Multi-step reasoning remains challenging for language models with limited capacity. While recent reasoning distillation approaches transfer chain-of-thought supervision from large teacher models, they typically apply uniform supervision across all Transformer components, overlooking the fact that different modules contribute unequally to reasoning. We propose Module-Aware Reasoning Distillation, a parameter-efficient framework that explicitly targets key Transformer components for effective reasoning transfer. Through systematic analysis, we identify the feed-forward network projections and the output projection of self-attention as primary bottlenecks for reasoning. Based on these findings, we introduce lightweight adapter modules at these components while freezing the backbone parameters, enabling focused and efficient distillation. Our approach adopts an offline distillation setting, where a strong teacher model provides reasoning trajectories in advance, and incorporates an adaptive supervision strategy that adjusts the strength of reasoning-related losses according to problem difficulty. Experiments on mathematical reasoning benchmarks demonstrate consistent improvements over strong baselines, and ablation studies confirm the importance of both module-aware placement and adaptive supervision.

Existing text-guided image editing methods primarily rely on end-to-end pixel-level inpainting paradigm. Despite its success in simple scenarios, this paradigm still significantly struggles with compositional editing tasks that require precise local control and complex multi-object spatial reasoning. This paradigm is severely limited by 1) the implicit coupling of planning and execution, 2) the lack of object-level control granularity, and 3) the reliance on unstructured, pixel-centric modeling. To address these limitations, we propose I2E, a novel "Decompose-then-Action” paradigm that revisits image editing as an actionable interaction process within a structured environment. I2E utilizes a Decomposer to transform unstructured images into discrete, manipulable object layers and then introduces a physics-aware Vision-Language-Action Agent to parse complex instructions into a series of atomic actions via Chain-of-Thought reasoning. Further, we also construct I2E-Bench, a benchmark designed for multi-instance spatial reasoning and high-precision editing. Experimental results on I2E-Bench and multiple public benchmarks demonstrate that I2E significantly outperforms state-of-the-art methods in handling complex compositional instructions, maintaining physical plausibility, and ensuring multi-turn editing stability.

Large Language Models (LLMs) have shown strong potential for text-attributed graph (TAG) learning, yet effectively integrating LLM semantics with graph structural information remains challenging. Embeddings obtained from frozen LLMs lack topology awareness, while fine-tuning LLMs is often computationally expensive. Moreover, LLM embeddings are high-dimensional, and naively reducing dimensionality tends to destroy semantics. To address these issues, we propose GASE, a framework for learning Graph-Aware Semantic Embeddings using frozen LLMs. GASE consists of two key stages: First, we introduce a Training-Free Structure-Aware Semantic Extraction (TSSE) module. Through inter-layer semantic feedback and progressive masked attention, it efficiently compresses and propagates semantic context from neighboring nodes without updating LLM parameters. Second, we propose a Subspace Decomposition and Structural Injection (SDSI) strategy. Embeddings obtained from TSSE are decomposed into a semantic-rich subspace and a structural injection subspace, and structural signals are injected into the latter, which preserves original semantics while integrating graph information. Experiments demonstrate that GASE outperforms state-of-the-art baselines on node classification and achieves a 5× speedup over fine-tuning-based methods.

As a more natural and intelligent interaction manner, multimodal task-oriented dialog system recently has received great attention and many remarkable progresses have been achieved. Nevertheless, almost all existing studies follow the pipeline to first learn intra-modal features separately and then conduct simple feature concatenation or attention-based feature fusion to generate responses, which hampers them from learning inter-modal interactions and conducting cross-modal feature alignment for generating more intention-aware responses. To address these issues, we propose UniTranSeR, a Unified Transformer Semantic Representation framework with feature alignment and intention reasoning for multimodal dialog systems. Specifically, we first embed the multimodal features into a unified Transformer semantic space to prompt inter-modal interactions, and then devise a feature alignment and intention reasoning (FAIR) layer to perform cross-modal entity alignment and fine-grained key-value reasoning, so as to effectively identify user’s intention for generating more accurate responses. Experimental results verify the effectiveness of UniTranSeR, showing that it significantly outperforms state-of-the-art approaches on the representative MMD dataset.

Accurate fact verification depends on performing fine-grained reasoning over crucial entities by capturing their latent logical relations hidden in multiple evidence clues, which is generally lacking in existing fact verification models. In this work, we propose a novel Global-to-Local Aggregation and Fission network (GLAF) to fill this gap. Instead of treating entire sentences or all semantic elements within them as nodes to construct a coarse-grained or unstructured evidence graph as in previous methods, GLAF constructs a fine-grained and structured evidence graph by parsing the rambling sentences into structural triple-level reasoning clues and regarding them as graph nodes to achieve fine-grained and interpretable evidence graph reasoning. Specifically, to capture latent logical relations between the clues, GLAF first employs a local fission reasoning layer to conduct fine-grained multi-hop reasoning, and then uses a global evidence aggregation layer to achieve information sharing and the interchange of evidence clues for final claim label prediction. Experimental results on the FEVER dataset demonstrate the effectiveness of GLAF, showing that it achieves the state-of-the-art performance by obtaining a 77.62% FEVER score.

Recent years has witnessed the remarkable success in end-to-end task-oriented dialog system, especially when incorporating external knowledge information. However, the quality of most existing models’ generated response is still limited, mainly due to their lack of fine-grained reasoning on deterministic knowledge (w.r.t. conceptual tokens), which makes them difficult to capture the concept shifts and identify user’s real intention in cross-task scenarios. To address these issues, we propose a novel intention mechanism to better model deterministic entity knowledge. Based on such a mechanism, we further propose an intention reasoning network (IR-Net), which consists of joint and multi-hop reasoning, to obtain intention-aware representations of conceptual tokens that can be used to capture the concept shifts involved in task-oriented conversations, so as to effectively identify user’s intention and generate more accurate responses. Experimental results verify the effectiveness of IR-Net, showing that it achieves the state-of-the-art performance on two representative multi-domain dialog datasets.