Five labs run by recent NeurIPS / ICML / ICLR / CVPR / ACL authors posted open PhD or postdoc positions in the past week. Each entry below follows the same structure: position type, intake, research focus, how to apply, and one honest note on lab culture or fit.

Research focus. Wang's lab sits at the intersection of online learning, bandits, and sequential decision-making under uncertainty — with two specific problem clusters: (1) entanglement routing in noisy quantum networks, where agents must verify network health from limited measurements; (2) communication-efficient learning in LLM agent networks, where groups of agents share knowledge while minimizing inter-agent messaging overhead. 1

Target background is solid probability theory, optimization, or theoretical CS. Strong collaborator network includes UMass Amherst, CUHK, CityUHK, NJU, and SJTU.

Lab culture note. Wang is building this lab fresh at HKBU, with an established publication track at NeurIPS / ICML / ICLR / AAAI / SIGMETRICS. Early PhD students will have high visibility and close mentorship; the flip side is that lab infrastructure is still forming. Short-term RA slots are genuinely available for undergraduates looking to build a research record before applying to PhD programs.

Research focus. Chen's work targets efficient and controllable large-scale models. Current threads: diffusion large language models (dLLMs) and masked diffusion; multi-objective deep learning with Pareto-optimal model merging; and LLM compression via low-rank adaptation and structured pruning. Recent highlights include an ICML 2025 paper on preference-aware model merging and two NeurIPS 2025 papers on masked diffusion sampling and multi-objective one-shot pruning. 2 The AI-for-Science angle is explicit in the group's positioning.

Lab culture note. CityUHK's HKIASI is a newer interdisciplinary institute, which means cross-disciplinary collaborations are actively encouraged. Chen publishes steadily at ICML / NeurIPS / ICLR; students targeting top-venue output from day one will find a direct fit.

Research focus. Li's Multimedia Lab concentrates on embodied AI, robotic manipulation, and multimodal foundation models. In 2025–2026 alone the group placed 13 papers at NeurIPS, 27 across ICML / CVPR / ICLR / AAAI, and 3 at ACL — one of the highest output rates among individual PIs in the region. Core threads span vision-language models, reinforcement learning for image generation, mobile GUI agents, and autonomous driving. 3

Lab culture note. Li was promoted to Full Professor effective August 2026, and consistently appears on Stanford's World's Top 2% Scientists list. This is a high-throughput group — students who want both volume and prestige will find it productive, but expect a competitive, fast-paced pace. Lab size means individual mentorship may be distributed across senior PhD students and postdocs.

Research focus. Xu's group works on natural language processing and large language models — specifically multilingual and cross-cultural LLMs, RL post-training for LLMs, long-context multi-turn agent evaluation, and cross-disciplinary AI applications spanning education, privacy, legal, and medical domains. A recent ACL 2026 paper on geolocation reasoning chains and an ACL 2024 Best Social Impact Award paper on cultural bias in LLMs are representative outputs. 4

Lab culture note. Xu co-organizes lab social activities with collaborator Alan Ritter's group (kayaking, dinners, group retreats visible on her homepage), which signals a deliberate investment in cohort culture. Atlanta is a mid-cost-of-living US city; the GRA stipend goes further here than at coastal institutions.

Research focus. Udell's group sits at the crossroads of large-scale optimization, machine learning, and data science: stochastic numerical linear algebra, adaptive step sizes, second-order optimization, low-rank matrix completion, safe and explainable AI, Gaussian process inference, and physics-informed neural networks. Application areas include PDE solvers, healthcare risk prediction, and energy systems. A NeurIPS 2025 paper on sketch-and-project Gaussian process inference is a recent flagship output. 5

Lab culture note. Udell explicitly looks for postdocs with "ideas with the potential for big impact in optimization and AI." This is a theory-leaning group — strong candidates will have a dissertation in optimization, numerical linear algebra, or applied probability. Stanford's MS&E department has unusually direct bridges to the business school, which suits postdocs planning future careers at the intersection of ML and operations research.

Openings here were identified by direct review of faculty homepages updated in the past 7 days. All five positions show active recruiting signals (updated homepage text, explicit recruitment language, or rolling contact). Two fields — Deadline and Stipend — often show "Not disclosed" or "Rolling" because CS PhD recruitment in Asia and Europe does not follow the uniform November–December cycle of North American programs; contact labs directly for the most current admissions timelines.

For NeurIPS / ICML / ICLR applicants, the November graduate admission deadline means the practical window to reach out is now through September — well before most portals open.