Institution profile

Kuaishou Technology

Industry researchasia · cn
Official website
Research library647linked papers
Opportunities0open roles
Selected work

Representative Papers

DreamOn: Diffusion Language Models For Code Infilling Beyond Fixed-size Canvas

Feb 01, 2026

Existing diffusion-based language models for code infilling are constrained by fixed-length masks, limiting their ability to generate variable-length outputs. This work proposes DreamOn, a framework that introduces two lightweight length-control states into the diffusion process, enabling dynamic adjustment of the generated sequence length without modifying the underlying model architecture. Experiments on DreamCoder-7B and DiffuCoder-7B demonstrate that DreamOn achieves performance comparable to state-of-the-art autoregressive models on the HumanEval-Infilling and SantaCoder-FIM benchmarks, while closely approaching the ideal performance attainable when the ground-truth target length is known. These results significantly enhance the practicality of diffusion models for real-world code infilling tasks.

14 citations1 influentialRead paper

CodeScaler: Scaling Code LLM Training and Test-Time Inference via Execution-Free Reward Models

Feb 04, 2026arXiv.org

This work addresses the limitations of existing reinforcement learning approaches for code large language models that rely on unit tests, which are often scarce and unreliable, hindering scalable training and inference. To overcome this, the authors propose CodeScaler—a test-case-free reward model for code generation—leveraging verified preference data, syntax-aware code extraction, and validity-preserving reward shaping to enable efficient scaling. Evaluated across four code benchmarks, CodeScaler outperforms execution-feedback-based RL by up to 4.23 points. Notably, when scaled to 44K problems without any test cases, it achieves a 14.64-point improvement while reducing inference latency by an order of magnitude. Furthermore, it surpasses prior methods by 3.3 points on coding tasks and by an average of 2.7 points across general and reasoning tasks in RM-Bench.

3 citationsRead paper

NTIRE 2025 Challenge on Short-form UGC Video Quality Assessment and Enhancement: KwaiSR Dataset and Study

Apr 21, 2025

This paper addresses the low image quality of user-generated content (UGC) on short-video platforms and the poor generalization of existing single-image super-resolution (SR) methods. To bridge this gap, the authors introduce KwaiSR—the first benchmark dataset tailored to real-world UGC scenarios—comprising 1,800 synthetically generated LR-HR image pairs and 1,900 authentic low-quality images selected via KVQ’s quality assessment model, enabling dual-domain co-modeling of synthetic ground truth and realistic degradation distributions. Leveraging KwaiSR, the authors organized the NTIRE 2025 Challenge on Short-Format UGC Video Quality Assessment and Enhancement, attracting over 30 participating teams. Extensive experiments reveal substantial performance degradation of mainstream SR methods on UGC data, highlighting critical challenges in realistic degradation modeling, quality-aware image selection, and cross-domain generalization. KwaiSR thus establishes a foundational data resource and a new research paradigm for UGC image enhancement.

3 citationsRead paper

DiaDem: Advancing Dialogue Descriptions in Audiovisual Video Captioning for Multimodal Large Language Models

Jan 27, 2026

This work addresses the challenges of ambiguous speaker attribution and inaccurate transcription in existing audio-visual description models within conversational scenarios. To this end, the authors propose DiaDem, a multimodal large language model that effectively integrates audio, visual, and textual information through supervised fine-tuning on synthetically generated data and a two-stage difficulty-stratified GRPO reinforcement learning strategy. The contributions include the construction of DiaDemBench—the first systematic evaluation benchmark for conversational description—and a novel difficulty-aware optimization approach. Experimental results demonstrate that DiaDem significantly outperforms the Gemini family of models on DiaDemBench, achieving state-of-the-art performance in both accuracy and faithfulness of conversational descriptions, while maintaining competitive results on general audio-visual captioning tasks.

2 citationsRead paper

FITMM: Adaptive Frequency-Aware Multimodal Recommendation via Information-Theoretic Representation Learning

Oct 27, 2025ACM Multimedia

This work proposes FITMM, a novel framework that addresses the limitations of existing spatial-domain approaches to multimodal recommendation, which often neglect frequency-domain structures and suffer from modality misalignment and redundancy. FITMM is the first to introduce the information bottleneck principle into frequency-domain multimodal recommendation. It constructs item representations via graph augmentation and performs orthogonal decomposition of each modality in the frequency domain to yield lightweight intra-band components. A task-adaptive gating mechanism fuses band-specific information, while intra-band independent modeling and cross-modal spectral consistency constraints enable adaptive band selection and redundancy suppression. Extensive experiments on three real-world datasets demonstrate that FITMM significantly outperforms state-of-the-art baselines, validating the effectiveness and generalizability of frequency-domain modeling for multimodal recommendation.

2 citationsRead paper
Recent publications

Latest Papers