This paper investigates an uplink rate-splitting multiple access (RSMA) network empowered by a Rydberg atomic quantum receiver (RAQR) to maximize energy efficiency (EE). To overcome the amplitude-only measurement limitation of the considered direct readout RAQR, we devise a phase alignment mechanism that linearizes the received signal while retaining full channel energy, thereby enabling tractable optimization. The resulting non-convex fractional EE problem is transformed via Dinkelbach's method. For IoT scenarios with symmetric rate requirements, we derive exact closed-form power allocation from the Karush-Kuhn-Tucker (KKT) conditions, which yields a globally optimal bisection search with negligible computational latency. For asymmetric rate requirements, we develop a generalized block coordinate descent-based power allocation (BCD-PA) framework that eliminates reliance on expensive interior-point solvers. We further propose a low-complexity paired allocation (LC-Paired) that exploits user pairing and closed-form bandwidth solutions via the Lambert-W function. Simulation results confirm that the KKT-based scheme attains the highest EE, while the BCD-PA offers a flexible complexity-performance trade-off. The LC-Paired scheme drastically reduces computational cost with only moderate EE degradation.
In this paper, we propose a novel uplink Hybrid Non-Orthogonal Multiple Access (NOMA) architecture assisted by Rydberg Atomic Quantum Receivers (RAQRs) to improve energy efficiency by addressing the physical constraints of atomic sensing. We formulate a joint optimization problem encompassing phase measurements and multi-user power allocation to minimize the total transmit power. This problem is particularly challenging due to the non-linear amplitude-only observations of atomic vapor cells. To decomposes the resulting non-convexity, we develop a hierarchical three-layer decomposition optimization framework that systematically decouples the global problem. The inner layer employs a customized Expectation Maximization-Gerchberg-Saxton (EM-GS) algorithm to resolve phase ambiguity and estimate equivalent channel gains. The middle layer derives closed-form optimal power allocation solutions via Karush-Kuhn-Tucker (KKT) conditions. The outer layer utilizes linear Minimum Mean Square Error (MMSE) detection for high-reliability signal recovery. Simulation results demonstrate the superiority of the proposed framework over conventional RF-based Hybrid NOMA. Notably, the atomic architecture achieves approximately a 93% reduction in total power consumption compared to Orthogonal Multiple Access (OMA) baselines while maintaining strict backward compatibility, thereby laying a solid foundation for practical deployment in next-generation green communication networks.
Tian Xueyun, MingHua Ma, Bingbing Xu Institute of Computing Technology, Chinese Academy of Sciences, Nuoyan Lyu, Wei Li ByteDance Inc., Heng Dong ByteDance Inc., Zheng Chu Harbin Institute of Technology, Yuanzhuo Wang Chinese Academy of Sciences 等 9 位作者
Supervised fine-tuning (SFT) on chain-of-thought (CoT) trajectories demonstrations is a common approach for enabling reasoning in large language models. Standard practices typically only retain trajectories with correct final answers (positives) while ignoring the rest (negatives). We argue that this paradigm discards substantial supervision and exacerbates overfitting, limiting out-of-domain (OOD) generalization. Specifically, we surprisingly find that incorporating negative trajectories into SFT yields substantial OOD generalization gains over positive-only training, as these trajectories often retain valid intermediate reasoning despite incorrect final answers. To understand this effect in depth, we systematically analyze data, training dynamics, and inference behavior, identifying 22 recurring patterns in negative chains that serve a dual role: they moderate loss descent to mitigate overfitting during training and boost policy entropy by 35.67% during inference to facilitate exploration. Motivated by these observations, we further propose Gain-based LOss Weighting (GLOW), an adaptive, sample-aware scheme that exploits such distinctive training dynamics by rescaling per-sample loss based on inter-epoch progress. Empirically, GLOW efficiently leverages unfiltered trajectories, yielding a 5.51% OOD gain over positive-only SFT on Qwen2.5-7B and boosting MMLU from 72.82% to 76.47% as an RL initialization. Code is available at Github.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Runxuan Liu, Xianhao Ou Harbin Institute of Technology, Xinyan Ma Harbin Institute of Technology, Jiyuan Wang, Jiafeng Liang Harbin Institute of Technology, Jiaqi Li Tianjin Normal University, Tao He Harbin Institute of Technology, Zheng Chu Harbin Institute of Technology 等 16 位作者
Long Chain-of-Thought (LCoT), achieved by Reinforcement Learning with Verifiable Rewards (RLVR), has proven effective in enhancing the reasoning capabilities of Large Language Models (LLMs). However, reasoning in current LLMs is primarily generated as plain text, where performing semantic evaluation on such unstructured data creates a computational bottleneck during training. Despite RLVR-based optimization, existing methods still suffer from coarse-grained supervision, reward hacking, high training costs, and poor generalization.To address these issues, we propose the Graph Reasoning Paradigm (GRP), which realizes structured and symbolic reasoning, implemented via graph-structured representations with step-level cognitive labels. Building upon GRP, we further design Process-Aware Stratified Clipping Group Relative Policy Optimization (PASC-GRPO), which leverages structured evaluation to replace semantic evaluation, achieves process-aware verification through graph-structured outcome rewards, and mitigates reward hacking via stratified clipping advantage estimation. Experiments demonstrate significant improvements across mathematical reasoning and code generation tasks. Data, models, and code will be released later.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表