Weight decay induces low-rank attention layers

• URL: http://arxiv.org/abs/2410.23819v1
• Date: Thu, 31 Oct 2024 11:04:07 GMT
• Title: Weight decay induces low-rank attention layers
• Abstract summary: We study the influence of weight decay as well as $L2$-regularization when training neural network models. We verify that the key-query and value-projection matrix products $W_KTW_Q, PW_V$ within attention layers, when optimized with weight decay, indeed induce a significant reduction in the rank.
• Score: 8.746774382296008
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The effect of regularizers such as weight decay when training deep neural networks is not well understood. We study the influence of weight decay as well as $L2$-regularization when training neural network models in which parameter matrices interact multiplicatively. This combination is of particular interest as this parametrization is common in attention layers, the workhorse of transformers. Here, key-query, as well as value-projection parameter matrices, are multiplied directly with each other: $W_K^TW_Q$ and $PW_V$. We extend previous results and show on one hand that any local minimum of a $L2$-regularized loss of the form $L(AB^\top) + \lambda (\|A\|^2 + \|B\|^2)$ coincides with a minimum of the nuclear norm-regularized loss $L(AB^\top) + \lambda\|AB^\top\|_*$, and on the other hand that the 2 losses become identical exponentially quickly during training. We thus complement existing works linking $L2$-regularization with low-rank regularization, and in particular, explain why such regularization on the matrix product affects early stages of training. Based on these theoretical insights, we verify empirically that the key-query and value-projection matrix products $W_K^TW_Q, PW_V$ within attention layers, when optimized with weight decay, as usually done in vision tasks and language modelling, indeed induce a significant reduction in the rank of $W_K^TW_Q$ and $PW_V$, even in fully online training. We find that, in accordance with existing work, inducing low rank in attention matrix products can damage language model performance, and observe advantages when decoupling weight decay in attention layers from the rest of the parameters.

Related papers

• Bug or Feature$^2$: Weight Drift, Activation Sparsity and Spikes [53.726365933748134]
  We analyze a negative weight drift induced by the interaction between standard losses and positively biased activation functions.<n>We characterize the sparsity-accuracy tradeoff across 79 configurations and identify a sharp accuracy cliff above $sim$70% activation sparsity.
  arXiv  Detail & Related papers  (2026-05-17T21:29:20Z)
• Wiener Chaos Expansion based Neural Operator for Singular Stochastic Partial Differential Equations [58.188204772101756]
  This paper explores how our recently developed Wiener Chaos Expansion (WCE)-based neural operator (NO) can be applied to singular partial differential equations.<n>We leverage feature-wise linear modulation (FiLM) to appropriately capture the dependency between the solution of singular SPDE and its smooth remainder.<n>The resulting WCE-FiLM-NO shows excellent performance on $boldsymbol4$, as measured by relative $L$ loss, out-of-distribution and autocorrelation score.
  arXiv  Detail & Related papers  (2026-03-09T10:50:30Z)
• Evolution Strategies at the Hyperscale [57.75314521465674]
  We introduce EGGROLL, an evolution strategies (ES) algorithm designed to scale backprop-free optimization to large population sizes.<n>ES is a set of powerful blackbox optimisation methods that can handle non-differentiable or noisy objectives.<n>EGGROLL overcomes these bottlenecks by generating random matrices $Ain mathbbRmtimes r, Bin mathbbRntimes r$ with $rll min(m,n)$ to form a low-rank matrix perturbation $A Btop$
  arXiv  Detail & Related papers  (2025-11-20T18:56:05Z)
• Generalization Bounds for Rank-sparse Neural Networks [40.956710248523294]
  We prove generalization bounds for neural networks which exploit the approximate low rank structure of the weight matrices if present.<n>As $p$ increases, the bound behaves more like a norm-based bound instead.
  arXiv  Detail & Related papers  (2025-10-24T18:19:32Z)
• Robust Layerwise Scaling Rules by Proper Weight Decay Tuning [50.11170157029911]
  In modern scale-invariant architectures, training quickly enters an degrading-governed steady state.<n>We introduce a weight-decay scaling rule for AdamW that preserves sublayer gain across widths.<n>Our results extend $mu$P beyond the near-init regime by explicitly controlling the steady-state scales set by parameters.
  arXiv  Detail & Related papers  (2025-10-17T02:58:35Z)
• Sparse Representations Improve Adversarial Robustness of Neural Network Classifiers [0.0]
  Deep neural networks perform remarkably well on image classification tasks but remain vulnerable to carefully crafted adversarial perturbations.<n>This work revisits linear dimensionality reduction as a simple, data-adapted defense.<n> Empirically, with a small non-linear network after the projection, SPCA consistently degrades more gracefully than PCA under strong white-box and black-box attacks.
  arXiv  Detail & Related papers  (2025-09-25T13:21:22Z)
• Tensor Decomposition Networks for Fast Machine Learning Interatomic Potential Computations [63.945006006152035]
  tensor decomposition networks (TDNs) achieve competitive performance with dramatic speedup in computations.<n>We evaluate TDNs on PubChemQCR, a newly curated molecular relaxation dataset containing 105 million DFT-calculated snapshots.
  arXiv  Detail & Related papers  (2025-07-01T18:46:27Z)
• FedSVD: Adaptive Orthogonalization for Private Federated Learning with LoRA [61.79405341803085]
  Low-Rank Adaptation (LoRA) is widely used for efficient fine-tuning of language models in federated learning (FL)<n>Low-Rank Adaptation (LoRA) is widely used for efficient fine-tuning of language models in federated learning (FL)
  arXiv  Detail & Related papers  (2025-05-19T07:32:56Z)
• Theoretical Insights into Fine-Tuning Attention Mechanism: Generalization and Optimization [27.907707931902547]
  We investigate two phenomena related to the attention mechanism during the fine-tuning of Large Language Models. The first phenomenon, termed "Unequal Importance of Attention Matrices," highlights the impact of fine-tuning different weight matrices. The second phenomenon, "Attention Matrices with Customized Learning Rate Leads to Better Convergence," emphasizes the importance of assigning distinct learning rates.
  arXiv  Detail & Related papers  (2024-10-03T06:37:37Z)
• Unveiling Induction Heads: Provable Training Dynamics and Feature Learning in Transformers [54.20763128054692]
  We study how a two-attention-layer transformer is trained to perform ICL on $n$-gram Markov chain data. We prove that the gradient flow with respect to a cross-entropy ICL loss converges to a limiting model.
  arXiv  Detail & Related papers  (2024-09-09T18:10:26Z)
• On Characterizing and Mitigating Imbalances in Multi-Instance Partial Label Learning [57.18649648182171]
  We make contributions towards addressing a problem that hasn't been studied so far in the context of MI-PLL. We derive class-specific risk bounds for MI-PLL, while making minimal assumptions. Our theory reveals a unique phenomenon: that $sigma$ can greatly impact learning imbalances.
  arXiv  Detail & Related papers  (2024-07-13T20:56:34Z)
• Decoupled Weight Decay for Any $p$ Norm [1.1510009152620668]
  We consider a simple yet effective approach to sparsification, based on the Bridge, $L_p$ regularization during training. We introduce a novel weight decay scheme, which generalizes the standard $L$ weight decay to any $p$ norm. We empirically demonstrate that it leads to highly sparse networks, while maintaining performance comparable to standard $L$ regularization.
  arXiv  Detail & Related papers  (2024-04-16T18:02:15Z)
• Asymmetry in Low-Rank Adapters of Foundation Models [47.310550805920585]
  This paper characterizes and leverages unexpected asymmetry in the importance of low-rank adapter matrices. We show that fine-tuning $B$ is inherently more effective than fine-tuning $A$, and that a random untrained $A$ should perform nearly as well as a fine-tuned one.
  arXiv  Detail & Related papers  (2024-02-26T18:59:12Z)
• A Unified Scheme of ResNet and Softmax [8.556540804058203]
  We provide a theoretical analysis of the regression problem: $| langle exp(Ax) + A x, bf 1_n rangle-1 ( exp(Ax) + Ax ) This regression problem is a unified scheme that combines softmax regression and ResNet, which has never been done before.
  arXiv  Detail & Related papers  (2023-09-23T21:41:01Z)
• Generalization and Stability of Interpolating Neural Networks with Minimal Width [37.908159361149835]
  We investigate the generalization and optimization of shallow neural-networks trained by gradient in the interpolating regime. We prove the training loss number minimizations $m=Omega(log4 (n))$ neurons and neurons $Tapprox n$. With $m=Omega(log4 (n))$ neurons and $Tapprox n$, we bound the test loss training by $tildeO (1/)$.
  arXiv  Detail & Related papers  (2023-02-18T05:06:15Z)
• Implicit Regularization Leads to Benign Overfitting for Sparse Linear Regression [16.551664358490658]
  In deep learning, often the training process finds an interpolator (a solution with 0 training loss) but the test loss is still low. One common mechanism for benign overfitting is implicit regularization, where the training process leads to additional properties for the interpolator. We show that training our new model via gradient descent leads to an interpolator with near-optimal test loss.
  arXiv  Detail & Related papers  (2023-02-01T05:41:41Z)
• Nonparametric regression with modified ReLU networks [77.34726150561087]
  We consider regression estimation with modified ReLU neural networks in which network weight matrices are first modified by a function $alpha$ before being multiplied by input vectors.
  arXiv  Detail & Related papers  (2022-07-17T21:46:06Z)
• Bounding the Width of Neural Networks via Coupled Initialization -- A Worst Case Analysis [121.9821494461427]
  We show how to significantly reduce the number of neurons required for two-layer ReLU networks. We also prove new lower bounds that improve upon prior work, and that under certain assumptions, are best possible.
  arXiv  Detail & Related papers  (2022-06-26T06:51:31Z)
• Training Multi-Layer Over-Parametrized Neural Network in Subquadratic Time [12.348083977777833]
  We consider the problem of training a multi-layer over-parametrized neural network to minimize the empirical risk induced by a loss function. In this work, we show how to reduce the training cost per iteration.
  arXiv  Detail & Related papers  (2021-12-14T18:13:36Z)
• Towards an Understanding of Benign Overfitting in Neural Networks [104.2956323934544]
  Modern machine learning models often employ a huge number of parameters and are typically optimized to have zero training loss. We examine how these benign overfitting phenomena occur in a two-layer neural network setting. We show that it is possible for the two-layer ReLU network interpolator to achieve a near minimax-optimal learning rate.
  arXiv  Detail & Related papers  (2021-06-06T19:08:53Z)
• Large-time asymptotics in deep learning [0.0]
  We consider the impact of the final time $T$ (which may indicate the depth of a corresponding ResNet) in training. For the classical $L2$--regularized empirical risk minimization problem, we show that the training error is at most of the order $mathcalOleft(frac1Tright)$. In the setting of $ellp$--distance losses, we prove that both the training error and the optimal parameters are at most of the order $mathcalOleft(e-mu
  arXiv  Detail & Related papers  (2020-08-06T07:33:17Z)

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.