Chunlai Zhou

Jihao Zhao, Shuaishuai Zu, Zhiyuan Ji et al.

As a typical open-ended generation task, creative writing lacks verifiable reference answers, which has long constrained reward modeling and automatic evaluation due to high human annotation costs, evaluative bias, and coarse feedback signals. To address these challenges, this paper first designs a multi-agent collaborative workflow based on Grounded Theory, performing dimensional decomposition and hierarchical induction of the problem to dynamically produce interpretable and reusable fine-grained criteria. Furthermore, we propose the Memory-augmented Replay Policy Optimization (MRPO) algorithm: on the one hand, without additional training, MRPO guides models to engage in self-reflection based on dynamic criteria, enabling controlled iterative improvement; on the other hand, we adopt the training paradigm that combines supervised fine-tuning with reinforcement learning to convert evaluation criteria into reward signals, achieving end-to-end optimization. Experimental results demonstrate that the automatically constructed criteria achieve performance gains comparable to human annotations. Writer-R1-4B models trained with this approach outperform baselines across multiple creative writing tasks and surpass some 100B+ parameter open-source models.

Peng Tang, Zhiqiang Xu, Pengfei Wei et al.

We propose an efficient deep learning method for single image defocus deblurring (SIDD) by further exploring inverse kernel properties. Although the current inverse kernel method, i.e., kernel-sharing parallel atrous convolution (KPAC), can address spatially varying defocus blurs, it has difficulty in handling large blurs of this kind. To tackle this issue, we propose a Residual and Recursive Kernel-sharing Atrous Convolution (R$^2$KAC). R$^2$KAC builds on a significant observation of inverse kernels, that is, successive use of inverse-kernel-based deconvolutions with fixed size helps remove unexpected large blurs but produces ringing artifacts. Specifically, on top of kernel-sharing atrous convolutions used to simulate multi-scale inverse kernels, R$^2$KAC applies atrous convolutions recursively to simulate a large inverse kernel. Specifically, on top of kernel-sharing atrous convolutions, R$^2$KAC stacks atrous convolutions recursively to simulate a large inverse kernel. To further alleviate the contingent effect of recursive stacking, i.e., ringing artifacts, we add identity shortcuts between atrous convolutions to simulate residual deconvolutions. Lastly, a scale recurrent module is embedded in the R$^2$KAC network, leading to SR-R$^2$KAC, so that multi-scale information from coarse to fine is exploited to progressively remove the spatially varying defocus blurs. Extensive experimental results show that our method achieves the state-of-the-art performance.

Likang Liu, Keke Sun, Chunlai Zhou et al.

In this paper, we provide two views of constrained differential private (DP) mechanisms. The first one is as belief revision. A constrained DP mechanism is obtained by standard probabilistic conditioning, and hence can be naturally implemented by Monte Carlo algorithms. The other is as belief update. A constrained DP is defined according to l2-distance minimization postprocessing or projection and hence can be naturally implemented by optimization algorithms. The main advantage of these two perspectives is that we can make full use of the machinery of belief revision and update to show basic properties for constrained differential privacy especially some important new composition properties. Within the framework established in this paper, constrained DP algorithms in the literature can be classified either as belief revision or belief update. At the end of the paper, we demonstrate their differences especially in utility in a couple of scenarios.

Wenhao Zhao, Qiushui Xu, Linjie Xu et al.

Recently, incorporating knowledge from pretrained language models (PLMs) into decision transformers (DTs) has generated significant attention in offline reinforcement learning (RL). These PLMs perform well in RL tasks, raising an intriguing question: what kind of knowledge from PLMs has been transferred to RL to achieve such good results? This work first dives into this problem by analyzing each head quantitatively and points out Markov head, a crucial component that exists in the attention heads of PLMs. It leads to extreme attention on the last-input token and performs well only in short-term environments. Furthermore, we prove that this extreme attention cannot be changed by re-training embedding layer or fine-tuning. Inspired by our analysis, we propose a general method GPT2-DTMA, which equips a pretrained DT with Mixture of Attention (MoA), to accommodate diverse attention requirements during fine-tuning. Extensive experiments corroborate our theorems and demonstrate the effectiveness of GPT2-DTMA: it achieves comparable performance in short-term environments while significantly narrowing the performance gap in long-term environments.

Luyao Wang, Pengnian Qi, Xigang Bao et al.

Multi-modal entity alignment (MMEA) aims to identify equivalent entities between two multi-modal knowledge graphs for integration. Unfortunately, prior arts have attempted to improve the interaction and fusion of multi-modal information, which have overlooked the influence of modal-specific noise and the usage of labeled and unlabeled data in semi-supervised settings. In this work, we introduce a Pseudo-label Calibration Multi-modal Entity Alignment (PCMEA) in a semi-supervised way. Specifically, in order to generate holistic entity representations, we first devise various embedding modules and attention mechanisms to extract visual, structural, relational, and attribute features. Different from the prior direct fusion methods, we next propose to exploit mutual information maximization to filter the modal-specific noise and to augment modal-invariant commonality. Then, we combine pseudo-label calibration with momentum-based contrastive learning to make full use of the labeled and unlabeled data, which improves the quality of pseudo-label and pulls aligned entities closer. Finally, extensive experiments on two MMEA datasets demonstrate the effectiveness of our PCMEA, which yields state-of-the-art performance.

Chunyan Zheng, Keke Sun, Wenhao Zhao et al.

Large pretrained language models (LLMs) have shown surprising In-Context Learning (ICL) ability. An important application in deploying large language models is to augment LLMs with a private database for some specific task. The main problem with this promising commercial use is that LLMs have been shown to memorize their training data and their prompt data are vulnerable to membership inference attacks (MIA) and prompt leaking attacks. In order to deal with this problem, we treat LLMs as untrusted in privacy and propose a locally differentially private framework of in-context learning(LDP-ICL) in the settings where labels are sensitive. Considering the mechanisms of in-context learning in Transformers by gradient descent, we provide an analysis of the trade-off between privacy and utility in such LDP-ICL for classification. Moreover, we apply LDP-ICL to the discrete distribution estimation problem. In the end, we perform several experiments to demonstrate our analysis results.

Jihao Zhao, Chunlai Zhou, Daixuan Li et al.

The collaborative paradigm of large and small language models (LMs) effectively balances performance and cost, yet its pivotal challenge lies in precisely pinpointing the moment of invocation when hallucinations arise in small LMs. Previous optimization efforts primarily focused on post-processing techniques, which were separate from the reasoning process of LMs, resulting in high computational costs and limited effectiveness. In this paper, we propose a practical invocation evaluation metric called AttenHScore, which calculates the accumulation and propagation of hallucinations during the generation process of small LMs, continuously amplifying potential reasoning errors. By dynamically adjusting the detection threshold, we achieve more accurate real-time invocation of large LMs. Additionally, considering the limited reasoning capacity of small LMs, we leverage uncertainty-aware knowledge reorganization to assist them better capture critical information from different text chunks. Extensive experiments reveal that our AttenHScore outperforms most baselines in enhancing real-time hallucination detection capabilities across multiple QA datasets, especially when addressing complex queries. Moreover, our strategies eliminate the need for additional model training and display flexibility in adapting to various transformer-based LMs.

Qiyu Li, Chunlai Zhou, Biao Qin et al.

In this paper, we propose two new definitions of local differential privacy for belief functions. One is based on Shafer's semantics of randomly coded messages and the other from the perspective of imprecise probabilities. We show that such basic properties as composition and post-processing also hold for our new definitions. Moreover, we provide a hypothesis testing framework for these definitions and study the effect of "don't know" in the trade-off between privacy and utility in discrete distribution estimation.