Ahmet Ezgi¹, Aytuğ Onan^2,*

CMC-Computers, Materials & Continua, Vol.86, No.3, 2026, DOI:10.32604/cmc.2025.071301 - 12 January 2026

Abstract Knowledge distillation has become a standard technique for compressing large language models into efficient student models, but existing methods often struggle to balance prediction accuracy with explanation quality. Recent approaches such as Distilling Step-by-Step (DSbS) introduce explanation supervision, yet they apply it in a uniform manner that may not fully exploit the different learning dynamics of prediction and explanation. In this work, we propose a task-structured curriculum learning (TSCL) framework that structures training into three sequential phases: (i) prediction-only, to establish stable feature representations; (ii) joint prediction–explanation, to align task outputs with rationale generation; and (iii)… More >

Moyi Zhang, Yixin Wang^*, Yu Cheng

CMC-Computers, Materials & Continua, Vol.86, No.3, 2026, DOI:10.32604/cmc.2025.071021 - 12 January 2026

Abstract Modern intelligent systems, such as autonomous vehicles and face recognition, must continuously adapt to new scenarios while preserving their ability to handle previously encountered situations. However, when neural networks learn new classes sequentially, they suffer from catastrophic forgetting—the tendency to lose knowledge of earlier classes. This challenge, which lies at the core of class-incremental learning, severely limits the deployment of continual learning systems in real-world applications with streaming data. Existing approaches, including rehearsal-based methods and knowledge distillation techniques, have attempted to address this issue but often struggle to effectively preserve decision boundaries and discriminative features… More >

Xin Ma^1,2, Jin Lei^3,4,*, Chenying Pei⁴, Chunming Wu⁴

CMC-Computers, Materials & Continua, Vol.86, No.2, pp. 1-17, 2026, DOI:10.32604/cmc.2025.069353 - 09 December 2025

Abstract This study proposes a lightweight apple detection method employing cascaded knowledge distillation (KD) to address the critical challenges of excessive parameters and high deployment costs in existing models. We introduce a Lightweight Feature Pyramid Network (LFPN) integrated with Lightweight Downsampling Convolutions (LDConv) to substantially reduce model complexity without compromising accuracy. A Lightweight Multi-channel Attention (LMCA) mechanism is incorporated between the backbone and neck networks to effectively suppress complex background interference in orchard environments. Furthermore, model size is compressed via Group_Slim channel pruning combined with a cascaded distillation strategy. Experimental results demonstrate that the proposed model More >

Guo-Feng Fan^1,2, Jia-Jing Qian¹, Li-Ling Peng¹, Xin-Hang Jia¹, Ling-Han Zuo¹, Jia-Can Yan¹, Jiang-Yan Chen¹, Anantkumar J. Umbarkar³, Wei-Chiang Hong^4,*

CMES-Computer Modeling in Engineering & Sciences, Vol.145, No.3, pp. 3813-3837, 2025, DOI:10.32604/cmes.2025.073615 - 23 December 2025

Abstract Accurate prediction of perovskite photovoltaic materials’ optoelectronic properties is crucial for developing efficient and stable materials, advancing solar technology. To address poor interpretability, high computational complexity, and inaccurate predictions in relevant machine learning models, this paper proposes a novel methodology. The technical route of this paper mainly centers on the random forest-knowledge distillation-bidirectional gated recurrent unit with attention technology (namely RF-KD-BIGRUA), which is applied in perovskite photovoltaic materials. Primarily, it combines random forest to quantitatively assess feature importance, selecting variables with significant impacts on photoelectric conversion efficiency. Subsequently, statistical techniques analyze the weight distribution of More >

Huan Zeng, Jianxun Zhang^*, Hongji Chen, Xinwei Zhu

CMC-Computers, Materials & Continua, Vol.85, No.2, pp. 3879-3896, 2025, DOI:10.32604/cmc.2025.066803 - 23 September 2025

Abstract Semantic segmentation in street scenes is a crucial technology for autonomous driving to analyze the surrounding environment. In street scenes, issues such as high image resolution caused by a large viewpoints and differences in object scales lead to a decline in real-time performance and difficulties in multi-scale feature extraction. To address this, we propose a bilateral-branch real-time semantic segmentation method based on semantic information distillation (BSDNet) for street scene images. The BSDNet consists of a Feature Conversion Convolutional Block (FCB), a Semantic Information Distillation Module (SIDM), and a Deep Aggregation Atrous Convolution Pyramid Pooling (DASP). More >

Jinhai Wang¹, Junwei Xue¹, Hongyan Zhang², Hui Xiao^3,4, Huiling Wei^3,4, Mingyou Chen^3,4, Jiang Liao², Lufeng Luo^3,4,*

CMC-Computers, Materials & Continua, Vol.84, No.1, pp. 1839-1861, 2025, DOI:10.32604/cmc.2025.064489 - 09 June 2025

Abstract Defect detection based on computer vision is a critical component in ensuring the quality of industrial products. However, existing detection methods encounter several challenges in practical applications, including the scarcity of labeled samples, limited adaptability of pre-trained models, and the data heterogeneity in distributed environments. To address these issues, this research proposes an unsupervised defect detection method, FLAME (Federated Learning with Adaptive Multi-Model Embeddings). The method comprises three stages: (1) Feature learning stage: this work proposes FADE (Feature-Adaptive Domain-Specific Embeddings), a framework employs Gaussian noise injection to simulate defective patterns and implements a feature discriminator… More >

Haocheng Sun, Ping Li, Ying Li^*

CMC-Computers, Materials & Continua, Vol.83, No.3, pp. 4899-4914, 2025, DOI:10.32604/cmc.2025.063029 - 19 May 2025

Abstract Traffic flow prediction is a key component of intelligent transportation systems, particularly in data-scarce regions where traditional models relying on complete datasets often fail to provide accurate forecasts. These regions are characterized by limited sensor coverage and sparse data collection, pose significant challenges for existing prediction methods. To address this, we propose a novel transfer learning framework called transfer learning with deep knowledge distillation (TL-DKD), which combines graph neural network (GNN) with deep knowledge distillation to enable effective knowledge transfer from data-rich to data-scarce domains. Our contributions are three-fold: (1) We introduce, for the first… More >

Erlin Tian¹, Zengchao Zhu^2,*, Fangmei Liu², Zuhe Li²

CMC-Computers, Materials & Continua, Vol.83, No.2, pp. 2305-2322, 2025, DOI:10.32604/cmc.2025.061145 - 16 April 2025

Abstract Within the realm of multimodal neural machine translation (MNMT), addressing the challenge of seamlessly integrating textual data with corresponding image data to enhance translation accuracy has become a pressing issue. We saw that discrepancies between textual content and associated images can lead to visual noise, potentially diverting the model’s focus away from the textual data and so affecting the translation’s comprehensive effectiveness. To solve this visual noise problem, we propose an innovative KDNR-MNMT model. The model combines the knowledge distillation technique with an anti-noise interaction mechanism, which makes full use of the synthesized graphic knowledge… More >

Min Yao^1,*, Guofeng Liu², Yaozu Zhang³, Guangjie Hu¹

CMC-Computers, Materials & Continua, Vol.83, No.2, pp. 2189-2205, 2025, DOI:10.32604/cmc.2025.060609 - 16 April 2025

Abstract Knowledge distillation (KD) is an emerging model compression technique for learning compact object detector models. Previous KD often focused solely on distilling from the logits layer or the feature intermediate layers, which may limit the comprehensive learning of the student network. Additionally, the imbalance between the foreground and background also affects the performance of the model. To address these issues, this paper employs feature-based distillation to enhance the detection performance of the bounding box localization part, and logit-based distillation to improve the detection performance of the category prediction part. Specifically, for the intermediate layer feature… More >

Jiyang Xu¹, Qi Wang^1,*, Xin Xiong², Weidong Min^1,3, Jiang Luo⁴, Di Gai¹, Qing Han^1,3

CMC-Computers, Materials & Continua, Vol.82, No.3, pp. 3921-3941, 2025, DOI:10.32604/cmc.2024.058586 - 06 March 2025

Abstract The unsupervised vehicle re-identification task aims at identifying specific vehicles in surveillance videos without utilizing annotation information. Due to the higher similarity in appearance between vehicles compared to pedestrians, pseudo-labels generated through clustering are ineffective in mitigating the impact of noise, and the feature distance between inter-class and intra-class has not been adequately improved. To address the aforementioned issues, we design a dual contrastive learning method based on knowledge distillation. During each iteration, we utilize a teacher model to randomly partition the entire dataset into two sub-domains based on clustering pseudo-label categories. By conducting contrastive… More >