PhD Roundup, Week of May 12–18 2026: Openings from NeurIPS, ICML, ICLR, and CVPR Authors

Eight labs publishing at this season's top conferences have active PhD or postdoc openings. This issue covers research directions, what each advisor explicitly asks from applicants, and the few cultural details they've put on the record.

How to read this roundup

Each entry is structured as: position type → institution → research focus → what the advisor asks → how to apply or contact. Lab-culture notes come directly from the advisor's own hiring page — no inference about personality or management style.

Open positions this week

Tianyi Chen — Cornell University (Cornell Tech)

Position: PhD students and postdocs, open intake

Chen's lab moved from RPI to Cornell in fall 2025 and sits inside the Cornell Tech campus in New York City. Research sits at the intersection of large language model alignment and analog AI hardware: bilevel optimization for LLM fine-tuning, multi-objective RLHF, and training neural networks directly on non-ideal physical substrates (resistive memory, photonic circuits) to address energy costs. 1

Recent top-conference placements include NeurIPS 2025 (Oral), ICML 2025, ICLR 2025, and ICML 2026. Notably, the NeurIPS 2025 oral was on analog in-memory training — an area that has almost no overlap with the "LLM alignment" crowd, which means the lab spans an unusually wide methods range.

What the advisor asks: Strong background in optimization, ML systems, or AI hardware. Send a CV to tianyi.chen [AT] cornell.edu.

Huan Zhang — University of Illinois Urbana-Champaign (UIUC)

Position: PhD students and postdocs, open intake

Zhang's lab at UIUC ECE works on trustworthy AI with formal and empirical methods: neural network verification (the α,β-CROWN framework, which won the VNN-COMP competition), certified robustness for LLMs and VLMs, and safe reinforcement learning for embodied agents. 2

The lab has maintained one of the densest NeurIPS/ICML/ICLR track records in the verification-and-robustness niche over the last five years — three NeurIPS 2025 spotlights, ICML 2025 (two papers), and four ICLR 2026 acceptances. That breadth across theory, LLMs, and robotics suggests students with different emphases can find a project.

What the advisor asks: Passion for research plus strong ML/AI technical background; experience in trustworthy ML, formal verification, AI safety, or agentic AI is preferred but not required. PhD applicants apply to UIUC ECE and/or CS, then email the lab. Postdocs and visiting students send CV + brief research statement to huan [AT] huan-zhang.com.

Chong Liu — University at Albany (UAlbany), SUNY

Position: 2 fully funded PhD positions (Fall 2026 — listed as closed, applications were reviewed)

Liu joined UAlbany in 2024 as part of a state-funded 27-faculty AI cluster hire, backed by $75M from New York State and a $20M IBM collaboration announced in 2023. The research combines Bayesian optimization, bandit algorithms, quantum machine learning, and AI for accelerated scientific discovery (including drug discovery). 3

The hiring page is now labeled "closed for Fall 2026," but the position description is worth reading for the next cycle: Liu is an Area Chair at ICML 2025 and ICLR 2026, has published at ICML, ICLR, AISTATS, and runs the AI4D3 (AI for Drug Discovery) workshop series since 2023.

What the advisor asks: First-author publication at one of the listed venues (ICML, NeurIPS, ICLR, CVPR, ACL, and several others) is explicitly required. Email cliu24 AT albany.edu with CV, undergrad transcript, and a representative writing sample. Subject line: UAlbany CS PhD Application by [[YOUR_NAME]].

Seulki Lee — KAIST, School of Electrical Engineering

Position: MS, PhD, postdoc, and intern positions for 2026 and 2027 intake

Lee's Embedded AI Lab (EAI Lab) at KAIST works on running ML models directly on embedded, mobile, and IoT hardware — what the lab calls "Embedded Physical AI." Sub-directions include on-device computer vision, embedded NLP, efficient inference and training, and edge AI agents. 4

The lab won an ICML 2025 Outstanding Position Paper Award (on the peer-review crisis), has an ICLR 2026 paper, and published at NeurIPS 2025 on long-term time series forecasting. A Gold reviewer badge at ICML 2026 and a 2026 Google Research Award ($50K cloud credits) signal active engagement with the community.

What the advisor asks: Interest in systems and AI research. Apply via the lab page at sites.google.com/view/embeddedai.

Shiqiang Wang — University of Exeter (UK)

Position: 2 fully funded PhD positions on LLM-based agentic AI (application deadline was November 30, 2025; check for re-posting)

Wang's CORE-AIx Lab focuses on distributed ML, federated learning, LLM training and routing, and agentic AI systems. 5 The November 2025 cohort was listed via jobs.ac.uk. 6

Wang has papers at NeurIPS in every year from 2020 to 2025 (including NeurIPS 2025 on LLM routing with service-level guarantees), ICML from 2022 to 2026, and ICLR from 2023 to 2026 (two ICLR 2026 papers on efficient fine-tuning). The breadth from federated optimization theory to production LLM deployment is notable.

What the advisor asks: Interested candidates should check the lab site for re-posted openings. Past postings asked for relevant theoretical or systems background.

Yusuke Matsui — University of Tokyo

Position: 1–2 PhD students, intake from April 2027 onward

Matsui's lab at the University of Tokyo (Department of Information and Communication Engineering) researches large-scale AI system infrastructure: vector databases, approximate nearest neighbor (ANN) search, retrieval-augmented generation (RAG), and learned data structures. 7 8

Recent conference output spans CVPR 2026, ICML 2026, and SIGMOD 2026. The lab explicitly values open-source release and has no enforced "core time" — relevant signals if you care about research culture and work style.

What the advisor asks: Applicants are expected to enter via the University of Tokyo graduate school entrance exam (Department of ICE, Graduate School of IST). Email a CV and research proposal, mentioning that you found the position through the website. Prior PhD applicants should have at least one first-author top-tier paper, or multiple second-tier first-author papers.

Hongsheng Li — The Chinese University of Hong Kong (CUHK)

Position: PhD students, intake for 2027

Li's Multimedia Laboratory at CUHK ECE covers embodied AI and robotic manipulation, multimodal large language models, vision generation and editing, 3D perception for autonomous driving, and medical image analysis. 9

The lab co-led foundational work on recent CVPR 2025 papers (including image generation and 3D scene understanding) and is actively publishing in CVPR, NeurIPS, and multimodal reasoning benchmarks. For applicants focused on embodied AI or vision-language models, this lab operates in the Hong Kong academic environment with strong ties to mainland Chinese AI industry.

What the advisor asks: Contact the lab with a CV; check the homepage for current application instructions.

Shaoyi Huang — Stevens Institute of Technology

Position: Fully funded PhD positions (rolling intake) + remote interns

Huang's lab at Stevens works on efficient ML algorithms, hardware-aware algorithm–system co-design for LLMs and diffusion models, privacy-preserving ML, and ML for chip design (EDA). 10

Conference output includes NeurIPS 2023 (LinGCN on homomorphic-encryption inference), ICML MAS 2025, and CVPR 2026. A distinctive angle: the group explicitly works at the intersection of ML algorithms and chip-level hardware, which is uncommon among purely ML-focused labs at teaching-focused research universities.

What the advisor asks: Send CV, transcript, and research interests to [email protected]. Remote interns are also accepted.

Quick reference table

Advisor	Institution	Conference track record	Research focus	Position type	Status
Tianyi Chen	Cornell (Cornell Tech)	NeurIPS, ICML, ICLR	LLM alignment, bilevel optimization, analog AI hardware	PhD + postdoc	Open
Huan Zhang	UIUC	NeurIPS, ICML, ICLR	Trustworthy AI, neural network verification, safe RL	PhD + postdoc	Open
Chong Liu	UAlbany (SUNY)	ICML, ICLR, AISTATS	Bayesian optimization, quantum ML, AI for science	2 funded PhDs	Closed for Fall 2026
Seulki Lee	KAIST	NeurIPS, ICML, ICLR	Embedded / on-device AI, edge ML systems	PhD + postdoc + intern	Open (2026–2027)
Shiqiang Wang	U Exeter	NeurIPS, ICML, ICLR	Federated learning, LLM routing, agentic AI	2 funded PhDs	Check for re-posting
Yusuke Matsui	U Tokyo	CVPR, ICML	Vector DBs, ANN search, RAG infrastructure	1–2 PhDs	Open (April 2027)
Hongsheng Li	CUHK	CVPR, NeurIPS	Embodied AI, VLMs, vision generation	PhDs	Open (2027)
Shaoyi Huang	Stevens Institute	NeurIPS, CVPR, ICML	Efficient ML, hardware co-design, EDA	PhD + remote intern	Open (rolling)

Notes for applicants

A few patterns across this week's listings worth flagging directly:

Publication bar varies widely. Chong Liu explicitly requires a first-author top-venue paper for PhD admission. Tianyi Chen, Huan Zhang, and Seulki Lee ask for strong backgrounds without listing a publication requirement. Know which bar you're applying against before emailing.

System vs. theory vs. application splits. The labs above split roughly into: pure optimization/theory (Chen, Zhang, Liu), systems/hardware (Huang, Wang, Lee), and vision/multimodal applications (Li, Matsui). Most are not pure ML theory groups — applicants whose strength is empirical deep learning will find more openings here than those whose sole skill is proving convergence rates.

Geography matters beyond prestige. UAlbany's $75M AI investment from New York State and proximity to IBM Research and GlobalFoundries is a real differentiator for students interested in semiconductor-adjacent AI research. KAIST sits in Daejeon with full funding in South Korean won; Exeter is a Russell Group university in southwest England. These are not minor footnotes when you're committing 4–6 years.

Next issue publishes Monday, May 25. Send corrections or additions to the channel.