Cognitive diagnosis aims to infer students' mastery levels over knowledge components from their learning interactions, supporting personalized education applications. However, existing models encode responses as binary correctness labels, discarding information about which specific option a student selected. This input-level information loss limits their ability to distinguish qualitatively different error types. As a result, providing interpretable diagnostic outputs becomes challenging. To address these limitations, we propose SACD, a semantic-augmented cognitive diagnosis framework that integrates LLM-based semantic analysis with student behavioral modeling. SACD comprises the following key components. First, an LLM-based exercise diagnostic generator analyzes exercise content and produces structured semantic annotations for each answer choice, capturing the specific misconceptions each option represents. Second, a kernel-based alignment mechanism projects semantic embeddings and behavioral representations into a unified kernel space, enabling effective fusion of heterogeneous information. Third, an interpretable diagnosis layer predicts student performance and generates fine-grained mastery estimates, which LLMs further process to produce actionable learning plans. Extensive experiments on three real-world datasets demonstrate that SACD achieves superior prediction accuracy while enabling interpretable, actionable diagnostics.

Knowledge tracing (KT) models students' knowledge states and predicts their future performance based on their historical interaction data. However, attention based KT models struggle to accurately capture diverse forgetting behaviors in ever-growing interaction sequences. First, existing models use uniform time decay matrices, conflating forgetting representations with problem relevance. Second, the fixed-length window prediction paradigm fails to model continuous forgetting processes in expanding sequences. To address these challenges, this paper introduces LefoKT, a unified architecture that enhances attention based KT models by incorporating proposed relative forgetting attention. LefoKT improves forgetting modeling through relative forgetting attention to decouple forgetting patterns from problem relevance. It also enhances attention based KT models' length extrapolation capability for capturing continuous forgetting processes in ever-growing interaction sequences. Extensive experimental results on three datasets validate the effectiveness of LefoKT.

Knowledge tracing (KT) involves using the historical records of student-learning interactions to anticipate their performance on forthcoming questions. Central to this process is the modeling of human cognition to gain deeper insights into how knowledge is acquired and retained. Human cognition is characterized by two key features: long-term cognitive trends, reflecting the gradual accumulation and stabilization of knowledge over time, and short-term cognitive fluctuations, which arise from transient factors such as forgetting or momentary lapses in attention. Although existing attention-based KT models effectively capture long-term cognitive trends, they often fail to adequately address short-term cognitive fluctuations. These limitations lead to overly smoothed cognitive features and reduced model performance, especially when the test data length exceeds the training data length. To address these problems, we propose FlucKT, a novel short-term cognitive fluctuations enhanced attention network for KT tasks. FlucKT improves the attention mechanism in two ways: First, by using a decomposition-based layer with causal convolution to separate and dynamically reweight long-term and short-term cognitive features. Second, by introducing a kernelized bias attention score penalty to enhance focus on short-term fluctuations, improving length generalization capabilities. Our contributions are validated through extensive experiments on three real-world datasets, demonstrating significant improvements in length generalization and prediction performance.

Step-level reward models (SRMs) can significantly enhance mathematical reasoning performance through process supervision or step-level preference alignment based on reinforcement learning. The performance of SRMs is pivotal, as they serve as critical guidelines, ensuring that each step in the reasoning process is aligned with desired outcomes. Recently, AlphaZero-like methods, where Monte Carlo Tree Search (MCTS) is employed for automatic step-level preference annotation, have proven particularly effective. However, the precise mechanisms behind the success of SRMs remain largely unexplored. To address this gap, this study delves into the counterintuitive aspects of SRMs, particularly focusing on MCTS-based approaches. Our findings reveal that the removal of natural language descriptions of thought processes has minimal impact on the efficacy of SRMs. Furthermore, we demonstrate that SRMs are adept at assessing the complex logical coherence present in mathematical language while having difficulty in natural language. These insights provide a nuanced understanding of the core elements that drive effective step-level reward modeling in mathematical reasoning. By shedding light on these mechanisms, this study offers valuable guidance for developing more efficient and streamlined SRMs, which can be achieved by focusing on the crucial parts of mathematical reasoning.

Mathematical reasoning remains a challenging area for large language models (LLMs), prompting the development of math-specific LLMs such as LLEMMA, DeepSeekMath, and Qwen2-Math, among others. These models typically follow a two-stage training paradigm: pre-training with math-related corpora and post-training with problem datasets for supervised fine-tuning (SFT). Despite these efforts, the improvements in mathematical reasoning achieved through continued pre-training (CPT) are often less significant compared to those obtained via SFT. This study addresses this discrepancy by exploring alternative strategies during the pre-training phase, focusing on the use of problem-solving data over general mathematical corpora. We investigate three primary research questions: (1) Can problem-solving data enhance the model's mathematical reasoning capabilities more effectively than general mathematical corpora during CPT? (2) Are synthetic data from the same source equally effective, and which synthesis methods are most efficient? (3) How do the capabilities developed from the same problem-solving data differ between the CPT and SFT stages, and what factors contribute to these differences? Our findings indicate that problem-solving data significantly enhances the model's mathematical capabilities compared to general mathematical corpora. We also identify effective data synthesis methods, demonstrating that the tutorship amplification synthesis method achieves the best performance. Furthermore, while SFT facilitates instruction-following abilities, it underperforms compared to CPT with the same data, which can be partially attributed to its poor learning capacity for more challenging problem-solving data. These insights provide valuable guidance for optimizing the mathematical reasoning capabilities of LLMs, culminating in our development of a powerful mathematical base model called MathGPT-8B.

Sparsely activated Mixture-of-Experts (MoE) models effectively increase the number of parameters while maintaining consistent computational costs per token. However, vanilla MoE models often suffer from limited diversity and specialization among experts, constraining their performance and scalability, especially as the number of experts increases. In this paper, we present a novel perspective on vanilla MoE with top-k routing inspired by sparse representation. This allows us to bridge established theoretical insights from sparse representation into MoE models. Building on this foundation, we propose a group sparse regularization approach for the input of top-k routing, termed Mixture of Group Experts (MoGE). MoGE indirectly regularizes experts by imposing structural constraints on the routing inputs, while preserving the original MoE architecture. Furthermore, we organize the routing input into a 2D topographic map, spatially grouping neighboring elements. This structure enables MoGE to capture representations invariant to minor transformations, thereby significantly enhancing expert diversity and specialization. Comprehensive evaluations across various Transformer models for image classification and language modeling tasks demonstrate that MoGE substantially outperforms its MoE counterpart, with minimal additional memory and computation overhead. Our approach provides a simple yet effective solution to scale the number of experts and reduce redundancy among them. Our code is available at: https://github.com/wangyuankl123/MoGE.

Sketching in Virtual Reality (VR) is challenging mainly due to the absence of physical surface support and virtual depth perception cues, which induce high cognitive and sensorimotor load. This paper presents WieldingCanvas, an interactive VR sketching platform that integrates canvas manipulations to draw lines and curves in 3D. Informed by real-life examples of two-handed creative activities, WieldingCanvas interprets users’ spatial gestures to move, swing, rotate, transform, or fold a virtual canvas, whereby users simply draw primitive strokes on the canvas, which are turned into finer and more sophisticated shapes via the manipulation of the canvas. We evaluated the capability and user experience of WieldingCanvas with two studies where participants were asked to sketch target shapes. A set of freehand sketches of high aesthetic qualities were created, and the results demonstrated that WieldingCanvas can assist users with creating 3D sketches.