Balancing Quality and Efficiency in Future AI Systems

Add to your list(s) Download to your calendar using vCal

• Shiqiang Wang - IBM
• Friday 29 November 2024, 11:00-12:00
• Computer Lab, LT1.

If you have a question about this talk, please contact Sally Matthews.

Quality and efficiency are both essential in AI systems as data sources become more diverse and model sizes grow. In this talk, I will present techniques to address challenges in data and model quality as well as their efficiency, which are essential for building high-performing and sustainable AI systems. I will first introduce a toolkit for enhancing the quality of datasets, which can be used in a broad range of learning tasks including the training or fine tuning large language models (LLMs), laying the groundwork for model training with good data. Then, considering the specific challenge where data is distributed unevenly across sources with varying sizes, quality, and availability, such as in the case of federated learning, I will introduce the FedAU algorithm. This algorithm dynamically adjusts aggregation weights in the model training process based on the availability of data sources, to prevent model bias and improve training convergence. Afterwards, I will introduce techniques to make both training and inference more efficient, focusing on a framework that optimizes model selection from a zoo of LLMs to minimize energy usage while maintaining model performance guarantees. Together, these approaches form a blueprint for future AI systems that are capable of learning effectively from a vast amount of data at diverse sources and delivering high quality models while enhancing resource efficiency in real-world applications.

Bio: Shiqiang Wang is a Staff Research Scientist at IBM T . J. Watson Research Center, NY, USA . He received his Ph.D. from Imperial College London, United Kingdom, in 2015. His research focuses on the intersection of distributed computing, machine learning, networking, and optimization, currently emphasizing on quality and efficiency aspects related to distributed data and models, which has a broad range of applications including distributed data analytics, efficient model training and inference, edge-based artificial intelligence (Edge AI), and large language models (LLMs). He received the IEEE Communications Society (ComSoc) Leonard G. Abraham Prize in 2021, IEEE ComSoc Best Young Professional Award in Industry in 2021, IBM Outstanding Technical Achievement Awards (OTAA) in 2019, 2021, 2022, and 2023, multiple Invention Achievement Awards from IBM since 2016. For more details, please visit his homepage at https://shiqiang.wang/

This talk is part of the Cambridge ML Systems Seminar Series series.

This talk is included in these lists:

• All Talks (aka the CURE list)
• Cambridge talks
• Computer Lab, LT1
• Department of Computer Science and Technology talks and seminars
• Interested Talks
• School of Technology
• Trust & Technology Initiative - interesting events
• bld31

Note that ex-directory lists are not shown.