Fully mining the differential features of different class samples in overlapping areas is the key and difficult point to improving imbalanced classification performance under complex distribution patterns. Although existing data-level and algorithm-level methods have achieved good results in dealing with overlapping problems, sample generation and classifier training heavily rely on distribution information, and the ability to mine the different information is limited. This paper proposes a dual imbalanced classification framework with feature transfer guided by memory compensation strategy, which enhances the model's ability to mine differential features by constructing a feature space with better inter-class separability. In the traditional classification branch, a feature extraction network maps original samples to feature space and a traditional classifier is used to classify the features. In the compensation classification branch, a feature memory module based on iterative clustering strategy is designed, separately obtaining and saving the correctly classified feature centers of different classes. Moreover, a feature transfer module based on vector combination theory is proposed, combining “push” and “pull” vectors to transfer the misclassified features to the non-overlapping areas corresponding to the same class feature memory module, thereby constructing a feature space with better inter-class separability. Finally, a classification compensation strategy based on feature similarity is designed, integrating the prediction results of the traditional classifier and feature memory module as the final classification results. Experimental results on 50 imbalanced datasets show the proposed method outperforms 28 typical imbalanced classification methods in F1-score and G-mean. Especially on 20 severely overlapping datasets, the performance improvement is more significant.

Generalized few-shot object detection aims to improve detection accuracy for novel classes while maintaining high performance on base classes. Traditional fine-tuning approaches often blur feature boundaries, leading to misclassification of novel samples as base classes or background. Additionally, differences in data distributions between base and novel classes can cause the model to “forget” base knowledge. This paper proposes a novel generalized few-shot detection method that leverages memory distillation of category prototypes. The approach includes two key components: a variational prototype refinement module (VPRM) and a memory bank of category prototypes (MBCP). The variational prototype refinement module introduces a class-agnostic feature fusion mechanism based on the original variational autoencoder. First, the mean and variance of the original distribution of base class are estimated in the base class training stage. The noise variables are converted into memory prototypes with strong generalization ability through reparameterization and stored. Second, the stored memory prototypes are fused with class-agnostic features of novel classes in the fine-tuning stage, which significantly alleviates the problem of base class bias when processing novel classes. In the base class training phase, the category prototype memory bank stores the base class memory prototypes extracted by the variational prototype refinement module and selects the best memory items by dynamically updating the category confidence and intersection-over-union threshold. This memory item can be used not only to constrain features of base classes to alleviate catastrophic forgetting of base classes but also to fuse with features of novel classes, adaptively extracting class-agnostic common information to strengthen the feature representation of the novel class. Experiments on PASCAL VOC and MS-COCO show superior average precision in both single-round and multi-round tests, outperforming existing state-of-the-art methods.

Multivariate time series (MTS), whose patterns change dynamically, often have complex temporal and dimensional dependence. Most existing reconstruction-based MTS anomaly detection methods only learn the point-wise information while ignoring the overall trend of time series, resulting in their incompetence in extracting high-level semantic information. Although a few contrastive learning-based approaches have been proposed recently to solve this problem, they forcibly increase the difference between the features of normal data, leading to the loss of useful information. This paper proposes an adversarial contrastive autoencoder (ACAE) for MTS anomaly detection. ACAE conducts feature combination and decomposition as the contrastive learning proxy task, which introduces adversarial training to learn the transformation-invariant representation of data, achieving a robust representation of MTS. Firstly, ACAE constructs positive and negative sample pairs through the multi-scale timestamp mask and random sampling. Secondly, the features of the original samples are combined with those of the positive and negative samples to generate the positive and negative composite features. Finally, ACAE trains the encoder and discriminator to decompose the negative composite features cooperatively to decrease the similarity between the features of negative pairs. In contrast, it adversarially decomposes the positive composite features to increase the similarity between the features of positive pairs. Experimental results show that ACAE outperforms 14 state-of-the-art baselines on five real-world datasets from different fields.

A programmable parallel digital signal processor (DSP) core for embedded applications is presented which combines the concepts of single instruction stream over multiple data streams (SIMD) and reconfigurable architecture. Equipped with eight SIMD-controlled 16-bit datapaths which can also be reconfigured as two 32-bit datapaths, the DSP core can process both 16-bit and 32-bit data in parallel, showing high performance, especially in the applications preferring parallel data flow computations, such as image processing. The SIMD scheme is extended with the instant-scalability of datapaths (ISSIMD), which offers the DSP a capability of dimension-controllable vector processing, so that to provide flexibility for different embedded applications. A first prototype in 0.18-/spl mu/m CMOS technology has been fabricated, which achieves IGMACS performance at the clock of 125 MHz.