Jing Zhu

Yuhang Zhou, Jing Zhu, Shengyi Qian et al.

Large Language Models (LLMs) are increasingly aligned with human preferences through Reinforcement Learning from Human Feedback (RLHF). Among RLHF methods, Group Relative Policy Optimization (GRPO) has gained attention for its simplicity and strong performance, notably eliminating the need for a learned value function. However, GRPO implicitly assumes a balanced domain distribution and uniform semantic alignment across groups, assumptions that rarely hold in real-world datasets. When applied to multi-domain, imbalanced data, GRPO disproportionately optimizes for dominant domains, neglecting underrepresented ones and resulting in poor generalization and fairness. We propose Domain-Informed Self-Consistency Policy Optimization (DISCO), a principled extension to GRPO that addresses inter-group imbalance with two key innovations. Domain-aware reward scaling counteracts frequency bias by reweighting optimization based on domain prevalence. Difficulty-aware reward scaling leverages prompt-level self-consistency to identify and prioritize uncertain prompts that offer greater learning value. Together, these strategies promote more equitable and effective policy learning across domains. Extensive experiments across multiple LLMs and skewed training distributions show that DISCO improves generalization, outperforms existing GRPO variants by 5% on Qwen3 models, and sets new state-of-the-art results on multi-domain alignment benchmarks. Our code and data are available at https://github.com/Tonyzhou98/disco_grpo.

Sihan Gao, Jing Zhu, Xiaoxuan Zhuang et al.

In recent years, human pose estimation has made significant progress through the implementation of deep learning techniques. However, these techniques still face limitations when confronted with challenging scenarios, including occlusion, diverse appearances, variations in illumination, and overlap. To cope with such drawbacks, we present the Spatial Attention-based Distribution Integration Network (SADI-NET) to improve the accuracy of localization in such situations. Our network consists of three efficient models: the receptive fortified module (RFM), spatial fusion module (SFM), and distribution learning module (DLM). Building upon the classic HourglassNet architecture, we replace the basic block with our proposed RFM. The RFM incorporates a dilated residual block and attention mechanism to expand receptive fields while enhancing sensitivity to spatial information. In addition, the SFM incorporates multi-scale characteristics by employing both global and local attention mechanisms. Furthermore, the DLM, inspired by residual log-likelihood estimation (RLE), reconfigures a predicted heatmap using a trainable distribution weight. For the purpose of determining the efficacy of our model, we conducted extensive experiments on the MPII and LSP benchmarks. Particularly, our model obtained a remarkable $92.10\%$ percent accuracy on the MPII test dataset, demonstrating significant improvements over existing models and establishing state-of-the-art performance.

Tara Safavi, Jing Zhu, Danai Koutra

Codifying commonsense knowledge in machines is a longstanding goal of artificial intelligence. Recently, much progress toward this goal has been made with automatic knowledge base (KB) construction techniques. However, such techniques focus primarily on the acquisition of positive (true) KB statements, even though negative (false) statements are often also important for discriminative reasoning over commonsense KBs. As a first step toward the latter, this paper proposes NegatER, a framework that ranks potential negatives in commonsense KBs using a contextual language model (LM). Importantly, as most KBs do not contain negatives, NegatER relies only on the positive knowledge in the LM and does not require ground-truth negative examples. Experiments demonstrate that, compared to multiple contrastive data augmentation approaches, NegatER yields negatives that are more grammatical, coherent, and informative -- leading to statistically significant accuracy improvements in a challenging KB completion task and confirming that the positive knowledge in LMs can be "re-purposed" to generate negative knowledge.

Jing Zhu, Yanan Xu, Yanmin Zhu

The aim of session-based recommendation is to predict the users' next clicked item, which is a challenging task due to the inherent uncertainty in user behaviors and anonymous implicit feedback information. A powerful session-based recommender can typically explore the users' evolving interests (i.e., a combination of his/her long-term and short-term interests). Recent advances in attention mechanisms have led to state-of-the-art methods for solving this task. However, there are two main drawbacks. First, most of the attention-based methods only simply utilize the last clicked item to represent the user's short-term interest ignoring the temporal information and behavior context, which may fail to capture the recent preference of users comprehensively. Second, current studies typically think long-term and short-term interests as equally important, but the importance of them should be user-specific. Therefore, we propose a novel Parallel Attention Network model (PAN) for Session-based Recommendation. Specifically, we propose a novel time-aware attention mechanism to learn user's short-term interest by taking into account the contextual information and temporal signals simultaneously. Besides, we introduce a gated fusion method that adaptively integrates the user's long-term and short-term preferences to generate the hybrid interest representation. Experiments on the three real-world datasets show that PAN achieves obvious improvements than the state-of-the-art methods.

Yunxiao Shi, Jing Zhu, Yi Fang et al.

Learning to predict scene depth and camera motion from RGB inputs only is a challenging task. Most existing learning based methods deal with this task in a supervised manner which require ground-truth data that is expensive to acquire. More recent approaches explore the possibility of estimating scene depth and camera pose in a self-supervised learning framework. Despite encouraging results are shown, current methods either learn from monocular videos for depth and pose and typically do so without enforcing multi-view geometry constraints between scene structure and camera motion, or require stereo sequences as input where the ground-truth between-frame motion parameters need to be known. In this paper we propose to jointly optimize the scene depth and camera motion via incorporating differentiable Bundle Adjustment (BA) layer by minimizing the feature-metric error, and then form the photometric consistency loss with view synthesis as the final supervisory signal. The proposed approach only needs unlabeled monocular videos as input, and extensive experiments on the KITTI and Cityscapes dataset show that our method achieves state-of-the-art results in self-supervised approaches using monocular videos as input, and even gains advantage to the line of methods that learns from calibrated stereo sequences (i.e. with pose supervision).

Jing Zhu, Yunxiao Shi, Mengwei Ren et al.

Monocular depth estimation is a challenging task that aims to predict a corresponding depth map from a given single RGB image. Recent deep learning models have been proposed to predict the depth from the image by learning the alignment of deep features between the RGB image and the depth domains. In this paper, we present a novel approach, named Structure-Attentioned Memory Network, to more effectively transfer domain features for monocular depth estimation by taking into account the common structure regularities (e.g., repetitive structure patterns, planar surfaces, symmetries) in domain adaptation. To this end, we introduce a new Structure-Oriented Memory (SOM) module to learn and memorize the structure-specific information between RGB image domain and the depth domain. More specifically, in the SOM module, we develop a Memorable Bank of Filters (MBF) unit to learn a set of filters that memorize the structure-aware image-depth residual pattern, and also an Attention Guided Controller (AGC) unit to control the filter selection in the MBF given image features queries. Given the query image feature, the trained SOM module is able to adaptively select the best customized filters for cross-domain feature transferring with an optimal structural disparity between image and depth. In summary, we focus on addressing this structure-specific domain adaption challenge by proposing a novel end-to-end multi-scale memorable network for monocular depth estimation. The experiments show that our proposed model demonstrates the superior performance compared to the existing supervised monocular depth estimation approaches on the challenging KITTI and NYU Depth V2 benchmarks.

Jing Zhu, Yi Fang, Husam Abu-Haimed et al.

Environment perception, including object detection and distance estimation, is one of the most crucial tasks for autonomous driving. Many attentions have been paid on the object detection task, but distance estimation only arouse few interests in the computer vision community. Observing that the traditional inverse perspective mapping algorithm performs poorly for objects far away from the camera or on the curved road, in this paper, we address the challenging distance estimation problem by developing the first end-to-end learning-based model to directly predict distances for given objects in the images. Besides the introduction of a learning-based base model, we further design an enhanced model with a keypoint regressor, where a projection loss is defined to enforce a better distance estimation, especially for objects close to the camera. To facilitate the research on this task, we construct the extented KITTI and nuScenes (mini) object detection datasets with a distance for each object. Our experiments demonstrate that our proposed methods outperform alternative approaches (e.g., the traditional IPM, SVR) on object-specific distance estimation, particularly for the challenging cases that objects are on a curved road. Moreover, the performance margin implies the effectiveness of our enhanced method.