End-to-end Temporal Action Detection (TAD) has achieved remarkable progress in recent years, driven by innovations in model architectures and the emergence of Video Foundation Models (VFMs). However, existing TAD methods that perform full fine-tuning of pretrained video models often incur substantial computational costs, which become particularly pronounced when processing long video sequences. Moreover, the need for precise temporal boundary annotations makes data labeling extremely expensive. In low-resource settings where annotated samples are scarce, direct fine-tuning tends to cause overfitting. To address these challenges, we introduce Dynamic Low-Rank Adapter (DyLoRA), a lightweight fine-tuning framework tailored specifically for the TAD task. Built upon the Low-Rank Adaptation (LoRA) architecture, DyLoRA adapts only the key layers of the pretrained model via low-rank decomposition, reducing the number of trainable parameters to less than 5% of full fine-tuning methods. This significantly lowers memory consumption and mitigates overfitting in low-resource settings. Notably, DyLoRA enhances the temporal modeling capability of pretrained models by optimizing temporal dimension weights, thereby alleviating the representation misalignment of temporal features. Experimental results demonstrate that DyLoRA-TAD achieves impressive performance, with 73.9% mAP on THUMOS14, 39.52% on ActivityNet-1.3, and 28.2% on Charades, substantially surpassing the best traditional feature-based methods.

In the field of intelligent education, the integration of artificial intelligence, especially deep learning technologies, has garnered significant attention. Knowledge tracing (KT) plays a pivotal role in this field by predicting students’ future performance through the analysis of historical interaction data, thereby assisting educators in evaluating knowledge mastery and tailoring instructional strategies. Traditional knowledge tracing methods, largely based on Recurrent Neural Networks (RNNs) and Transformer models, primarily focus on capturing long-term interaction patterns in sequential data. However, these models may neglect crucial short-term dynamics and other relevant features. This paper introduces a novel approach to knowledge tracing by leveraging a pure Multilayer Perceptron (MLP) architecture. We propose MixerKT, a knowledge tracing model based on the HyperMixer framework, which uniquely integrates global and local Mixer feature extractors. This architecture enables more effective extraction of both long-term interaction trends and recent learning behaviors, addressing limitations in current models that may overlook these key aspects. Empirical evaluations on two widely-used datasets, ASSISTments2009 and Algebra2005, demonstrate that MixerKT consistently outperforms several state-of-the-art models, including DKT, SAKT, and Separated Self-Attentive Neural Knowledge Tracing (SAINT). Specifically, MixerKT achieves higher prediction accuracy, highlighting its effectiveness in capturing the nuances of learners’ knowledge states. These results indicate that our model provides a more comprehensive representation of student learning patterns, enhancing the ability to predict future performance with greater precision.