J. S. Deng

Jiaxin Deng, Junbiao Pang, Baochang Zhang et al.

Sharpness-Aware Minimization (SAM) enhances generalization by reducing a Max-Sharpness (MaxS). Despite the practical success, we empirically found that the MAxS behind SAM's generalization enhancements face the "Flatness Indicator Problem" (FIP), where SAM only considers the flatness in the direction of gradient ascent, resulting in a next minimization region that is not sufficiently flat. A better Flatness Indicator (FI) would bring a better generalization of neural networks. Because SAM is a greedy search method in nature. In this paper, we propose to utilize the difference between the training loss and the minimum loss over the neighborhood surrounding the current weight, which we denote as Min-Sharpness (MinS). By merging MaxS and MinS, we created a better FI that indicates a flatter direction during the optimization. Specially, we combine this FI with SAM into the proposed Bilateral SAM (BSAM) which finds a more flatter minimum than that of SAM. The theoretical analysis proves that BSAM converges to local minima. Extensive experiments demonstrate that BSAM offers superior generalization performance and robustness compared to vanilla SAM across various tasks, i.e., classification, transfer learning, human pose estimation, and network quantization. Code is publicly available at: https://github.com/ajiaaa/BSAM.

Jiaxin Deng, Shiyao Wang, Song Lu et al.

State-of-the-art sequential recommendation models heavily rely on transformer's attention mechanism. However, the quadratic computational and memory complexities of self attention have limited its scalability for modeling users' long range behaviour sequences. To address this problem, we propose ELASTIC, an Efficient Linear Attention for SequenTial Interest Compression, requiring only linear time complexity and decoupling model capacity from computational cost. Specifically, ELASTIC introduces a fixed length interest experts with linear dispatcher attention mechanism which compresses the long-term behaviour sequences to a significantly more compact representation which reduces up to 90% GPU memory usage with x2.7 inference speed up. The proposed linear dispatcher attention mechanism significantly reduces the quadratic complexity and makes the model feasible for adequately modeling extremely long sequences. Moreover, in order to retain the capacity for modeling various user interests, ELASTIC initializes a vast learnable interest memory bank and sparsely retrieves compressed user's interests from the memory with a negligible computational overhead. The proposed interest memory retrieval technique significantly expands the cardinality of available interest space while keeping the same computational cost, thereby striking a trade-off between recommendation accuracy and efficiency. To validate the effectiveness of our proposed ELASTIC, we conduct extensive experiments on various public datasets and compare it with several strong sequential recommenders. Experimental results demonstrate that ELASTIC consistently outperforms baselines by a significant margin and also highlight the computational efficiency of ELASTIC when modeling long sequences. We will make our implementation code publicly available.

Jiaxin Deng, Junbiao Pang

Sharpness-Aware Minimization (SAM) improves model generalization but doubles the computational cost of Stochastic Gradient Descent (SGD) by requiring twice the gradient calculations per optimization step. To mitigate this, we propose Adaptively sampling-Reusing-mixing decomposed gradients to significantly accelerate SAM (ARSAM). Concretely, we firstly discover that SAM's gradient can be decomposed into the SGD gradient and the Projection of the Second-order gradient onto the First-order gradient (PSF). Furthermore, we observe that the SGD gradient and PSF dynamically evolve during training, emphasizing the growing role of the PSF to achieve a flat minima. Therefore, ARSAM is proposed to the reused PSF and the timely updated PSF still maintain the model's generalization ability. Extensive experiments show that ARSAM achieves state-of-the-art accuracies comparable to SAM across diverse network architectures. On CIFAR-10/100, ARSAM is comparable to SAM while providing a speedup of about 40\%. Moreover, ARSAM accelerates optimization for the various challenge tasks (\textit{e.g.}, human pose estimation, and model quantization) without sacrificing performance, demonstrating its broad practicality.% The code is publicly accessible at: https://github.com/ajiaaa/ARSAM.

Jiaxin Deng, Junbiao Pang, Baochang Zhang et al.

Sharpness-aware Minimization (SAM) has been proposed recently to improve model generalization ability. However, SAM calculates the gradient twice in each optimization step, thereby doubling the computation costs compared to stochastic gradient descent (SGD). In this paper, we propose a simple yet efficient sampling method to significantly accelerate SAM. Concretely, we discover that the gradient of SAM is a combination of the gradient of SGD and the Projection of the Second-order gradient matrix onto the First-order gradient (PSF). PSF exhibits a gradually increasing frequency of change during the training process. To leverage this observation, we propose an adaptive sampling method based on the variation of PSF, and we reuse the sampled PSF for non-sampling iterations. Extensive empirical results illustrate that the proposed method achieved state-of-the-art accuracies comparable to SAM on diverse network architectures.

Jiaxin Deng, Qingcheng Zhu, Junbiao Pang et al.

Little research explores the correlation between the expressive ability and generalization ability of the low-rank adaptation (LoRA). Sharpness-Aware Minimization (SAM) improves model generalization for both Convolutional Neural Networks (CNNs) and Transformers by encouraging convergence to locally flat minima. However, the connection between sharpness and generalization has not been fully explored for LoRA due to the lack of tools to either empirically seek flat minima or develop theoretical methods. In this work, we propose Flat-LoRA and its efficient version i.e., EFlat-LoRA, to seek flat minima for LoRA. Concretely, we theoretically demonstrate that perturbations in the full parameter space can be transferred to the low-rank subspace. This approach eliminates the potential interference introduced by perturbations across multiple matrices in the low-rank subspace. Our extensive experiments on large language models and vision-language models demonstrate that EFlat-LoRA achieves optimize efficiency comparable to that of LoRA while simultaneously attaining comparable or even better performance. For example, on the GLUE dataset with RoBERTa-large, EFlat-LoRA outperforms LoRA and full fine-tuning by 1.0% and 0.5% on average, respectively. On vision-language models e.g., Qwen-VL-Chat shows performance improvements of 1.5% and 1.0% on SQA and VizWiz datasets, respectively. These empirical results also verify that the generalization of LoRA is closely related to sharpness, which is omitted by previous methods.

Jiaxin Deng, Junbiao Pang, Baochang Zhang

Sharpness-Aware Minimization (SAM) has emerged as a promising approach for effectively reducing the generalization error. However, SAM incurs twice the computational cost compared to base optimizer (e.g., SGD). We propose Asymptotic Unbiased Sampling with respect to iterations to accelerate SAM (AUSAM), which maintains the model's generalization capacity while significantly enhancing computational efficiency. Concretely, we probabilistically sample a subset of data points beneficial for SAM optimization based on a theoretically guaranteed criterion, i.e., the Gradient Norm of each Sample (GNS). We further approximate the GNS by the difference in loss values before and after perturbation in SAM. As a plug-and-play, architecture-agnostic method, our approach consistently accelerates SAM across a range of tasks and networks, i.e., classification, human pose estimation and network quantization. On CIFAR10/100 and Tiny-ImageNet, AUSAM achieves results comparable to SAM while providing a speedup of over 70%. Compared to recent dynamic data pruning methods, AUSAM is better suited for SAM and excels in maintaining performance. Additionally, AUSAM accelerates optimization in human pose estimation and model quantization without sacrificing performance, demonstrating its broad practicality.