Chengjun Li

Xuyi Yu, Yaohua Liu, Ziming Song et al.

Knowledge distillation (KD) typically relies on the fixed perspective of a single teacher, limiting the diversity of supervisory signals. While multi-teacher distillation addresses this by aggregating knowledge from multiple models, it incurs prohibitive computational and storage costs. To balance efficiency and diversity, recent research has focused on generating virtual views from a single teacher. However, existing methods face a trade-off: random perturbation approaches offer efficiency but lack controlled diversity, while structured augmentation methods require multi-stage training and incur linear parameter growth. We observe that this trade-off stems from a common design choice: using the teacher's strong but static features to generate views. Instead, we propose Shift-Augmented Knowledge Distillation (SAKD), a simple yet effective framework that leverages the student's evolving features as a dynamic condition for perturbation generation. This shift in perspective enables single-stage training while producing adaptive, diverse views through a parameter-free cyclic shift. Extensive experiments on CIFAR-100 and ImageNet demonstrate that SAKD consistently outperforms random perturbation methods and achieves accuracy on par with two-stage approaches, while using significantly fewer parameters and eliminating pre-training requirements.

Chengjun Li, Yang Li

Differential Evolution (DE) is quite powerful for real parameter single objective optimization. However, the ability of extending or changing search area when falling into a local optimum is still required to be developed in DE for accommodating extremely complicated fitness landscapes with a huge number of local optima. We propose a new flow of DE, termed DE with individuals redistribution, in which a process of individuals redistribution will be called when progress on fitness is low for generations. In such a process, mutation and crossover are standardized, while trial vectors are all kept in selection. Once diversity exceeds a predetermined threshold, our opposition replacement is executed, then algorithm behavior returns to original mode. In our experiments based on two benchmark test suites, we apply individuals redistribution in ten DE algorithms. Versions of the ten DE algorithms based on individuals redistribution are compared with not only original version but also version based on complete restart, where individuals redistribution and complete restart are based on the same entry criterion. Experimental results indicate that, for most of the DE algorithms, version based on individuals redistribution performs better than both original version and version based on complete restart.

Tai-Ling Yuan, Zhe Zhu, Kun Xu et al.

We introduce Chinese Text in the Wild, a very large dataset of Chinese text in street view images. While optical character recognition (OCR) in document images is well studied and many commercial tools are available, detection and recognition of text in natural images is still a challenging problem, especially for more complicated character sets such as Chinese text. Lack of training data has always been a problem, especially for deep learning methods which require massive training data. In this paper we provide details of a newly created dataset of Chinese text with about 1 million Chinese characters annotated by experts in over 30 thousand street view images. This is a challenging dataset with good diversity. It contains planar text, raised text, text in cities, text in rural areas, text under poor illumination, distant text, partially occluded text, etc. For each character in the dataset, the annotation includes its underlying character, its bounding box, and 6 attributes. The attributes indicate whether it has complex background, whether it is raised, whether it is handwritten or printed, etc. The large size and diversity of this dataset make it suitable for training robust neural networks for various tasks, particularly detection and recognition. We give baseline results using several state-of-the-art networks, including AlexNet, OverFeat, Google Inception and ResNet for character recognition, and YOLOv2 for character detection in images. Overall Google Inception has the best performance on recognition with 80.5% top-1 accuracy, while YOLOv2 achieves an mAP of 71.0% on detection. Dataset, source code and trained models will all be publicly available on the website.

Chengjun Li, Jia Wu

In distributed evolutionary algorithms, migration interval is used to decide migration moments. Nevertheless, migration moments predetermined by intervals cannot match the dynamic situation of evolution. In this paper, a scheme of setting the success rate of migration based on subpopulation diversity at each interval is proposed. With the scheme, migration still occurs at intervals, but the probability of immigrants entering the target subpopulation will be determined by the diversity of this subpopulation according to a proposed formula. An analysis shows that the time consumption of our scheme is acceptable. In our experiments, the basement of parallelism is an evolutionary algorithm for the traveling salesman problem. Under different value combinations of parameters for the formula, outcomes for eight benchmark instances of the distributed evolutionary algorithm with the proposed scheme are compared with those of a traditional one, respectively. Results show that the distributed evolutionary algorithm based on our scheme has a significant advantage on solutions especially for high difficulty instances. Moreover, it can be seen that the algorithm with the scheme has the most outstanding performance under three value combinations of above-mentioned parameters for the formula.