Invitation to RIAIoT Lunch Seminar Series on 13 March: “Revolutionizing Proteomics:AI-Assisted De Novo Peptide Sequencing” & “PolyLink: Efficient Geo-Distributed LLM Inference via Collaborative Edge Computing”

We are excited to announce the RIAIoT Lunch Seminar Series, featuring two engaging talks from our members: “Revolutionizing Proteomics:AI-Assisted De Novo Peptide Sequencing” by Prof. Shujun Wang (1st speaker) and “PolyLink: Efficient Geo-Distributed LLM Inference via Collaborative Edge Computing” by Dr Mingjin Zhang(2nd speaker). This event will be held on 13 March, 2026, at The Hong Kong Polytechnic University.

Abstracts

• Revolutionizing Proteomics: AI-Assisted De Novo Peptide Sequencing

Abstract: Peptide identification is a central task in bottom-up proteomics, as it converts raw tandem mass spectrometry data into biologically meaningful insights that underpin biomarker discovery, drug development, mechanistic biology, and personalized medicine. Traditional approaches, including database search and spectral library matching, are highly effective for known peptides but remain fundamentally constrained by reference availability, limiting their ability to uncover novel biology. De novo peptide sequencing offers a promising alternative by inferring peptide sequences directly from mass spectra without relying on predefined databases, yet its performance is hindered by two major challenges: missing fragmentation, where informative ions are absent, and noisy fragmentation, where irrelevant signals interfere with true sequence inference. In this talk, I will discuss how artificial intelligence can address these challenges and enable more accurate and robust peptide identification. I will also highlight our recent work on AI-assisted de novo sequencing, focusing on two key computational problems and the corresponding solutions we propose, and frame this effort within the broader shift from conventional experiment-driven proteomics toward AI-assisted scientific discovery.

• PolyLink: Efficient Geo-Distributed LLM Inference via Collaborative Edge Computing

Abstract: Large Language Models (LLMs) have achieved significant success in content generation and intelligent decision-making. Traditionally, LLMs are primarily deployed in the cloud, leading to high latency, bandwidth costs, and potential privacy risks. In recent years, edge computing has been seen as an effective way to address these issues, deploying LLMs on edge devices closer to the data source. However, LLMs are inherently computationally intensive and resource-intensive, and a single edge device often struggles to handle their operational demands.

To address this, we propose designing and developing a novel Collaborative Edge Computing (CEC) system to connect these geographically distributed edge devices with cloud servers, building a unified joint resource pool to support low-cost, high-efficiency, and cross-regional distributed LLM inference. This approach offers several advantages: low-cost inference, improved overall resource utilization, and reduced reliance on centralized cloud.

In this report, I will introduce PolyLink, a geo-distributed LLM inference platform built on the CEC concept. This platform aims to connect idle GPU servers from multiple stakeholders globally, providing low-cost, high-efficiency LLM inference services. PolyLink is publicly available and has successfully integrated with over 100  geo-distributed GPU servers from several universities. I will focus on two main research questions addressed by PolyLink: 1) an edge-native LLM inference framework supporting resource-adaptive model partitioning and automated deployment; and 2) an efficient KV cache migration mechanism to handle dynamic scenarios such as node overload and planned exits. By integrating these innovations into the PolyLink system and validating them through actual deployment, we are committed to building a scalable and efficient infrastructure for next-generation LLM services.

Details

• • Title:
  1. Revolutionizing Proteomics:AI-Assisted De Novo Peptide Sequencing (1^st Speaker: Prof. Shujun Wang)
  2. PolyLink: Efficient Geo-Distributed LLM Inference via Collaborative Edge Computing (2^nd Speaker: Dr Mingjin Zhang)
  • Date: 13 March 2026 (Friday)
  • Time: 12:00 n.n. – 1:45 p.m.
  • Venue: CD620, 6/F, Hong Kong Chinese Manufacturers' Association Building (Block C), The Hong Kong Polytechnic University (Capacity: 30, first-registered, first-served)
  • Lunch Arrangement: A light lunch set (Food & drink from the PolyU gourmet shop) will be provided if your registered the seminar before 4pm on 12 March. (Please note that food orders placed after this time may not be counted, but you are still welcome to join the seminar)
  • Register Methods: click https://polyu.hk/LRZXb or scan the QR code on the poster above