Non-Exemplar Class Incremental Learning (NECIL) strives to preserve classification performance in an evolving data stream without revisiting old-class exemplars. Current methods mitigate catastrophic forgetting by replaying and augmenting historical prototypes as surrogates for old classes. However, they treat prototypes as holistic representations for global-level augmentations, which overlook dimensional semantic disparity and old-new class relationships, failing to maintain old-class discriminability and adaptability to the evolving feature space. To address this challenge, we propose Dimensionally-Allocated Prototype Refinement (DiAPR), a granular framework that progressively refines prototypes to exhibit class separability in the new feature space through three modules. Specifically, Distribution-aware Pairing (DAP) captures old-new class semantic consistency to guide Granular Semantic Allocation (GSA) in dimension-wise conflation, while Cross-Dimensional Transition (CDT) enhances cross-dimensional dependencies. The resulting prototypes sharpen classifier decision boundaries. Moreover, CDT inherently enables softened feature alignment, thereby yielding a more compatible feature space. Extensive experiments demonstrate DiAPR’s superiority, with improvements over SOTA by 2.35%, 0.70%, 0.96% on three CIFAR-100 settings, 1.03%, 0.54%, 0.40% on Tiny-ImageNet, and 0.60% on ImageNet-Subset.

Few-shot image classification stands as a pivotal task within the realm of computer vision. However, obtaining accurate class prototypes from limited annotated samples is a challenging problem. In recent years, many methods based on prototype networks have shown excellent performance. Nevertheless, existing methods overlook the discriminative semantic information lost due to sample scarcity and the hidden category information in the query set, failing to address the issue of unreliable prototypes generated from limited annotated samples. In this paper, we propose a feature enhancement network for few-shot classification. To improve the accuracy and robustness of few-shot classification models, we first enhance the support set through learning a weight matrix and then align the enhanced support set prototypes with textual semantics. To avoid being influenced by introduced prior noise, we fuse between semantically aligned prototypes and mean prototypes and ultimately utilize query prototypes for dynamic updating to obtain more accurate class prototypes. Extensive experiments demonstrate that our method achieves competitive performance on miniImageNet and tieredImageNet datasets. Furthermore, it exhibits excellent results in cross-domain few-shot classification.

Challenges in gait recognition arise from the difficulty of subject identification through subtle spatial differences. Learning the temporal representation helps, however, straightforward temporal modeling like 3D Convolution not only incurs excessive computational complexity, but may also further interfere with the already limited spatial details. In this paper, we present a Spatial-Temporal Decoupled Network (STDN) to balance the trade-off between the cross-frame interference, computational complexity, and model performance. Specifically, motivated by the concept of spatial-temporal decoupling, we propose a Decoupled Parallel Feature Extractor (DPFE) for noise-suppressed spatial-temporal features extraction. By incorporating DPFE as the basic component, we build a Efficient Spatial-temporal Awareness Block (ESA Block) that accounts for the unique properties of gait data, namely, restricted spatial information and abundant temporal information, to minimize the computational overhead imposed by temporal modeling. Experiments on CASIA-B and OU-MVLP datasets demonstrate the state-of-the-art performance of our proposed method, with an average rank-1 accuracy of 93.7% on the CASIA-B and 90.5% on the OU-MVLP. The source code is available on https://github.com/tk59854/stdn.

Alzheimer's disease (AD) is one of the most challenging neurodegenerative diseases because of its complicated and progressive mechanisms, and multiple risk factors. Increasing research evidence demonstrates that genetics may be a key factor responsible for the occurrence of the disease. Although previous reports identified quite a few AD-associated genes, they were mostly limited owing to patient sample size and selection bias. There is a lack of comprehensive research aimed to identify AD-associated risk mutations systematically. To address this challenge, we hereby construct a large-scale AD mutation and co-mutation framework ('AD-Syn-Net'), and propose deep learning models named Deep-SMCI and Deep-CMCI configured with fully connected layers that are capable of predicting cognitive impairment of subjects effectively based on genetic mutation and co-mutation profiles. Next, we apply the customized frameworks to data sets to evaluate the importance scores of the mutations and identified mutation effectors and co-mutation combination vulnerabilities contributing to cognitive impairment. Furthermore, we evaluate the influence of mutation pairs on the network architecture to dissect the genetic organization of AD and identify novel co-mutations that could be responsible for dementia, laying a solid foundation for proposing future targeted therapy for AD precision medicine. Our deep learning model codes are available open access here: https://github.com/Pan-Bio/AD-mutation-effectors.