényi Graphs: The Sparse Regime and Optimal Breakdown Point [3.793609515750114]
  We study the problem of robustly estimating the edge density of ErdHos-R'enyi random graphs $G(n, dcirc/n)$.<n>Our algorithm is based on the sum-of-squares hierarchy.
  arXiv  Detail & Related papers  (2025-03-05T21:45:17Z)
• Optimal Sketching for Residual Error Estimation for Matrix and Vector Norms [50.15964512954274]
  We study the problem of residual error estimation for matrix and vector norms using a linear sketch. We demonstrate that this gives a substantial advantage empirically, for roughly the same sketch size and accuracy as in previous work. We also show an $Omega(k2/pn1-2/p)$ lower bound for the sparse recovery problem, which is tight up to a $mathrmpoly(log n)$ factor.
  arXiv  Detail & Related papers  (2024-08-16T02:33:07Z)
• A Sub-Quadratic Time Algorithm for Robust Sparse Mean Estimation [6.853165736531941]
  We study the algorithmic problem of sparse mean estimation in the presence of adversarial outliers. Our main contribution is an algorithm for robust sparse mean estimation which runs in emphsubquadratic time using $mathrmpoly(k,log d,1/epsilon)$ samples.
  arXiv  Detail & Related papers  (2024-03-07T18:23:51Z)
• How to Capture Higher-order Correlations? Generalizing Matrix Softmax Attention to Kronecker Computation [12.853829771559916]
  We study a generalization of attention which captures triple-wise correlations. This generalization is able to solve problems about detecting triple-wise connections that were shown to be impossible for transformers. We show that our construction, algorithms, and lower bounds naturally generalize to higher-order tensors and correlations.
  arXiv  Detail & Related papers  (2023-10-06T07:42:39Z)
• Fast $(1+\varepsilon)$-Approximation Algorithms for Binary Matrix Factorization [54.29685789885059]
  We introduce efficient $(1+varepsilon)$-approximation algorithms for the binary matrix factorization (BMF) problem. The goal is to approximate $mathbfA$ as a product of low-rank factors. Our techniques generalize to other common variants of the BMF problem.
  arXiv  Detail & Related papers  (2023-06-02T18:55:27Z)
• Fast Attention Requires Bounded Entries [19.17278873525312]
  inner product attention computation is a fundamental task for training large language models such as Transformer, GPT-1, BERT, GPT-2, GPT-3 and ChatGPT. We investigate whether faster algorithms are possible by implicitly making use of the matrix $A$. This gives a theoretical explanation for the phenomenon observed in practice that attention computation is much more efficient when the input matrices have smaller entries.
  arXiv  Detail & Related papers  (2023-02-26T02:42:39Z)
• Sketching Algorithms and Lower Bounds for Ridge Regression [65.0720777731368]
  We give a sketching-based iterative algorithm that computes $1+varepsilon$ approximate solutions for the ridge regression problem. We also show that this algorithm can be used to give faster algorithms for kernel ridge regression.
  arXiv  Detail & Related papers  (2022-04-13T22:18:47Z)
• Low-degree learning and the metric entropy of polynomials [44.99833362998488]
  We prove that any (deterministic or randomized) algorithm which learns $mathscrF_nd$ with $L$-accuracy $varepsilon$ requires at least $Omega(sqrtvarepsilon)2dlog n leq log mathsfM(mathscrF_n,d,|cdot|_L,varepsilon) satisfies the two-sided estimate $$c (1-varepsilon)2dlog
  arXiv  Detail & Related papers  (2022-03-17T23:52:08Z)
• Logarithmic Regret from Sublinear Hints [76.87432703516942]
  We show that an algorithm can obtain $O(log T)$ regret with just $O(sqrtT)$ hints under a natural query model. We also show that $o(sqrtT)$ hints cannot guarantee better than $Omega(sqrtT)$ regret.
  arXiv  Detail & Related papers  (2021-11-09T16:50:18Z)
• Online Sub-Sampling for Reinforcement Learning with General Function Approximation [111.01990889581243]
  In this paper, we establish an efficient online sub-sampling framework that measures the information gain of data points collected by an RL algorithm. For a value-based method with complexity-bounded function class, we show that the policy only needs to be updated for $proptooperatornamepolylog(K)$ times. In contrast to existing approaches that update the policy for at least $Omega(K)$ times, our approach drastically reduces the number of optimization calls in solving for a policy.
  arXiv  Detail & Related papers  (2021-06-14T07:36:25Z)
• Learning a Latent Simplex in Input-Sparsity Time [58.30321592603066]
  We consider the problem of learning a latent $k$-vertex simplex $KsubsetmathbbRdtimes n$, given access to $AinmathbbRdtimes n$. We show that the dependence on $k$ in the running time is unnecessary given a natural assumption about the mass of the top $k$ singular values of $A$.
  arXiv  Detail & Related papers  (2021-05-17T16:40:48Z)
• Streaming Complexity of SVMs [110.63976030971106]
  We study the space complexity of solving the bias-regularized SVM problem in the streaming model. We show that for both problems, for dimensions of $frac1lambdaepsilon$, one can obtain streaming algorithms with spacely smaller than $frac1lambdaepsilon$.
  arXiv  Detail & Related papers  (2020-07-07T17:10:00Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.