Efficient AI Systems
Fast and fair large language model decoding across heterogeneous edge resources, including distributed speculative decoding.
Efficient LLM · Speculative decoding
Sydney, Australia
Distributed intelligence for real‑world systems.
The Distributed compUting, optimizAtion, and Learning (DUAL) Group connects rigorous theory with intelligent systems that can operate at scale.
Solving…
We study resource-constrained environments where communication, compute, fairness, and reliability matter just as much as model accuracy.
Research themes
Building dependable intelligence, together.
Federated & Edge Learning
Collaborative learning that respects resource, privacy, and communication constraints.
Federated learning · Edge intelligence
Network Systems & Optimization
Scalable algorithms with robust convergence, and reliable meaning-aware communication under uncertainty.
Distributed optimization · Semantic communication
Generative AI
Generative modeling — flow matching, diffusion, and beyond — for dependable real-world use.
Flow matching · Diffusion
News
Signals from the lab.
- 2026-08 paper Our paper FAVoR: Measuring and Mitigating Author-Style Homogenization in Federated Personalized Generation was accepted to the EMNLP 2026 main track.
- 2026-07 paper Our paper NashPG: A Policy Gradient Method with Iteratively Refined Regularization for Finding Nash Equilibria was accepted to TMLR.
- 2026-07 paper We published our FedKO work in IEEE IoT Journal, on federated anomaly detection for multivariate time series.
- 2026-06 paper We published FORCE in IEEE Networking Letters — our orthogonality-aware LoRA for communication-efficient federated fine-tuning.
- 2026-06 paper We presented FOUL, our communication-efficient federated unlearning method, at CVPR 2026.
People & collaboration
Curious minds are welcome.
DUAL brings together researchers and students across machine learning, distributed systems, optimization, and edge intelligence.
