European Conference of Artificial Intelligence (ECAI) 2024

Xinpeng Hong (University of Oxford, xinpeng.hong@eng.ox.ac.uk)

I had the privilege of attending the European Conference of Artificial Intelligence (ECAI) 2024, which took place from October 19th to 24th in the historic city of Santiago de Compostela, Spain. This conference provided an invaluable opportunity to engage with leading researchers in the field of AI, exchange ideas, and gain insights into the latest advancements. It was also an honour to present my work, titled “In-Network Machine Learning for Real-Time Transaction Fraud Detection”, during the conference. The discussions and feedback I received were highly beneficial, contributing to the refinement of my research. My participation in ECAI was made possible thanks to the student grant from AISB. It not only facilitated my attendance but also allowed me to make meaningful connections with fellow researchers and explore potential collaborations. The experience has been incredibly enriching, both academically and professionally, and I look forward to continuing to build on the relationships and knowledge gained during the conference.

Presented Research

The research I presented enables real-time transaction fraud detection using in-network machine learning. Machine learning plays a crucial role in combating transaction fraud and real-time fraud detection with ultra-low latency is also essential for swift prevention, but balancing both remains a challenge for financial institutions. We introduced a solution that performs machine-learning-based fraud detection within off-the-shelf programmable network devices such as Intel Tofino switch and NVIDIA BlueField-2 DPU. Tested with three public transaction datasets, our solution achieves a throughput of 4.8 billion transactions per second with microsecond-scale latency. Compared to server-based solutions, it processes 800x more transactions per second and reduces latency by 1300x, while maintaining nearly the same classification performance with minimal loss. This leads to significant savings in computing equipment usage, reducing costs and energy consumption, while also enhancing security for customers.

paper

In-Network Machine Learning for Real-Time Transaction Fraud Detection

poster

ECAI Conference Overview

ECAI is Europe’s leading AI research event, bringing together top researchers, students, and industry professionals. It offers a diverse program, including technical papers, workshops, tutorials, invited talks, special sessions, and the PAIS conference focused on AI applications. ECAI 2024 has over 1,500 registered participants, with 547 accepted submissions and an acceptance rate of 23%. Attending this conference not only provided me with an in-depth understanding of the latest advancements in AI research but also allowed me to engage with experts from various domains, expanding my network and fostering valuable discussions. The exposure to groundbreaking technologies and innovative methodologies has significantly broadened my perspective and enriched my approach to research in the field of AI.

Inspirational Sessions and Research

The conference featured a range of insightful sessions, such as four keynote talks covering key topics: large language models by Prof. Iryna Gurevych, symbolic AI by Prof. Marijn Heule, robotics by Prof. Iolanda Leite, and the application of machine learning in economic development by Prof. César A. Hidalgo. These talks provided a broad perspective on how AI is tackling real-world challenges. One of the standout events was the 50th Anniversary Session on the History of AI in Europe, where I gained valuable insights into the evolution of AI. I also found the "Frontiers in AI" series to be highly engaging, as it showcased exciting new research and techniques across various fields like AI for declarative processes and proportional representation for AI. The short talks, integrated with technical sessions, allowed me to delve deeper into areas that sparked my interest. In addition, the demos and outreach activities also provided practical insights into AI applications, while panels on AI regulation, economic impact, and future directions sparked discussions on the broader implications of AI. In particular, the discussion on AI regulation highlighted the importance of the European strategy in addressing the supervision of high-risk AI systems. The panel also examined the implications of the recently adopted AI Act and the EU’s proposed supervision model, comparing them with global regulatory frameworks.

Networking and Social Opportunities

The conference offered numerous networking and social opportunities. It allowed me to engage directly with my peers and established professionals in a variety of informal settings. Notable social events included the Welcome Reception at the Auditorio de Galicia, the Opening Ceremony at the Galicia Conference and Exhibition Centre, and the "Lunch with a EurAI Fellow". During the latter, small groups of PhD students had the opportunity to engage with a EurAI Fellow to discuss their scientific careers. I had the privilege of speaking with Prof. Roman Barták, who shared valuable insights and offered helpful guidance for advancing our careers. Another highlight was the outdoor social event, where participants walked the final section of Santiago’s French Way, covering the stretch from Monte do Gozo Hill to Obradoiro Square. This walk encouraged conversations while allowing participants to enjoy the historic atmosphere. Also, The Whova app also facilitated online networking before the conference began, and various post-conference meetups brought together individuals with shared research interests or industry ties. I am thankful to AISB for their financial support, which made my participation at ECAI possible. As a way of giving back to the community, I also served as a volunteer during the conference.

About the Author

Xinpeng Hong is a final year Ph.D. student in Engineering Science at the University of Oxford with a research focus on time-sensitive applications of in-network machine learning. He is an alumnus of the Heidelberg Laureate Forum, and his research will be featured in the 2024 edition of the Prototypes for Humanity. Before attending Oxford, he received a bachelor’s degree in software engineering from Tongji University in Shanghai.