We propose universally slimmable self-supervised learning (dubbed as US3L) to achieve better accuracy-efficiency trade-offs for deploying self-supervised models across different devices. We observe that direct adaptation of self-supervised learning (SSL) to universally slimmable networks misbehaves as the training process frequently collapses. We then discover that temporal consistent guidance is the key to the success of SSL for universally slimmable networks, and we propose three guidelines for the loss design to ensure this temporal consistency from a unified gradient perspective. Moreover, we propose dynamic sampling and group regularization strategies to simultaneously improve training efficiency and accuracy. Our US3L method has been empirically validated on both convolutional neural networks and vision transformers. With only once training and one copy of weights, our method outperforms various state-of-the-art methods (individually trained or not) on benchmarks including recognition, object detection and instance segmentation. Our code is available at https://github.com/megvii-research/US3L-CVPR2023.

Compressing deep neural networks (DNNs) is important for real-world applications operating on resource-constrained devices. However, it is difficult to change the model size once the training is completed, which needs retraining to configure models suitable for different devices. In this paper, we propose a novel method that enables DNNs to flexibly change their size after training. We factorize the weight matrices of the DNNs via singular value decomposition (SVD) and change their ranks according to the target size. In contrast with existing methods, we introduce simple criteria that characterize the importance of each basis and layer, which enables to effectively compress the error and complexity of models as little as possible. In experiments on multiple image-classification tasks, our method exhibits favorable performance compared with other methods. As part of the great progress made in deep learning, deep neural network (DNN) models with higher performance have been proposed for various machine-learning tasks (LeCun et al., 2015). However, these performance improvements require a higher number of parameters and greater computational complexity. Therefore, it is important to compress them without sacrificing the performance for running the models on resource-constrained devices. Han et al. (2016) reduced the memory requirement for devices by pruning and quantizing weight coefficients after training the models.

Slimmable networks are a family of neural networks that can instantly adjust the runtime width. The width can be chosen from a predefined widths set to adaptively optimize accuracy-efficiency trade-offs at runtime. In this work, we propose a systematic approach to train universally slimmable networks (US-Nets), extending slimmable networks to execute at arbitrary width, and generalizing to networks both with and without batch normalization layers. We further propose two improved training techniques for US-Nets, named the sandwich rule and inplace distillation, to enhance training process and boost testing accuracy. We show improved performance of universally slimmable MobileNet v1 and MobileNet v2 on ImageNet classification task, compared with individually trained ones and 4-switch slimmable network baselines. We also evaluate the proposed US-Nets and improved training techniques on tasks of image super-resolution and deep reinforcement learning. Extensive ablation experiments on these representative tasks demonstrate the effectiveness of our proposed methods. Our discovery opens up the possibility to directly evaluate FLOPs-Accuracy spectrum of network architectures. Code and models will be available at: https://github.com/JiahuiYu/slimmable_networks