Distributed AI training
I analyse how data, tensor and pipeline parallel training move information between computing nodes, including communication in large language models.
AI SYSTEMS · OPTICAL INTERCONNECTS · COMPUTING EDUCATION
Efficient systems and interconnects for large-scale AI
Senior Lecturer in Computing · School of Computing, Eastern Institute of Technology (EIT), New Zealand
Dr Fei (Travis) Dai is a Senior Lecturer in Computing at the Eastern Institute of Technology. He received his PhD in Computer Science from the University of Otago.
His research focuses on communication-efficient distributed AI training: collective communication, optical and electronic interconnects, and the co-design of algorithms and computing systems. He also studies performance and energy use in large-scale AI workloads, including distributed large language model training.
Alongside this, he is developing a research programme in GenAI-enabled computing education, covering authentic and AI-aware assessment, reflective learning, and transnational computing education between China and New Zealand. He welcomes research collaboration and postgraduate supervision enquiries.

CURRENT PROJECT · 2026
Communication Analysis of LLM Training for Optical Interconnect Co-design
EIT internal research fundRESEARCH
My research focuses on the algorithms and networks that move data during distributed training, and on how students learn to build and evaluate such systems.
Communication-efficient and interconnect-aware AI systems
I analyse how data, tensor and pipeline parallel training move information between computing nodes, including communication in large language models.
I design and evaluate all-reduce, all-gather and related operations used to coordinate large-scale training.
I study how communication algorithms and optical or electronic networks can be designed together to reduce training time and energy use.
GenAI-enabled computing education and authentic learning
A complementary programme on how computing students move from theory to practice, and how generative AI changes teaching and assessment.
PUBLICATIONS
Full list below. See Google Scholar for citations.
From Survey to Lived Experience: Explaining Task-Level Theory-to-Practice Barriers in a Sino–New Zealand Joint Undergraduate Computing Programme
ACM Transactions on Computing Education (TOCE) Conditionally accepted
State of the Art in Parallel and Distributed Systems: Emerging Trends and Challenges
Electronics, 14(4), 677
A Preliminary Investigation into Theory-Practice Barriers in Sino-New Zealand Undergraduate Computing Education
17th International Conference on Computer Supported Education (CSEDU 2025)
Efficient Algorithm for All-Gather Operation in Optical Interconnect Systems
IEEE Open Journal of the Communications Society (OJ-COMS), vol. 5, pp. 664–676
Wrht: Efficient All-Reduce for Distributed DNN Training in Optical Interconnect Systems
52nd International Conference on Parallel Processing (ICPP 2023), pp. 556–565 Oral
OpTree: An Efficient Algorithm for All-gather Operation in Optical Interconnect Systems
IEEE International Conference on Communications (ICC 2023), pp. 428–434
Efficient All-Reduce for Distributed DNN Training in Optical Interconnect Systems
28th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming (PPoPP 2023), pp. 422–424 Poster
Performance Comparison of Distributed DNN Training on Optical Versus Electrical Interconnect Systems
23rd International Conference on Algorithms and Architectures for Parallel Processing (ICA3PP 2023), LNCS 14487, pp. 401–418
Comparing the Performance of Multi-Layer Perceptron Training on Electrical and Optical Network-on-Chips
The Journal of Supercomputing, 79(10), pp. 10725–10746
Accelerating Deep Neural Network Training on Optical Interconnect Systems
PhD thesis, University of Otago
Performance Comparison of Multi-layer Perceptron Training on Electrical and Optical Network-on-Chips
22nd International Conference on Parallel and Distributed Computing, Applications and Technologies (PDCAT 2021), LNCS 13148, pp. 129–141
Open research artefacts
Reproducibility artefact for layer-wise hybrid parallelism in distributed DNN training on optical interconnects: transport scheduler, collective candidates (WRHT, OpTree and others), memory-aware configuration search and an independent schedule verifier.
Simulator and reproduction package for collective-aware core allocation in fully connected neural network training on a ring optical network-on-chip. Every table, figure and quoted number is regenerated from the repository.
NEWS
Paper on task-level theory-to-practice barriers in a Sino–New Zealand joint computing programme conditionally accepted in ACM Transactions on Computing Education (TOCE).
Lead Guest Editor of the Electronics Special Issue on AI-enabled intelligent embedded architectures. Submissions open until 6 February 2027.
Released public reproducibility artefacts for LHAS and optical network-on-chip core allocation.
Promoted to Senior Lecturer in Computing at the Eastern Institute of Technology.
Paper on theory–practice barriers in Sino–New Zealand undergraduate computing education published at CSEDU 2025, where I also served as Session Chair.
Review State of the Art in Parallel and Distributed Systems: Emerging Trends and Challenges published in Electronics.
All-gather paper for optical interconnect systems published in IEEE OJ-COMS.
Wrht all-reduce paper accepted to ICPP 2023 (oral presentation).
OpTree all-gather paper accepted to IEEE ICC 2023.
Distributed DNN training performance comparison accepted to ICA3PP 2023.
Poster on all-reduce for optical interconnect systems accepted to PPoPP 2023; Workshop and Session Chair at PMAM 2023.
Multi-layer perceptron training on electrical and optical network-on-chips published in The Journal of Supercomputing.
This Special Issue brings together practical research on embedded AI, edge intelligence and real-time systems. We welcome original research and review articles that connect AI methods with efficient system design.
Guest Editors: Dr Fei (Travis) Dai (EIT, New Zealand), Dr Shuangyao Huang (Xi'an Jiaotong-Liverpool University, China), Dr Fanfan Shen (Nanjing Audit University, China)
TEACHING
My teaching connects computing concepts with the systems students design, build and evaluate.
Current areas
I use hands-on tasks and step-by-step guidance to help students move from core ideas to working systems.
My assessment design is authentic and AI-aware: students use appropriate tools while showing that they understand their decisions and learning.
ACADEMIC SERVICE
Editorial
Conferences
Journal reviewing
GLOBAL REACH
Loading visitor data…