论文检索

Xi Chen, Aske Plaat, Niki Van Stein

Chain‑of‑thought (CoT) prompting boosts Large Language Models accuracy on multi‑step tasks, yet whether the generated ``thoughts'' reflect the true internal reasoning process is unresolved. We present the first feature‑level causal study of CoT faithfulness. Combining sparse autoencoders with activation patching, we extract monosemantic features from Pythia‑70M and Pythia‑2.8B while they tackle GSM8K math problems under CoT and plain (noCoT) prompting. Swapping a small set of CoT‑reasoning features into a noCoT run raises answer log‑probabilities significantly in the 2.8B model, but has no reliable effect in 70M, revealing a clear contrast for these two scales. CoT also leads to significantly higher activation sparsity and feature interpretability scores in the larger model, signalling more modular internal computation. For example, the model's confidence in generating correct answers improves from 1.2 to 4.3. We introduce patch‑curves and random‑feature patching baselines, showing that useful CoT information is not only present in the top-K patches but widely distributed. Overall, our results indicate that CoT can induce more interpretable internal structures in high-capacity LLMs, validating its role as a structured prompting method.

Haozhuo Zheng, Cheng Wang, Yang Liu

The de novo generation of molecules with desirable properties is a critical challenge, where diffusion models are computationally intensive and autoregressive models struggle with error propagation. In this work, we introduce the Graph VQ-Transformer (GVT), a two-stage generative framework that achieves both high accuracy and efficiency. The core of our approach is a novel Graph Vector Quantized Variational Autoencoder (VQ-VAE) that compresses molecular graphs into high-fidelity discrete latent sequences. By synergistically combining a Graph Transformer with canonical Reverse Cuthill-McKee (RCM) node ordering and Rotary Positional Embeddings (RoPE), our VQ-VAE achieves near-perfect reconstruction rates. An autoregressive Transformer is then trained on these discrete latents, effectively converting graph generation into a well-structured sequence modeling problem. Crucially, this mapping of complex graphs to high-fidelity discrete sequences bridges molecular design with the powerful paradigm of large-scale sequence modeling, unlocking potential synergies with Large Language Models (LLMs). Extensive experiments show that GVT achieves state-of-the-art or highly competitive performance across major benchmarks like ZINC250k, MOSES, and GuacaMol, and notably outperforms leading diffusion models on key distribution similarity metrics such as FCD and KL Divergence. With its superior performance, efficiency, and architectural novelty, GVT not only presents a compelling alternative to diffusion models but also establishes a strong new baseline for the field, paving the way for future research in discrete latent-space molecular generation.

Shangshang Yang, Xuewen Duan, Xiaoshan Yu, Ziwen Wang, Haiping Ma, Xingyi Zhang

Cognitive diagnosis (CD), inferring student knowledge mastery based on historical response records, is crucial for personalized educational services such as adaptive practice and learning path planning. Existing CD models were built based on the assumption that student's response data is integral, overlooking the nonrandom missingness of data caused by student answering exercises selectively. This missingness generally leads to biased and incomplete observations, where confounders, such as selection bias and exposure bias, significantly undermine the accuracy of student knowledge modeling. To address missingness, we propose a Debiased Cognitive Diagnosis (DBCD) framework through the perspective of counterfactual modeling to remove exogenous confounders from the response data. Specifically, the proposed DBCD achieves debiasing for CD by applying the idea of contrastive learning to constrain the model's prediction distributions on both factual and counterfactual data. For a student, the factual data is his/her original response records, while the counterfactual data is generated by sampling the same number of exercises from all exercises of each concept through a similarity-based counterfactual sampling strategy. Considering the difficulty of directly removing the exogenous confounders for student, we devise a β-Variational Autoencoder to model their exogenous confounders within the latent representations of knowledge proficiency by leveraging exercise priors and student response patterns. Then, the learned representations are further combined with the vanilla student's ability embedding via a gating mechanism-based fusion for final diagnosis prediction of the model. Extensive experiments on real-world educational datasets demonstrate that the proposed DBCD effectively mitigates confounders and even outperforms existing methods, thereby validating the feasibility and effectiveness of the DBCD framework.

Md Shahriar Rahim Siddiqui, Moshe Eliasof, Eldad Haber

Flow matching casts sample generation as learning a continuous-time velocity field that transports noise to data. Existing flow matching networks typically predict each point's velocity independently, considering only its location and time along its flow trajectory, and ignoring neighboring points. However, this pointwise approach may overlook correlations between points along the generation trajectory that could enhance velocity predictions, thereby improving downstream generation quality. To address this, we propose Graph Flow Matching (GFM), a lightweight enhancement that decomposes the learned velocity into a reaction term -- any standard flow matching network -- and a diffusion term that aggregates neighbor information via a graph neural module. This reaction-diffusion formulation retains the scalability of deep flow models while enriching velocity predictions with local context, all at minimal additional computational cost. Operating in the latent space of a pretrained variational autoencoder, GFM consistently improves Fréchet Inception Distance (FID) and recall across five image generation benchmarks (LSUN Church, LSUN Bedroom, FFHQ, AFHQ-Cat, and CelebA-HQ at 256 × 256), demonstrating its effectiveness as a modular enhancement to existing flow matching architectures.

Haokai Hong, Wanyu Lin, Ming Yang, Kay Chen Tan

Can we train a 3D molecule generator using data from dense regions to generate samples in sparse regions? This challenge can be framed as an out-of-distribution (OOD) generation problem. While prior research on OOD generation predominantly targets property shifts, structural shifts, such as differences in molecular scaffolds or functional groups, represent an equally critical source of distributional shifts. This work introduces the Geometric OOD Diffusion Model (GODD), a novel diffusion-based framework that enables training on data-abundant molecular distributions while generalizing to data-scarce distributions under distributional structural shifts. Central to our approach is a designated equivariant asymmetric autoencoder to capture distributional structural priors. The asymmetric design allows the model to generalize to unseen structural variations by capturing distributional priors representing distinct distributions. The encoded structural-grained priors guide generation toward sparse regions without requiring explicit training on such data. Evaluated across standard benchmarks encompassing OOD structural shifts (e.g., scaffolds, rings), GODD achieves an improvement of 12.6% in success rate, defined based on molecular validity, uniqueness, and novelty. Furthermore, the framework demonstrates promising performance and generalization on canonical fragment-based drug design tasks, highlighting its utility in learning-based molecular discovery.

Xinshun Feng, Mingzhe Liu, Yi Qiao, Tongyu Zhu, Leilei Sun, Shuai Wang

Recent advances in explainable recommendation have explored the integration of language models to analyze natural language rationales for user–item interactions. Despite their potential, existing methods often rely on ID-based representations that obscure semantic meaning and impose structural constraints on language models, thereby limiting their applicability in open-ended scenarios. These challenges are intensified by the complex nature of real-world interactions, where diverse user intents are entangled and collaborative signals rarely align with linguistic semantics. To overcome these limitations, we propose BEAT, a unified and transferable framework that tokenizes user and item behaviors into discrete, interpretable sequences. We construct a behavior vocabulary via a vector-quantized autoencoding process that disentangles macro-level interests and micro-level intentions from graph-based representations. We then introduce multi-level semantic supervision to bridge the gap between behavioral signals and language space. A semantic alignment regularization mechanism is designed to embed behavior tokens directly into the input space of frozen language models. Experiments on three public datasets show that BEAT improves zero-shot recommendation performance while generating coherent and informative explanations. Further analysis demonstrates that our behavior tokens capture fine-grained semantics and offer a plug-and-play interface for integrating complex behavior patterns into large language models.

Xin Chen, Qi Zhao, Wei Zeng, Zongben Xu

Images are generally represented by pixel intensities or color values, which are usually used as direct inputs for learning. This study innovatively proposes a geometric image representation method and refreshes the general learning model (e.g., autoencoder) in the diffeomorphic space. Based on the theory of geometric optimal transport and quasiconformal mapping, we equivalently transform the intensity representation into a shape representation. The image space becomes a diffeomorphic space, where any image can be uniquely represented as a Beltrami coefficient function defined on a uniform grid reference, and vice versa. This innovative geometric image representation (G-IR) captures the fine-grained structure inherent in the entire image, which is different from the traditional feature extraction that focuses on the internal geometric objects of the image (such as boundaries and axes). The diffeomorphic property preserves structure in the generation process, which is very necessary in the field of real physics. It can be assembled into existing pipelines as a plug-in, providing structure-preserving properties for the entire framework. Experiments on image restoration and interpolation validated the high efficiency, efficacy and applicability of the G-IR method, demonstrating its superior performance compared to common pixel-level image appearance representations.

Yingji Zhang, Marco Valentino, Danilo Carvalho, André Freitas

Incorporating explicit reasoning rules within the latent space of language models (LMs) offers a promising pathway to enhance generalisation, interpretability, and controllability. While current Transformer-based language models have shown strong performance on Natural Language Inference (NLI) tasks, they often rely on memorisation rather than explicit rule-based generalisation. This work investigates how human-interpretable reasoning rules can be explicitly encoded within LMs with the support of Language Variational Autoencoders (VAEs), as a mechanism for generative control. We propose a complete pipeline for learning reasoning rules within Transformer-based language VAEs. This pipeline encompasses three rule-based reasoning tasks, a supporting theoretical framework, and a practical end-to-end architecture. The experiment illustrates the following findings: Disentangled reasoning: Under explicit signal supervision, reasoning rules (viewed as functional mappings) can be disentangled within the encoder’s parametric space. This separation results in distinct clustering of rules in the output feature space. Prior knowledge injection: injecting rule-based constraints into the Query enables the model to more effectively retrieve the stored Value from memory based on Key. This approach offers a simple method for integrating prior knowledge into decoder-only language models. Moreover, we found that FFN layers are better than attention layers at preserving the separation of reasoning rules in the model's parameters.

Ruijie Zhang, Bixin Zeng, Shengpeng Wang, Fuhui Zhou, Wei Wang

Millimeter-wave radar offers a promising sensing modality for autonomous systems thanks to its robustness in adverse conditions and low cost. However, its utility is significantly limited by the sparsity and low resolution of radar point clouds, which poses challenges for tasks requiring dense and accurate 3D perception. Despite that recent efforts have shown great potential by exploring generative approaches to address this issue, they often rely on dense voxel representations that are inefficient and struggle to preserve structural detail. To fill this gap, we make the key observation that latent diffusion models (LDMs), though successful in other modalities, have not been effectively leveraged for radar-based 3D generation due to a lack of compatible representations and conditioning strategies. We introduce RaLD, a framework that bridges this gap by integrating scene-level frustum-based LiDAR autoencoding, order-invariant latent representations, and direct radar spectrum conditioning. These insights lead to a more compact and expressive generation process. Experiments show that RaLD produces dense and accurate 3D point clouds from raw radar spectrums, offering a promising solution for robust perception in challenging environments.

Jialin Liu, Zhaorui Zhang, Ray C.C. Cheung

Graph neural networks (GNNs) have revolutionized recommender systems by effectively modeling complex user-item interactions, yet data sparsity and the item cold-start problem significantly impair performance, particularly for new items with limited or no interaction history. While multimodal content offers a promising solution, existing methods result in suboptimal representations for new items due to noise and entanglement in sparse data. To address this, we transform multimodal recommendation into discrete semantic tokenization. We present Sparse-Regularized Multimodal Tokenization for Cold-Start Recommender Systems (MoToRec), a framework centered on a sparsely-regularized Residual Quantized Variational Autoencoder (RQ-VAE) that generates a compositional semantic code of discrete, interpretable tokens, promoting disentangled representations. MoToRec’s architecture is enhanced by three synergistic components: (1) a sparsely regularized RQ-VAE that promotes disentangled representations, (2) a novel adaptive rarity amplification that promotes prioritized learning for cold-start items, and (3) a hierarchical multi-source graph encoder for robust signal fusion with collaborative signals. Extensive experiments on three large-scale datasets demonstrate MoToRec’s superiority over state-of-the-art methods in both overall and cold-start scenarios. Our work validates that discrete tokenization provides an effective and scalable alternative for mitigating the long-standing cold-start challenge.

Chuan He, Yongchao Liu, Qiang Li, Chuntao Hong, Wenliang Zhong, Xin-Wei Yao

Cold-start item recommendation is a significant challenge in recommendation systems, particularly when new items are introduced without any historical interaction data. While existing methods leverage multi-modal content to alleviate the cold-start issue, they often neglect the inherent multi-view structure of modalities, namely the distinction between shared and modality-specific features. In this paper, we propose Multi-Modal Multi-View Variational AutoEncoder (M²VAE), a generative model that addresses the challenges of modeling common and unique views in attribute and multi-modal features, as well as user preferences over single-typed item features. Specifically, we generate type-specific latent variables for item IDs, categorical attributes, and image features, and use Product-of-Experts (PoE) to derive a common representation. A disentangled contrastive loss decouples the common view from unique views while preserving feature informativeness. To model user inclinations, we employ a user-aware hierarchical Mixture-of-Experts (MoE) to adaptively fuse representations. We further incorporate co-occurrence signals via contrastive learning, eliminating the need for pretraining. Extensive experiments on real-world datasets validate the effectiveness of our approach.

Ya Zou, Jingfeng Yao, Siyuan Yu, Shuai Zhang, Wenyu Liu, Xinggang Wang

There is a growing demand for deploying large generative AI models on mobile devices. For recent popular video generative models, however, the Variational AutoEncoder (VAE) represents one of the major computational bottlenecks. Both large parameter sizes and mismatched kernels cause out-of-memory errors or extremely slow inference on mobile devices. To address this, we propose a low-cost solution that efficiently transfers widely used video VAEs to mobile devices. (1) We analyze redundancy in existing VAE architectures and get empirical design insights. By integrating 3D depthwise separable convolutions into our model, we significantly reduce the number of parameters. (2) We observe that the upsampling techniques in mainstream video VAEs are poorly suited to mobile hardware and form the main bottleneck. In response, we propose a decoupled 3D pixel shuffle scheme that slashes end-to-end delay. Building upon these, we develop a universal mobile-oriented VAE decoder, Turbo-VAED. (3) We propose an efficient VAE decoder training method. Since only the decoder is used during deployment, we distill it to Turbo-VAED instead of retraining the full VAE, enabling fast mobile adaptation with minimal performance loss. To our knowledge, our method enables real-time 720p video VAE decoding on mobile devices for the first time. This approach is widely applicable to most video VAEs. When integrated into four representative models, with training cost as low as $95, it accelerates original VAEs by up to 84.5× at 720p resolution on GPUs, uses as low as 17.5% of original parameter count, and retains 96.9% of the original reconstruction quality. Compared to mobile-optimized VAEs, Turbo-VAED achieves a 2.9× speedup in FPS and better reconstruction quality on the iPhone 16 Pro.

Weishang Wu, Yifei Shi, Zhiping Cai

Task-oriented dexterous grasping remains challenging in robotic manipulations of open-world objects under severe partial observation, where significant missing data invalidates generic shape completion. In this paper, to overcome this limitation, we study \emph{Task-Oriented Shape Completion}, a new task that focuses on completing the potential contact regions rather than the entire shape. We argue that shape completion for grasping should be explicitly guided by the downstream manipulation task. To achieve this, we first generate multiple task-oriented shape completion candidates by leveraging the zero-shot capabilities of object functional understanding from several pre-trained foundation models. A 3D discriminative autoencoder is then proposed to evaluate the plausibility of each generated candidate and optimize the most plausible one from a global perspective. A conditional flow-matching model named FlowGrasp is developed to generate task-oriented dexterous grasps from the optimized shape. Our method achieves state-of-the-art performance in task-oriented dexterous grasping and task-oriented shape completion, improving the Grasp Displacement and the Chamfer Distance over the state-of-the-art by 16.17% and 55.26%, respectively. In particular, it shows good capabilities in grasping objects with severe missing data. It also demonstrates good generality in handling open-set categories and tasks.

Xuan Wang, Kai Ruan, Liyang Qian, Guo Zhi Zhi, Chang Su, Gaoang Wang

Text-driven motion generation has attracted increasing attention due to its broad applications in virtual reality, animation, and robotics. While existing methods typically model human and animal motion separately, a joint cross-species approach offers key advantages, such as a unified representation and improved generalization. However, morphological differences across species remain a key challenge, often compromising motion plausibility. To address this, we propose X-MoGen, the first unified framework for cross-species text-driven motion generation covering both humans and animals. X-MoGen adopts a two-stage architecture. First, a conditional graph variational autoencoder learns canonical T-pose priors, while an autoencoder encodes motion into a shared latent space regularized by morphological loss. In the second stage, we perform masked motion modeling to generate motion embeddings conditioned on textual descriptions. During training, a morphological consistency module is employed to promote skeletal plausibility across species. To support unified modeling, we construct UniMo4D, a large-scale dataset of 115 species and 119k motion sequences, which integrates human and animal motions under a shared skeletal topology for joint training. Extensive experiments on UniMo4D demonstrate that X-MoGen outperforms state-of-the-art methods on both seen and unseen species.

Ruizhi Wang, Weihan Li, Zunlei Feng, Haofei Zhang, Mingli Song, Jiayu Wang, Jie Song, Li Sun

Real-time, high-fidelity monocular depth estimation from remote sensing imagery is crucial for numerous applications, yet existing methods face a stark trade-off between accuracy and efficiency. Although using Vision Transformer (ViT) backbones for dense prediction is fast, they often exhibit poor perceptual quality. Conversely, diffusion models offer high fidelity but at a prohibitive computational cost. To overcome these limitations, we propose Depth Detail Diffusion for Remote Sensing Monocular Depth Estimation (D³-RSMDE), an efficient framework designed to achieve an optimal balance between speed and quality. Our framework first leverages a ViT-based module to rapidly generate a high-quality preliminary depth map construction, which serves as a structural prior, effectively replacing the time-consuming initial structure generation stage of diffusion models. Based on this prior, we propose a Progressive Linear Blending Refinement (PLBR) strategy, which uses a lightweight U-Net to refine the details in only a few iterations. The entire refinement step operates efficiently in a compact latent space supported by a Variational Autoencoder (VAE). Extensive experiments demonstrate that D³-RSMDE achieves a notable 11.85% reduction in the Learned Perceptual Image Patch Similarity (LPIPS) perceptual metric over leading models like Marigold, while also achieving over a 40× speedup in inference and maintaining VRAM usage comparable to lightweight ViT models.

Haotian Wang, Yuzhe Weng, Jun Du, Haoran Xu, Xiaoyan Wu, Shan He, Bing Yin, Cong Liu, Jianqing Gao, Qingfeng Liu

The introduction of diffusion models has brought significant advances to the field of audio-driven talking head generation. However, the extremely slow inference speed severely limits the practical implementation of diffusion-based talking head generation models. In this study, we propose READ, a real-time diffusion-transformer-based talking head generation framework. Our approach first learns a spatiotemporal highly compressed video latent space via a temporal VAE, significantly reducing the token count to accelerate generation. To achieve better audio-visual alignment within this compressed latent space, a pre-trained Speech Autoencoder (SpeechAE) is proposed to generate temporally compressed speech latent codes corresponding to the video latent space. These latent representations are then modeled by a carefully designed Audio-to-Video Diffusion Transformer (A2V-DiT) backbone for efficient talking head synthesis. Furthermore, to ensure temporal consistency and accelerated inference in extended generation, we propose a novel asynchronous noise scheduler (ANS) for both the training and inference processes of our framework. The ANS leverages asynchronous add-noise and asynchronous motion-guided generation in the latent space, ensuring consistency in generated video clips. Experimental results demonstrate that READ outperforms state-of-the-art methods by generating competitive talking head videos with significantly reduced runtime, achieving an optimal balance between quality and speed while maintaining robust metric stability in long-time generation.

Chao Wang, Zijin Yang, Yaofei Wang, Weiming Zhang, Kejiang Chen

The rapid advancement of image-generation technologies has made it possible for anyone to create photorealistic images using generative models, raising significant security concerns. To mitigate malicious use, tracing the origin of such images is essential. Reconstruction-based attribution methods offer a promising solution, but they often suffer from reduced accuracy and high computational costs when applied to state‑of‑the‑art (SOTA) models. To address these challenges, we propose AEDR (AutoEncoder Double-Reconstruction), a novel training‑free attribution method designed for generative models with continuous autoencoders. Unlike existing reconstruction‑based approaches that rely on the value of a single reconstruction loss, AEDR performs two consecutive reconstructions using the model’s autoencoder, and adopts the ratio of these two reconstruction losses as the attribution signal. This signal is further calibrated using the image homogeneity metric to improve accuracy, which inherently cancels out absolute biases caused by image complexity, with autoencoder‑based reconstruction ensuring superior computational efficiency. Experiments on eight top latent diffusion models show that AEDR achieves 25.5% higher attribution accuracy than existing reconstruction‑based methods, with requiring only 1% of the computational time.

Nagur Shareef Shaik, Teja Krishna Cherukuri, Adnan Masood, Dong Hye Ye

The integration of medical images with clinical context is essential for generating accurate and clinically interpretable radiology reports. However, current automated methods often rely on resource-heavy Large Language Models (LLMs) or static knowledge graphs and struggle with two fundamental challenges in real-world clinical data: (1) missing modalities, such as incomplete clinical context , and (2) feature entanglement, where mixed modality-specific and shared information leads to suboptimal fusion and clinically unfaithful hallucinated findings. To address these challenges, we propose the DiA-gnostic VLVAE, which achieves robust radiology reporting through Disentangled Alignment. Our framework is designed to be resilient to missing modalities by disentangling shared and modality-specific features using a Mixture-of-Experts (MoE) based Vision-Language Variational Autoencoder (VLVAE). A constrained optimization objective enforces orthogonality and alignment between these latent representations to prevent suboptimal fusion. A compact LLaMA-X decoder then uses these disentangled representations to generate reports efficiently. On the IU X-Ray and MIMIC-CXR datasets, DiA has set new state-of-the-art BLEU@4 scores of 0.266 and 0.134, respectively. Experimental results show that the proposed method significantly outperforms state-of-the-art models.

Zhen Liu, Diedong Feng, Hai Jiang, Liaoyuan Zeng, Hao Wang, Chaoyu Feng, Lei Lei, Bing Zeng, Shuaicheng Liu

RGB-to-RAW reconstruction, or the reverse modeling of a camera Image Signal Processing (ISP) pipeline, aims to recover high-fidelity RAW data from RGB images. Despite notable progress, existing learning-based methods typically treat this task as a direct regression objective and still struggle with detail inconsistency and color deviation, due to the ill-posed nature of inverse ISP and the inherent information loss in quantized RGB images. To address these limitations, we pioneer a generative perspective by reformulating RGB-to-RAW reconstruction as a deterministic latent transport problem and introduce a novel framework named RAW-Flow, which leverages flow matching to learn a deterministic vector field in latent space, to effectively bridge the gap between RGB and RAW representations and enable accurate reconstruction of structural details and color information. To further enhance latent transport, we introduce a cross-scale context guidance module that injects hierarchical RGB features into the flow estimation process. Moreover, we design a Dual-domain Latent Autoencoder (DLAE) with a feature alignment constraint to support the proposed latent transport framework, which jointly encodes RGB and RAW inputs while promoting stable training and high-fidelity reconstruction. Extensive experiments demonstrate that RAW-Flow outperforms state-of-the-art approaches both quantitatively and visually.

Hyundo Lee, Suhyung Choi, Inwoo Hwang, Byoung-Tak Zhang

Image generation models trained on large datasets can synthesize high-quality images but often produce spatially inconsistent and distorted images due to limited information about the underlying structures and spatial layouts. In this work, we leverage intrinsic scene properties (e.g., depth, segmentation maps) that provide rich information about the underlying scene, unlike prior approaches that solely rely on image-text pairs or use intrinsics as conditional inputs. Our approach aims to co-generate both images and their corresponding intrinsics, enabling the model to implicitly capture the underlying scene structure and generate more spatially consistent and realistic images. Specifically, we first extract rich intrinsic scene properties from a large image dataset with pre-trained estimators, eliminating the need for additional scene information or explicit 3D representations. We then aggregate various intrinsic scene properties into a single latent variable using an autoencoder. Building upon pre-trained large-scale Latent Diffusion Models (LDMs), our method simultaneously denoises the image and intrinsic domains by carefully sharing mutual information so that the image and intrinsic reflect each other without degrading image quality. Experimental results demonstrate that our method corrects spatial inconsistencies and produces a more natural layout of scenes while maintaining the fidelity and textual alignment of the base model (e.g., Stable Diffusion).