On-policy self-distillation (OPSD) enables a language model to learn from a Teacher instantiated from the same model but given access to privileged information, such as a reference solution. We find that access to sufficient privileged information allows the Teacher to generate highly accurate trajectories; however, ac...
Lin-Fei Ji, Jia-Yu Li, Hao-Zhao Wang et al.· 0 citations
Large-scale multipurpose cohort studies and biobanks often omit covariates needed for specific downstream analyses. We study target-population inference for generalized linear regression when key covariates are completely absent from the target data but observed in a related source population. Standard missing covariat...
Hua-Li Zhao, Kerstin Bunte Bernoulli Institute of Mathematics, Statistics et al.· 0 citations
Teleoperation, a human-in-the-loop control scheme, allows a human operator to remotely command and guide a mobile robot to navigate in off-road environments, yet fluent and user-friendly tele-navigation requires an alignment of traversability evaluation between human and robot. In the teleoperation system, the human op...
Le-Wei Feng, Qi Chen, Wen-Shuo Wang et al.· 0 citations
Agent harness bugs exhibit unique characteristics and remain challenging for state-of-the-art software agents to repair. Progress in this area is further hindered by existing benchmarks, which contain only a small and fixed number of executable harness bugs while requiring hundreds of human hours to construct. This wor...
Realtime-Venus, a proactive full-duplex interaction system with two separately trained 9B models: Realtime-Venus-Omni for audio-visual interaction and Realtime-Venus-Audio for spoken interaction, which leads the compared models on MMAU, Llama Questions, and Speech CMMLU.
The results show that explicitly selecting and integrating evidence improves long-context reasoning while enabling more compact generation, and the strongest overall result among the evaluated open-source models is obtained.
This work identifies a new implicit attack surface for skill selection: even when the user prompt and skill description appear benign in isolation, their semantic relationship can still be strategically shaped to favor an attacker-chosen skill.
Qi-Kai Wang, Yong-Zhao Zhang, Zhi-Wei Chen et al.· 1 citation
This paper presents a tutorial-style, collective-centric taxonomy for collective communication, which organizes recent advances into three layers: communication planning, which generates topology-aware collective schedules; communication execution and adaptation, which maps these schedules onto GPU runtimes and hardwar...
Xue-Bin Song, Menghao Zhang, Yue Liu et al.· IEEE Network· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
