CPIL / MBZUAI / McGill University / Mila
Advancing trustworthy, efficient, and agentic AI for the cyber-physical world.
CPIL is an academic research lab studying intelligent systems for the cyber-physical world. Led by Prof. Xue (Steve) Liu, the lab brings together researchers across MBZUAI, McGill University, and Mila to advance AI, learning, reasoning, optimization, and intelligent infrastructure.
Overview
Cyber-Physical Intelligence Lab is led by Prof. Xue (Steve) Liu. The lab develops methods and systems for trustworthy, efficient, and agentic intelligence in environments where computation, data, infrastructure, and physical processes interact.
Research spans large models, efficient inference, retrieval-augmented reasoning, representation learning, reinforcement learning, federated learning, optimization, trustworthy evaluation, and AI for infrastructure.
CPIL connects foundational AI research with applications in cyber-physical systems, digital twins, wireless networks, data centers, energy systems, and autonomous control.
Research Areas
Theme 01
Methods and systems for making large language models more efficient, scalable, and deployable, including quantization, pruning recovery, early-exit inference, heterogeneous inference, and KV cache compression.
Theme 02
Benchmarks, training methods, and runtime frameworks for AI agents that retrieve evidence, use tools, receive feedback, and operate across long-horizon workflows.
Theme 03
Robust and transferable representations, stable model alignment, reinforcement learning from human feedback, and data synthesis for knowledge distillation.
Theme 04
Diffusion language models and conditional generative modeling for offline black-box optimization, design problems, and settings where labels or experiments are expensive.
Theme 05
Efficient and trustworthy learning methods for distributed, resource-constrained, and heterogeneous environments, including edge devices and non-IID settings.
Theme 06
AI for intelligent infrastructure, data centers, wireless networks, digital twins, energy systems, autonomous control, and real-world optimization constraints.
Projects
A CPIL research theme bringing diffusion/generative models and offline model-based optimization (MBO) to large language models, including an LLM-oriented Design-Bench 2.0.
This theme studies the intersection of diffusion and flow-based generative models, offline black-box / model-based optimization (MBO), and LLM pretraining. A current focus is Design-Bench 2.0, an LLM-oriented benchmark that adapts offline MBO algorithms to LLM-related tasks, alongside diffusion- and flow-based methods for black-box and multi-objective optimization.
Lead: Ye Yuan
Members: Can Chen, Dheeraj, Linfeng Du, Zipeng Sun, Weixu Zhang, Xiuyuan Hu, Yonghan (Harry) Yang, Haolun Wu
A CPIL-led collaboration to integrate AI-field papers into Evidence through interactive, executable, and MyST-compatible dynamic paper experiences.
This project explores how Evidence can support interactive AI papers by combining MyST-native publishing, PreTeXt-style computational layout, containerized demos, and LLM-assisted code and dataset discovery.
Lead: Dun Yuan
Members: Yongjun (Richard) Du, Changjiang Han
A CPIL-led red-team/blue-team effort to systematically attack and defend OpenClaw-like autonomous agents, with a focus on their skills, tools, and high-privilege execution surfaces.
This project studies the security of self-hosted, skill-extensible autonomous agents in the style of OpenClaw. It pairs offensive research (prompt injection, skill poisoning, sandbox escape, and exfiltration) with defensive research (sandboxing, permission models, guardrails, and runtime monitoring), supported by a reproducible agent environment and a curated attack-defense dataset.
Lead: Bowei He
Members: Tao Ni (KAUST), Yankai Chen, Kuiyi Gao (UGRIP Intern), Sachhyam Shrestha (UGRIP Intern)
Publications
CPIL
Venue to be updated, 2026
A sample CPIL publication entry showing the fields maintainers and contributors should fill in when adding a real publication.
News
CPIL discusses OpenClaw as a window into the future of agentic AI, highlighting runtime systems, gateways, memory, skills, safety, and task trajectories as core research challenges.
CPIL has 6 papers accepted to ICML 2026, including 1 Spotlight paper. The works cover diffusion models, black-box optimization, retrieval-augmented reasoning, efficient inference, robust representation learning, and federated learning.
CPIL has 9 papers accepted to ICLR 2026, covering large model optimization, efficient inference, knowledge distillation, trustworthy evaluation, reinforcement learning, representation learning, and alignment.
Join CPIL
CPIL welcomes PhD Students, Postdoctoral Researchers, Visiting Scholars, Research Interns, Undergraduate and Master’s Research Students, Academic Collaborators, Industry Collaborators interested in trustworthy, efficient, and agentic AI systems for the cyber-physical world.