Nexus Lab, the joint research laboratory of INT2 and VNU-ITI, hosts the August session of the AI Lunch & Seminar Series on Friday, 21 August 2026, at 10:00 AM, in Room 505, Building E3, VNU Hanoi (144 Xuan Thuy, Cau Giay, Ha Noi).
The session welcomes Prof. Nguyen Le Minh from the Japan Advanced Institute of Science and Technology (JAIST), who will present Speculative Decoding: Towards Efficient Inference for Large Language Models. The seminar is chaired by Dr. Nguyen Viet Cuong of Nexus Lab, INT2 & VNU-ITI.
Large Language Models (LLMs) have brought remarkable progress, yet their greatest obstacle today is the computational cost of sequential, token-by-token generation. The talk introduces Speculative Decoding, a breakthrough approach that accelerates LLM inference without compromising the quality of the model’s output. At its heart is the draft-and-verify mechanism: instead of decoding one token at a time, a lightweight draft model rapidly proposes candidate token sequences, which the target LLM then verifies in parallel.
The presentation features an in-depth study of UNISPEC, a self-operating speculative decoding framework that requires no retraining and provably preserves the output distribution. UNISPEC stands out with device-aware automatic tuning of the draft length, an advanced n-gram confidence estimation algorithm, and a confidence-driven tree expansion strategy. The talk closes with a discussion of the future of adaptive, hardware-aware inference and multilingual optimisation.
Prof. Nguyen Le Minh is a Professor at JAIST and head of Nguyen Lab, which focuses on Machine Learning and Natural Language Understanding. His research interests span Large Language Models, Natural Language Processing, Explainable AI (XAI), Legal AI, and methods for optimising LLM inference performance. His group has earned the SAC Highlights Award at ACL 2025 for SPECTRA and at ACL 2026 for UNISPEC, opening new directions for high-performance AI deployment.
The AI Lunch & Seminar Series is a lunchtime discussion series hosted by Nexus Lab at Building E3, Xuan Thuy Campus, Vietnam National University, Ha Noi, tracking AI developments in Viet Nam and worldwide. Sessions are deliberately brief and fast-paced, wrapping up before 2:00 PM.
Source: Nguyen Lab, JAIST