Liran Ma

Jingyi Huang, Ruohan Zong, Yujun Feng et al.

Reinforcement Learning from Human Feedback (RLHF) is critical for aligning Large Language Models (LLMs) with human preferences. However, its efficacy is often compromised by the inherent inconsistency and subjectivity of human annotations. Existing preference optimization frameworks, such as Direct Preference Optimization (DPO), typically treat ambiguous pairs with high annotator disagreement identically to those with unanimous consensus, forcing models to overfit to inconsistent supervision signals and leading to suboptimal alignment. In this work, we propose Reliability-Guided Preference Optimization (RGPO), a robust framework designed to mitigate the impact of inconsistent human feedback. RGPO estimates annotator reliability and infers latent ground truth labels from noisy human feedback to identify robust preferences. Furthermore, we introduce a reliability-aware consistency optimization that dynamically modulates the training objective based on the consensus level of annotations, ensuring the model prioritizes high-consensus supervision signals. Extensive experiments on LLM alignment benchmarks demonstrate that RGPO effectively reduces inconsistency and noise in training data and achieves superior performance compared to widely adopted RLHF baselines. Our code and configurations are available at https://github.com/GenieHuang/RGPO.

Qinghe Gao, Yan Huo, Tao Jing et al.

The Internet-of-Things (IoT) emerges as a paradigm to achieve ubiquitous connectivity via wireless communications between kinds of physical objects. Due to the wireless broadcasting nature and the energy constraint of physical objects, concerns on IoT security have triggered research on cooperative jamming based physical layer security. With the help of a cooperative jammer, existing solutions can effectively fight against eavesdroppers. However, these schemes are of high energy cost due to continuously transmitting jamming signals. To reduce the energy consumption, we propose a new idea of intermittent jamming and design five specific intermittent jamming schemes (IJSs). By taking the transmit frame formate into account, we optimize these IJSs from three aspects, including the jamming power, the jamming method, and the jamming positions. Then we analyze the applicability of the proposed IJSs according to different requirements on the synchronization, the available jamming energy and the jamming power constraints. Extensive MATLAB experiments are conducted on the basis of the WLAN Toolbox, which demonstrate the proposed IJSs can effectively degrade the reception of the eavesdropper and outperform the widespread continuous jamming scheme (CJS) when the available jamming energy is limited.