Hong Kong Machine Learning Meetup

You need to register in the below link to attend the meetup
Register here

NOTE: This meetup is exclusive and the final list of attendees is at the discretion of the host. You will receive an invitation via email before the meetup date.

We are delighted to announce that our next meetup will be in cooperation with Bullish.
Special thanks to our sponsor, Bullish, for making this event happen.

----------------------------------------------------------------------
Speaker 1: Sébastien Borget
Topic: Unchaining Creators by moving the Creators Economy onchain
Bio: He is an entrepreneur and father with 14 years of experience in growing startups, including 4 years in blockchain and 11 years in mobile gaming. He is the Co-founder and COO of The Sandbox, a pioneering virtual world where players build, own, and monetize gaming experiences using NFTs and SAND, the platform’s utility token. In 2020, he became President of the Blockchain Game Alliance, representing over 250 industry leaders. Recognized among CoinTelegraph’s Top 100 most influential people in crypto, Sébastien holds a Computer Science Engineering degree from Telecom SudParis, one of France’s leading ICT graduate schools.
Summary: He will share how his journey in user-generated content and gaming led him to co-found one of the largest decentralized virtual worlds, partnering with more than 400 major brands. He’ll highlight why creators across all platforms still struggle to get discovered, build communities, and grow their economies; despite the industry’s progress. These challenges inspired him to launch SANDchain, designed as the financial backbone of creativity, empowering creators to introduce loyalty points and access on-chain financialization in a transparent, interoperable way.
----------------------------------------------------------------------
Speaker 2: Olaf Torne
Title: Accelerating equity exotics RFQ response times using ML
Bio: He leads the APAC front-office equity derivatives quant team at Barclays, with over 15 years of experience in quantitative finance spanning flow and exotic derivatives and quantitative investment strategies (QIS). Holding a PhD in mathematics from Brussels University and a graduate certificate in artificial intelligence from Stanford University, he combines classical modeling with machine learning to tackle complex financial challenges. His career includes roles at Merrill Lynch and research collaboration at Ecole Centrale Paris, where he published academic papers and built advanced models. As a team leader, he values collaboration, integrity, and innovation, mentoring others to drive excellence and impact.
Summary: We present several use cases arising in equity structured product issuance, where time-consuming Monte Carlo or Finite Difference based approaches can potentially be enhanced using Machine Learning models.

----------------------------------------------------------------------

Speaker 3: MA Jianfei
Title: Enhancing LLMs' Phonological Reasoning in Chinese via a Fine-tuning and Reinforcement Learning Framework
Bio: He holds a BA in Translation and Linguistics from The Hong Kong Polytechnic University （PolyU） and is currently a first-year Ph.D. student there. His research interests lie in computational linguistics. Under his supervisor's guidance, he is currently working on speech disorders in autism, with future plans to explore AI for therapy. He is currently including one first-author EMNLP (oral) and one ACL (oral, second author), alongside three workshop papers. His work primarily focuses on analyzing linguistic phenomena within model mechanisms （interpretability） and leveraging these phenomena to enhance model performance.
Summary: This talk addresses a key challenge in Chinese NLP: the language's complex phonological system leads to a vast number of homophones. Netizens often exploit this for wordplay or to circumvent content filters, which impacts tasks like high-quality web scraping and sensitive content detection. While Large Language Models (LLMs) possess some innate phonological knowledge, we argue that explicitly teaching them to align phonetic and textual information is crucial. To this end, we constructed a comprehensive Chinese homophone dataset, encompassing typing errors, ASR misrecognitions, and real-world online puns. We then developed a framework combining Supervised Fine-tuning (SFT) to mimic human-like reasoning and Reinforcement Learning (RL) to stabilize the acquired skills. Our results show that even smaller models can achieve phonological reasoning abilities comparable to their larger teacher models.
----------------------------------------------------------------------
Time: Thursday, December 18th, 2025, 18:30 PM to 9:00 PM HKT (Registration start at 6:45pm)
Venue: The Centrium, Building
Address: Bullish limited, 31/F, The Centrium building, 60 Wyndham St, Central