GENERATIVE MODELING · REPRESENTATION LEARNING · EFFICIENT INFERENCE
I work on generative modeling, representation learning, and efficient inference.
I am a senior research engineer at Huawei Hong Kong Research Center. My recent work spans full-duplex spoken dialogue models and neural audio codecs for music generation. My work includes synthetic data generation, post-training, self-supervised representations, vector quantization, and flow-matching decoders.
I received my Ph.D. in Electrical and Electronic Engineering from the University of Hong Kong, supervised by Prof. Ngai Wong and Prof. Edmund Y. Lam. My earlier research explored efficient neural representations and inference; collaborative work also spans GUI agents and multimodal evaluation.
Explore my research01 / RESEARCH
I explore how models learn useful representations, generate structured outputs, and run efficiently. My recent applications include full-duplex spoken dialogue and neural audio codecs for music generation.
SPOKEN DIALOGUE
Full-duplex dialogue requires more than recognizing speech. I work on synthetic dialogue data and post-training for turn-taking, backchannels, and interruptions, adapting a turn-based model for continuous interaction.
Spoken dialogue · Synthetic data · Post-training
REPRESENTATION LEARNING & GENERATION
My neural audio codec work brings together self-supervised encoders, vector quantization, and flow-matching decoders. I have led a team of four engineers and researchers exploring these components for music generation, with a focus on representation learning and generative modeling.
Neural audio codecs · Flow matching · Representation learning
EFFICIENCY & INFERENCE
From compact neural representations to LLM decoding, I study how model design and inference algorithms affect computational cost. My work spans activation sharing, lookup tables, low-bit quantization, and parallel and speculative decoding.
Neural fields · Compression · Inference acceleration
02 / SELECTED WORK
Selected publications from my work on GUI agents and efficient neural representations, explained through the problem each addresses. Full citations and resources follow below.
EMNLP 2025
KG-RAG uses structured knowledge of app interactions to guide an agent’s decisions through retrieval-augmented generation.
KG-RAGICLR 2024
ASMR explores activation sharing across multi-resolution coordinate networks to make implicit neural representation inference more efficient.
ASMRIJCAI 2024
Hundred-Kilobyte Lookup Tables explores compact lookup-based models for efficient single-image super-resolution.
HKLUTAAAI 2024
SCONE explores spatially collaged Fourier bases for implicit neural representations, learning how to represent signals across space.
SCONE03 / BLOG
· AGENTS & PROBABILISTIC DECISIONS
A milestone example connects our trajectory scorer to Jev’s structured decisions. Compare progress, proximity, expected scores, and confidence—and explore what they could change about BFS.
04 / PUBLICATIONS
Peer-reviewed publications and preprints.
* Equal contribution. My name is shown in bold.
15 publications
IEEE TIP 35, 4571–4586 (2026)
SID Symposium Digest 56(1), 282–285 (2025). Distinguished Student Paper
arXiv:2405.05573 (2024)Preprint
No matching publications.
05 / BACKGROUND
2024 — PRESENT
Senior Research Engineer, Audio Speech Lab
Generative AI research spanning full-duplex spoken dialogue models and neural audio codecs for music generation, alongside work on speech recognition. Previously a research intern at Fermat Lab, working on GUI agents.
2023 — 2024
Research Intern
LLM inference acceleration, parallel and speculative decoding, and efficient diffusion models for edge devices.
2020 — 2024
Ph.D., Electrical and Electronic Engineering
Supervised by Prof. Ngai Wong and Prof. Edmund Y. Lam. Previously B.Eng. in Electronic Engineering, Class of 2020.
I have also worked on mobile and language applications and served as a teaching assistant at HKU. My CV contains the full experience and education history ↗
GET IN TOUCH
For research discussions and professional enquiries:
jasonlcl@connect.hku.hk