Network pruning is widely used in model compression due to its simplicity and efficiency. Existing methods typically introduce sparse loss regularization to learn masks. However, this sparse regularization approach lacks a clear criterion for evaluating channel importance and relies on manually defined rules, leading to a decline in model performance. In this article, a Self-Supervised Mask Learning (SSML) method for global channel pruning is proposed, casting mask learning as a self-supervised binary classification task to automatically identify less important channels. Specifically, a dedicated pretext task is designed for the channelwise masks, which leverages the original network to generate pseudo-labels from the mask itself to guide mask learning. Then, a polarization mask loss function is proposed, transforming the discrete mask learning problem into a differentiable binary classification problem. The proposed loss function distinguishes the similarity between pseudo-labels and masks, clustering similar masks together in the feature space and separating dissimilar masks, ultimately allowing channels with masks of 0 to be safely removed without damaging the performance of the pruned model. In addition, SSML can train from scratch to yield a compact model. Extensive experiments on CIFAR-10, CIFAR-100 and ImageNet datasets demonstrate that SSML outperforms state-of-the-art methods. For instance, SSML prunes 52.7% of the FLOPs of ResNe34 on the ImageNet dataset with only 0.01% drop in Top-1 accuracy. Moreover, the generalization of SSML is verified on downstream tasks.

Channel pruning has been demonstrated as a highly effective approach to compress large convolutional neural networks. Existing differentiable channel pruning methods usually use deterministic soft masks to scale the channelwise outputs and explore an appropriate threshold on the masks to remove unimportant channels, which sometimes causes unexpected damage to the network accuracy when there are no sweet spots that clearly separate important channels from redundant ones. In this article, we introduce a new differentiable channel pruning method based on polarization of probabilistic channelwise soft masks (PPSMs). We use variational inference to approximate the posterior distributions of the masks and simultaneously exploit a polarization regularization to push the probabilistic masks towards either 0 or 1; thus, the channels with near-zero masks can be safely eliminated with little hurt on network accuracy. Our method significantly relieves the difficulty faced by the existing methods to find an appropriate threshold on the masks. The joint inference and polarization of probabilistic soft masks enable PPSM to yield better pruning results than the state of the arts. For instance, our method prunes 65.91% FLOPs of ResNet50 on the ImageNet dataset with only 0.7% model accuracy degradation.

Multi-modal Magnetic Resonance Imaging (MRI) provides complementary information for clinical diagnosis, yet its large-scale storage, privacy sensitivity, and annotation cost pose significant challenges. Inspired by biological vision systems, which integrate multi-sensory inputs and compress experiences into compact memory representations, we propose a bio-inspired framework termed Contrast-Guided Multi-modal Dataset Distillation (CGMDD). In biological perception, different sensory channels observe the same environment from complementary perspectives, while hierarchical neural processing ensures perceptual consistency across modalities. Meanwhile, memory systems such as the associated medial temporal lobe structures consolidate redundant experiences into efficient representations for long-term storage. Motivated by these principles, CGMDD treats multi-modal MRI as multi-view perceptual signals and introduces a hierarchical cross-modal contrastive learning mechanism that enforces perceptual alignment across modalities, analogous to multi-level processing in the visual cortex. Furthermore, we design a dynamic dataset distillation strategy that mimics memory consolidation by compressing large-scale data into compact, informative synthetic representations through gradient-based optimization. The proposed framework jointly optimizes perceptual alignment and memory compression in an end-to-end manner, achieving a biologically plausible integration of perception and learning. Experimental results on two MRI datasets demonstrate that CGMDD can compress the original dataset to 5% of its size while maintaining competitive performance, even with only 30% of the labels. These findings highlight the effectiveness of bio-inspired mechanisms in building efficient, robust, and privacy-preserving computer vision systems.