Human motion prediction (HMP) involves forecasting future human motion based on historical data. Graph Convolutional Networks (GCNs) have garnered widespread attention in this field for their proficiency in capturing relationships among joints in human motion. However, existing GCN-based methods tend to focus on either temporal-domain or spatial-domain features, or they combine spatio-temporal features without fully leveraging the complementarity and cross-dependency of these two features. In this paper, we propose the Spatial-Temporal Multi-Subgraph Graph Convolutional Network (STMS-GCN) to capture complex spatio-temporal dependencies in human motion. Specifically, we decouple the modeling of temporal and spatial dependencies, enabling cross-domain knowledge transfer at multiple scales through a spatio-temporal information consistency constraint mechanism. Besides, we utilize multiple subgraphs to extract richer motion information and enhance the learning associations of diverse subgraphs through a homogeneous information constraint mechanism. Extensive experiments on the standard HMP benchmarks demonstrate the superiority of our method.

Exploring the bridge between historical and future motion behaviors remains a central challenge in human motion prediction. While most existing methods incorporate a reconstruction task as an auxiliary task into the decoder, thereby improving the modeling of spatio-temporal dependencies, they overlook the potential conflicts between reconstruction and prediction tasks. In this paper, we propose a novel approach: Temporal Decoupling Decoding with Inverse Processing (TD2IP). Our method strategically separates reconstruction and prediction decoding processes, employing distinct decoders to decode the shared motion features into historical or future sequences. Additionally, inverse processing reverses motion information in the temporal dimension and reintroduces it into the model, leveraging the bidirectional temporal correlation of human motion behaviors. By alleviating the conflicts between reconstruction and prediction tasks and enhancing the association of historical and future information, TD2IP fosters a deeper understanding of motion patterns. Extensive experiments demonstrate the adaptability of our method within existing methods.

Reinforcement Learning from Human Feedback (RLHF) has emerged as a standard and effective approach for training large language models (LLMs) with human preferences. In this framework, a learned reward model approximates human preferences and guides policy optimization, making it crucial to develop an accurate reward model. However, without the ``true'' reward function, challenges arise when the reward model is an imperfect proxy for human preference. Since the policy optimization continuously shifts the human preference training dataset's distribution. The fixed reward model suffers from this problem of off-distribution, especially the on policy methods. While collecting new preference data can mitigate this issue, it is costly and challenging to optimize. Thus, reusing the policy interaction samples becomes a possible way to further refine the reward model. To tackle these challenges, we introduce a novel method \textbf{U}ncertainty-\textbf{G}radient based \textbf{D}ata \textbf{A}ugmentation (\textbf{UGDA} for short) to enhance reward modeling by leveraging policy samples to maintain on-distribution performance. Specifically, UGDA selects interaction samples based on the uncertainty of the reward ensembles and the gradient based influence of policy optimization. After the reward relabeling of selected samples, we use supervised learning to refine the reward ensembles, then get the retrained policy. Extensive experiments demonstrate that by leveraging UGDA to select a few samples without the costly human preference data collection, we can improve the ability of the policy and surpass the state-of-the-art methods.

Large language models (LLMs) have demonstrated significant improvements in contextual understanding. However, their ability to attend to truly critical information during long-context reasoning and generation still falls behind the pace. Specifically, our preliminary experiments reveal that certain distracting patterns can misdirect the model’s attention during inference, and removing these patterns substantially improves reasoning accuracy and generation quality. We attribute this phenomenon to spurious correlations in the training data, which obstruct the model’s capacity to infer authentic causal instruction–response relationships. This phenomenon may induce redundant reasoning processes, potentially resulting in significant inference overhead and, more critically, the generation of erroneous or suboptimal responses. To mitigate this, we introduce a two-stage framework called Learning to Focus (LeaF) leveraging intervention-based inference to disentangle confounding factors. In the first stage, LeaF employs gradient-based comparisons with an advanced teacher to automatically identify confounding tokens based on causal relationships in the training corpus. Then, in the second stage, it prunes these tokens during distillation to enact intervention, aligning the student’s attention with the teacher’s focus distribution on truly critical context tokens. Experimental results demonstrate that LeaF not only achieves an absolute improvement in various mathematical reasoning, code generation and multi-hop question answering benchmarks but also effectively suppresses attention to confounding tokens during inference, yielding a more interpretable and reliable reasoning model.

Yiju Guo, Ganqu Cui, Lifan Yuan, Ning Ding, Zexu Sun, Bowen Sun, Huimin Chen, Ruobing Xie, Jie Zhou, Yankai Lin, Zhiyuan Liu, Maosong Sun. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.