Yifei Cheng

Yifei Cheng, Xianglin Yang, Guoxia Wang et al.

Sharpness-aware minimization (SAM) seeks the minima with a flat loss landscape to improve the generalization performance in machine learning tasks, including fine-tuning. However, its extra parameter perturbation step doubles the computation cost, which becomes the bottleneck of SAM in the practical implementation. In this work, we propose an approach SL-SAM to break this bottleneck by introducing the sparse technique to layers. Our key innovation is to frame the dynamic selection of layers for both the gradient ascent (perturbation) and descent (update) steps as a multi-armed bandit problem. At the beginning of each iteration, SL-SAM samples a part of the layers of the model according to the gradient norm to participate in the backpropagation of the following parameter perturbation and update steps, thereby reducing the computation complexity. We then provide the analysis to guarantee the convergence of SL-SAM. In the experiments of fine-tuning models in several tasks, SL-SAM achieves the performances comparable to the state-of-the-art baselines, including a \#1 rank on LLM fine-tuning. Meanwhile, SL-SAM significantly reduces the ratio of active parameters in backpropagation compared to vanilla SAM (SL-SAM activates 47\%, 22\% and 21\% parameters on the vision, moderate and large language model respectively while vanilla SAM always activates 100\%), verifying the efficiency of our proposed algorithm.

Joya Chen, Dong Liu, Tong Xu et al.

To train accurate deep object detectors under the extreme foreground-background imbalance, heuristic sampling methods are always necessary, which either re-sample a subset of all training samples (hard sampling methods, \eg biased sampling, OHEM), or use all training samples but re-weight them discriminatively (soft sampling methods, \eg Focal Loss, GHM). In this paper, we challenge the necessity of such hard/soft sampling methods for training accurate deep object detectors. While previous studies have shown that training detectors without heuristic sampling methods would significantly degrade accuracy, we reveal that this degradation comes from an unreasonable classification gradient magnitude caused by the imbalance, rather than a lack of re-sampling/re-weighting. Motivated by our discovery, we propose a simple yet effective \emph{Sampling-Free} mechanism to achieve a reasonable classification gradient magnitude by initialization and loss scaling. Unlike heuristic sampling methods with multiple hyperparameters, our Sampling-Free mechanism is fully data diagnostic, without laborious hyperparameters searching. We verify the effectiveness of our method in training anchor-based and anchor-free object detectors, where our method always achieves higher detection accuracy than heuristic sampling methods on COCO and PASCAL VOC datasets. Our Sampling-Free mechanism provides a new perspective to address the foreground-background imbalance. Our code is released at \url{https://github.com/ChenJoya/sampling-free}.

Yifei Cheng, Li Shen, Hao Sun et al.

Sharpness-Aware Minimization (SAM) optimizer enhances the generalization ability of the machine learning model by exploring the flat minima landscape through weight perturbations. Despite its empirical success, SAM introduces an additional hyper-parameter, the perturbation radius, which causes the sensitivity of SAM to it. Moreover, it has been proved that the perturbation radius and learning rate of SAM are constrained by problem-dependent parameters to guarantee convergence. These limitations indicate the requirement of parameter-tuning in practical applications. In this paper, we propose the algorithm LightSAM which sets the perturbation radius and learning rate of SAM adaptively, thus extending the application scope of SAM. LightSAM employs three popular adaptive optimizers, including AdaGrad-Norm, AdaGrad and Adam, to replace the SGD optimizer for weight perturbation and model updating, reducing sensitivity to parameters. Theoretical results show that under weak assumptions, LightSAM could converge ideally with any choices of perturbation radius and learning rate, thus achieving parameter-agnostic. We conduct preliminary experiments on several deep learning tasks, which together with the theoretical findings validate the the effectiveness of LightSAM.

Shuheng Shen, Yifei Cheng, Jingchang Liu et al.

Distributed parallel stochastic gradient descent algorithms are workhorses for large scale machine learning tasks. Among them, local stochastic gradient descent (Local SGD) has attracted significant attention due to its low communication complexity. Previous studies prove that the communication complexity of Local SGD with a fixed or an adaptive communication period is in the order of $O (N^{\frac{3}{2}} T^{\frac{1}{2}})$ and $O (N^{\frac{3}{4}} T^{\frac{3}{4}})$ when the data distributions on clients are identical (IID) or otherwise (Non-IID), where $N$ is the number of clients and $T$ is the number of iterations. In this paper, to accelerate the convergence by reducing the communication complexity, we propose \textit{ST}agewise \textit{L}ocal \textit{SGD} (STL-SGD), which increases the communication period gradually along with decreasing learning rate. We prove that STL-SGD can keep the same convergence rate and linear speedup as mini-batch SGD. In addition, as the benefit of increasing the communication period, when the objective is strongly convex or satisfies the Polyak-Łojasiewicz condition, the communication complexity of STL-SGD is $O (N \log{T})$ and $O (N^{\frac{1}{2}} T^{\frac{1}{2}})$ for the IID case and the Non-IID case respectively, achieving significant improvements over Local SGD. Experiments on both convex and non-convex problems demonstrate the superior performance of STL-SGD.

Xianfeng Liang, Shuheng Shen, Jingchang Liu et al.

To accelerate the training of machine learning models, distributed stochastic gradient descent (SGD) and its variants have been widely adopted, which apply multiple workers in parallel to speed up training. Among them, Local SGD has gained much attention due to its lower communication cost. Nevertheless, when the data distribution on workers is non-identical, Local SGD requires $O(T^{\frac{3}{4}} N^{\frac{3}{4}})$ communications to maintain its \emph{linear iteration speedup} property, where $T$ is the total number of iterations and $N$ is the number of workers. In this paper, we propose Variance Reduced Local SGD (VRL-SGD) to further reduce the communication complexity. Benefiting from eliminating the dependency on the gradient variance among workers, we theoretically prove that VRL-SGD achieves a \emph{linear iteration speedup} with a lower communication complexity $O(T^{\frac{1}{2}} N^{\frac{3}{2}})$ even if workers access non-identical datasets. We conduct experiments on three machine learning tasks, and the experimental results demonstrate that VRL-SGD performs impressively better than Local SGD when the data among workers are quite diverse.

Shuheng Shen, Linli Xu, Jingchang Liu et al.

With the increase in the amount of data and the expansion of model scale, distributed parallel training becomes an important and successful technique to address the optimization challenges. Nevertheless, although distributed stochastic gradient descent (SGD) algorithms can achieve a linear iteration speedup, they are limited significantly in practice by the communication cost, making it difficult to achieve a linear time speedup. In this paper, we propose a computation and communication decoupled stochastic gradient descent (CoCoD-SGD) algorithm to run computation and communication in parallel to reduce the communication cost. We prove that CoCoD-SGD has a linear iteration speedup with respect to the total computation capability of the hardware resources. In addition, it has a lower communication complexity and better time speedup comparing with traditional distributed SGD algorithms. Experiments on deep neural network training demonstrate the significant improvements of CoCoD-SGD: when training ResNet18 and VGG16 with 16 Geforce GTX 1080Ti GPUs, CoCoD-SGD is up to 2-3$\times$ faster than traditional synchronous SGD.