Author name cluster

Debing Zhang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

11 papers

2 author rows

AAAI Conference 2026 Conference Paper

LiteLong: Resource-Efficient Long-Context Data Synthesis for LLMs

Junlong Jia
Xing Wu
Chaochen Gao
Ziyang Chen
Zijia Lin
Zhongzhi Li
Weinong Wang
Haotian Xu

High-quality long-context data is essential for training large language models (LLMs) capable of processing extensive documents, yet existing synthesis approaches using relevance-based aggregation face challenges of computational efficiency. We present LiteLong, a resource-efficient method for synthesizing long-context data through structured topic organization and multi-agent debate. Our approach leverages the BISAC book classification system to provide a comprehensive hierarchical topic organization, and then employs a debate mechanism with multiple LLMs to generate diverse, high-quality topics within this structure. For each topic, we use lightweight BM25 retrieval to obtain relevant documents and concatenate them into 128K-token training samples. Experiments on HELMET and Ruler benchmarks demonstrate that LiteLong achieves competitive long-context performance and can seamlessly integrate with other long-dependency enhancement methods. LiteLong makes high-quality long-context data synthesis more accessible by reducing both computational and data engineering costs, facilitating further research in long-context language training.

PDF Details DOI

AAAI Conference 2026 Conference Paper

Put the Space of LoRA Initialization to the Extreme to Preserve Pre-trained Knowledge

Pengwei Tang
Xiaolin Hu
Yong Liu
Lizhong Ding
Dongjie Zhang
Xing Wu
Debing Zhang

Low-Rank Adaptation (LoRA) is the leading parameter-efficient fine-tuning method for Large Language Models (LLMs), but it still suffers from catastrophic forgetting. Recent work has shown that specialized LoRA initialization can alleviate catastrophic forgetting. There are currently two approaches to LoRA initialization aimed at preventing knowledge forgetting during fine-tuning: (1) making residual weights close to pre-trained weights, and (2) ensuring the space of LoRA initialization is orthogonal to pre-trained knowledge. The former is what current methods strive to achieve, while the importance of the latter is not sufficiently recognized. We find that the space of LoRA initialization is the key to preserving pre-trained knowledge rather than the residual weights. Existing methods like MiLoRA propose making the LoRA initialization space orthogonal to pre-trained weights. However, MiLoRA utilizes the null space of pre-trained weights. Compared to pre-trained weights, the input activations of pre-trained knowledge take into account the parameters of all previous layers as well as the input data, while pre-trained weights only contain information from the current layer. Moreover, we find that the effective ranks of input activations are much smaller than those of pre-trained weights. Thus, the null space of activations is more accurate and contains less pre-trained knowledge information compared to that of weights. Based on these, we introduce LoRA-Null, our proposed method that initializes LoRA in the null space of activations. Experimental results show that LoRA-Null effectively preserves the pre-trained world knowledge of LLMs while achieving good fine-tuning performance, as evidenced by extensive experiments.

PDF Details DOI

ICML Conference 2025 Conference Paper

David and Goliath: Small One-step Model Beats Large Diffusion with Score Post-training

Weijian Luo
Colin Zhang
Debing Zhang
Zhengyang Geng

We propose Diff-Instruct (DI ), a data-efficient post-training approach to one-step text-to-image generative models to improve its human preferences without requiring image data. Our method frames alignment as online reinforcement learning from human feedback (RLHF), which optimizes a human reward function while regularizing the generator to stay close to a reference diffusion process. Unlike traditional RLHF approaches, which rely on the KL divergence for regularization, we introduce a novel score-based divergence regularization that substantially improves performance. Although such a score-based RLHF objective seems intractable when optimizing, we derive a strictly equivalent tractable loss function in theory that can efficiently compute its gradient for optimizations. Building upon this framework, we train DI -SDXL-1step, a 1-step text-to-image model based on Stable Diffusion-XL (2. 6B parameters), capable of generating 1024x1024 resolution images in a single step. The 2. 6B DI -SDXL-1step model outperforms the 12B FLUX-dev model in ImageReward, PickScore, and CLIP score on the Parti prompts benchmark while using only 1. 88% of the inference time. This result strongly supports the thought that with proper post-training, the small one-step model is capable of beating huge multi-step models. We will open-source our industry-ready model to the community.

Details

NeurIPS Conference 2025 Conference Paper

LongMagpie: A Self-synthesis Method for Generating Large-scale Long-context Instructions

Chaochen Gao
Xing W
Zijia Lin
Debing Zhang
Songlin Hu

High-quality long-context instruction data is essential for aligning long-context large language models (LLMs). Despite the public release of models like Qwen and Llama, their long-context instruction data remains proprietary. Human annotation is costly and challenging, while template-based synthesis methods limit scale, diversity, and quality. We introduce LongMagpie, a self-synthesis framework that automatically generates large-scale long-context instruction data. Our key insight is that aligned long-context LLMs, when presented with a document followed by special tokens preceding a user turn, auto-regressively generate contextually relevant queries. By harvesting these document-query pairs and the model's responses, LongMagpie produces high-quality instructions without human effort. Experiments on HELMET, RULER, and Longbench v2 demonstrate that LongMagpie achieves leading performance on long-context tasks while maintaining competitive performance on short-context tasks, establishing it as a simple and effective approach for open, diverse, and scalable long-context instruction data synthesis.

PDF Details

ICML Conference 2025 Conference Paper

NExtLong: Toward Effective Long-Context Training without Long Documents

Chaochen Gao
Xing Wu 0002
Zijia Lin
Debing Zhang
Songlin Hu 0001

Large language models (LLMs) with extended context windows have made significant strides yet remain a challenge due to the scarcity of long documents. Existing methods tend to synthesize long-context data but lack a clear mechanism to reinforce the long-range dependency modeling. To address this limitation, we propose NExtLong, a novel framework for synthesizing long-context data through Negative document Extension. NExtLong decomposes a document into multiple meta-chunks and extends the context by interleaving hard negative distractors retrieved from pretraining corpora. This approach compels the model to discriminate long-range dependent context from distracting content, enhancing its ability to model long-range dependencies. Extensive experiments demonstrate that NExtLong achieves significant performance improvements on the HELMET and RULER benchmarks compared to existing long-context synthesis approaches and leading models, which are trained on non-synthetic long documents. These findings highlight NExtLong’s ability to reduce reliance on non-synthetic long documents, making it an effective framework for developing advanced long-context LLMs.

Details

ICLR Conference 2025 Conference Paper

Rethinking Reward Model Evaluation: Are We Barking up the Wrong Tree?

Xueru Wen
Jie Lou
Yaojie Lu 0001
Hongyu Lin
XingYu
Xinyu Lu
Ben He 0001
Xianpei Han

Reward Models (RMs) are crucial for aligning language models with human preferences. Currently, the evaluation of RMs depends on measuring accuracy against a validation set of manually annotated preference data. Although this method is straightforward and widely adopted, the relationship between RM accuracy and downstream policy performance remains under-explored. In this work, we conduct experiments in a synthetic setting to investigate how differences in RM measured by accuracy translate into gaps in optimized policy performance. Our findings reveal that while there is a weak positive correlation between accuracy and downstream performance, policies optimized towards RMs with similar accuracy can exhibit quite different performance. Moreover, we discover that the way of measuring accuracy significantly impacts its ability to predict the final policy performance. Through the lens of the Regressional Goodhart effect, we recognize that accuracy, when used for measuring RM quality, can fail to fully capture the potential RM overoptimization. This underscores the inadequacy of relying solely on accuracy to reflect their impact on policy optimization.

Details

ICML Conference 2025 Conference Paper

The Devil Is in the Details: Tackling Unimodal Spurious Correlations for Generalizable Multimodal Reward Models

Zichao Li
Xueru Wen
Jie Lou
Yuqiu Ji
Yaojie Lu 0001
Xianpei Han
Debing Zhang
Le Sun 0001

Multimodal Reward Models (MM-RMs) are crucial for aligning Large Language Models (LLMs) with human preferences, particularly as LLMs increasingly interact with multimodal data. However, we find that MM-RMs trained on existing datasets often struggle to generalize to out-of-distribution data due to their reliance on unimodal spurious correlations, primarily text-only shortcuts within the training distribution, which prevents them from leveraging true multimodal reward functions. To address this, we introduce a Shortcut-aware MM-RM learning algorithm that mitigates this issue by dynamically reweighting training samples, shifting the distribution toward better multimodal understanding, and reducing dependence on unimodal spurious correlations. Our experiments demonstrate significant improvements in generalization, downstream task performance, and scalability, establishing a more robust framework for multimodal reward modeling. Our source code is provided on https: //github. com/alignrm/Generalizable-MM-RM.

Details

NeurIPS Conference 2025 Conference Paper

Uni-Instruct: One-step Diffusion Model through Unified Diffusion Divergence Instruction

Yifei Wang
Weimin Bai
colin zhang
Debing Zhang
Weijian Luo
He Sun

In this paper, we unify more than 10 existing one-step diffusion distillation approaches, such as Diff-Instruct, DMD, SIM, SiD, $f$-distill, etc, inside a theory-driven framework which we name the \textbf{\emph{Uni-Instruct}}. Uni-Instruct is motivated by our proposed diffusion expansion theory of the $f$-divergence family. Then we introduce key theories that overcome the intractability issue of the original expanded $f$-divergence, resulting in an equivalent yet tractable loss that effectively trains one-step diffusion models by minimizing the expanded $f$-divergence family. The novel unification introduced by Uni-Instruct not only offers new theoretical contributions that help understand existing approaches from a high-level perspective but also leads to state-of-the-art one-step diffusion generation performances. On the CIFAR10 generation benchmark, Uni-Instruct achieves record-breaking Frechet Inception Distance (FID) values of \textbf{\emph{1. 46}} for unconditional generation and \textbf{\emph{1. 38}} for conditional generation. On the ImageNet-$64\times 64$ generation benchmark, Uni-Instruct achieves a new SoTA one-step diffusion FID value of \textbf{\emph{1. 06}}, which outperforms its 79-step teacher diffusion with a significant improvement margin of 1. 29 (1. 06 vs 2. 35). We also apply Uni-Instruct on broader tasks like text-to-3D generation. For text-to-3D generation, Uni-Instruct gives decent results, which slightly outperforms previous methods, such as SDS and VSD, in terms of both generation quality and diversity. Both the solid theoretical and empirical contributions of Uni-Instruct will potentially help future studies on one-step diffusion distillation and knowledge transferring of diffusion models.

PDF Details

NeurIPS Conference 2021 Conference Paper

A Bilingual, OpenWorld Video Text Dataset and End-to-end Video Text Spotter with Transformer

威佳吴
Debing Zhang
Yuanqiang Cai
Sibo Wang
Jiahong Li
Zhuang Li
Yejun Tang
Hong Zhou

Most existing video text spotting benchmarks focus on evaluating a single language and scenario with limited data. In this work, we introduce a large-scale, Bilingual, Open World Video text benchmark dataset(BOVText). There are four features for BOVText. Firstly, we provide 1, 850+ videos with more than 1, 600, 000+ frames, 25 times larger than the existing largest dataset with incidental text in videos. Secondly, our dataset covers 30+ open categories with a wide selection of various scenarios, Life Vlog, Driving, Movie, etc. Thirdly, abundant text types annotation (i. e. , title, caption, or scene text) are provided for the different representational meanings in the video. Fourthly, the MOVText provides multilingual text annotation to promote multiple cultures' live and communication. Besides, we propose an end-to-end video text spotting framework with Transformer, termed TransVTSpotter, which solves the multi-orient text spotting in video with a simple, but efficient attention-based query-key mechanism. It applies object features from the previous frame as a tracking query for the current frame and introduces a rotation angle prediction to fit the multi-orient text instance. On ICDAR2015(video), TransVTSpotter achieves state-of-the-art performance with 44. 2% MOTA, 13 fps. The dataset and code of TransVTSpotter can be found at https: //github. com/weijiawu/BOVText-Benchmark and https: //github. com/weijiawu/TransVTSpotter, respectively.

PDF Details

IJCAI Conference 2013 Conference Paper

A Unified Approximate Nearest Neighbor Search Scheme by Combining Data Structure and Hashing

Debing Zhang
Genmao Yang
Yao Hu
Zhongming Jin
Deng Cai
Xiaofei He

Nowadays, Nearest Neighbor Search becomes more and more important when facing the challenge of big data. Traditionally, to solve this problem, researchers mainly focus on building effective data structures such as hierarchical k-means tree or using hashing methods to accelerate the query process. In this paper, we propose a novel unified approximate nearest neighbor search scheme to combine the advantages of both the effective data structure and the fast Hamming distance computation in hashing methods. In this way, the searching procedure can be further accelerated. Computational complexity analysis and extensive experiments have demonstrated the effectiveness of our proposed scheme.

PDF Details DOI

IJCAI Conference 2013 Conference Paper

Active Learning via Neighborhood Reconstruction

Yao Hu
Debing Zhang
Zhongming Jin
Deng Cai
Xiaofei He

In many real world scenarios, active learning methods are used to select the most informative points for labeling to reduce the expensive human action. One direction for active learning is selecting the most representative points, ie. , selecting the points that other points can be approximated by linear combination of the selected points. However, these methods fails to consider the local geometrical information of the data space. In this paper, we propose a novel framework named Active Learning via Neighborhood Reconstruction (ALNR) by taking into account the locality information directly during the selection. Speciﬁcally, for the linear reconstruction of target point, the nearer neighbors should have a greater effect and the selected points distant from the target point should be penalized severely. We further develop an efﬁcient two-stage iterative procedure to solve the ﬁnal optimization problem. Our empirical study shows encouraging results of the proposed algorithms in comparison to other state-of-the-art active learning algorithms on both synthetic and real visual data sets.

PDF Details DOI