Fan Gao, Cheng Huang, Yutong Liu, Nyima Tashi, Xiangxiang Wang, Thupten Tsering, Ban Ma-bao, Renzeng Duojie, Gadeng Luosang, Rinchen Dongrub, Dorje Tashi, Xiao Feng Cd, Yongbin Yu, Hao Wang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

To enhance the performance of a Takagi-Sugeno-Kang fuzzy classifier (TSKFC) on classification tasks, multiple single TSKFC models regarded as blocks to construct a deep TSKFC in series. Moreover, a novel evolutionary algorithm (EA) named hybrid search evolutionary algorithm with competitive behavior (C-SaHSEA) was proposed to search the best architecture of the deep TSKFNN for the different classification tasks. C-SaHSEA is a novel algorithm combined with two search algorithms owning strong exploration and exploitation characteristics separately which can be support to each other to enhance the search ability and stability through the hybrid search. Moreover, self-adaptive update laws for the parameters of mutation and crossover operators were adopted to alleviate the difficulty of complex design and the influence of additional parameters on the search ability. Considering the computation cost of the optimization task of the optimal architecture search, a mechanism called competitive behavior was deployed into the proposed search algorithm. To demonstrate the superiority of the C-SaHSEA, eight EAs was conducted as comparison methods on five test functions. The results of mean fitness and standard deviation fitness demonstrated the search ability and stability of the C-SaHSEA. Then, eight dataset was used to validate the performance improvement of the deep TSKFC optimized by the C-SaHSEA. By comparing with the other nine methods, the high-performance of the optimized deep TSKFC has been improved.

Medical report generation can be helpful in diagnoses. Despite the previous efforts of researchers, current models still need improvements in the extraction of image features and quality of generated reports. In this paper, we propose an attention enhanced network with semantic inspector (AENSI) as a new automatic medical report generation model, which serves to help doctors get a high-quality report. For the model, we propose double-weighted multi-head attention as our attention module, where different heads are aggregated with double weights (DWMHA) to enhance its power in catching subtle features and drawing correlations between images and texts. To prevent the drawback of imprecise multi-label classification modules used in current generation models, we design a novel module following decoder that treats tags as inspectors of the generated reports, namely Tag Inspector, as a substitute for the previous classification module. Experimental results of AENSI achieve to the level of state-of-the-art. On IU X-ray, our model surpasses all previous works on every metrics; on PEIR Gross, our model ranks first on BLEU-4 and ROUGE and closely approaches the best on other metrics.

Matrix operation is easy to be paralleled by hardware, and the memristor network can realize a parallel matrix computing model with in-memory computing. This article proposes a matrix-friendly genetic algorithm (MGA), in which the population is represented by a matrix and the evolution of population is realized by matrix operations. Compared with the performance of a baseline genetic algorithm (GA) on solving the maximum value of the binary function, MGA can converge better and faster. In addition, MGA is more efficient because of its parallelism on matrix operations, and MGA runs 2.5 times faster than the baseline GA when using the NumPy library. Considering the advantages of the memristor in matrix operations, memristor circuits are designed for the deployment of MGA. This deployment method realizes the parallelization and in-memory computing (memristor is both memory and computing unit) of MGA. In order to verify the effectiveness of this deployment, a feature selection experiment of logistic regression (LR) on Sonar datasets is completed. LR with MGA-based feature selection uses 46 fewer features and achieves 11.9% higher accuracy.

Periodic disturbances occur during repetitive operation of machines in industrial production. Compensation for the periodic disturbances is an important issue to realize proper machine works beacause the periodic disturbances deteriorate machining precision. In order to eliminate the periodic disturbances, an adaptive periodic-disturbance observer (APDOB) has been proposed as an effective method that can also estimate and compensate for frequency-varying periodic disturbances. However, the APDOB has a problem that design of the APDOB is complicated owing to its six design parameters, which need to be empirically adjusted. Here, we propose an approach based on a genetic algorithm (GA) including a Lévy flight to automatically adjust the six design parameters. The proposed method can remove the conventional empirical design. Moreover, the Lévy flight could improve the exploration ability of the GA by optimizing mutation operator and the best solution found by the GA including Lévy flight could improve the performance of the APDOB.