Cornell University, USA
Quantization and Compression in ML Systems
Low-precision arithmetic has been popular in machine learning for both training and inference for almost a decade. This quantization helps reduce the time, memory, and energy needed to learn big models on big data. This talk will cover some of the history of quantization, and compression more generally, in machine learning, discussing how it interacts differently with training and inference. Lastly, I will present some recent results from my lab that use a principled theory of quantization error to do post-training quantization of large language models down to 2 bits per weight.
Christopher Matthew De Sa is currently an Assistant Professor in the Computer Science department at Cornell University. research interests include algorithmic, software, and hardware techniques for high-performance machine learning, with a focus on relaxed-consistency variants of stochastic algorithms such as asynchronous and low-precision stochastic gradient descent (SGD) and Markov chain Monte Carlo. He has authored numerous papers in leading journals and conferences in Machine Learning. He holds BS and MS degrees, both in Electrical Engineering, from Stanford University, and completed his PhD from the same department at Stanford University. For more information about his work please see https://www.cs.cornell.edu/~cdesa/.
Google Research, Harvard University, USA
Integrating ML+Optimization: Driving Social Impact in public health and conservation
For more than 15 years, my team and I have been focused on AI for social impact, deploying end-to-end systems in areas of public health, conservation and public safety. In this talk, I will highlight the results from our deployments for social impact in public health and conservation, as well as required innovations in integrating machine learning and optimization. First in terms of public health, I will present recent results from our work in India with the world’s two largest mobile health programs for maternal and child care that have served millions of beneficiaries. Additionally, I will highlight results from earlier projects on HIV prevention and others. In terms of conservation, I will highlight efforts for protecting endangered wildlife in national parks around the globe. To address challenges of ML+optimizaton common to all of these applications, we have advanced the state of the art in decision focused learning, restless multi-armed bandits, influence maximization in social networks and green security games. In pushing this research agenda, our ultimate goal is to facilitate local communities and non-profits to directly benefit from advances in AI tools and techniques
Alfred Deakin Professor
Co-Director of Applied AI Institute, Deakin University, Australia.
Sample efficient AI with applications in health care and advanced manufacturing
From Guglielmo Marconi who developed the radio telegraph to the Wright brothers who invented flying machines, curiosity driven experimentation has powered human innovation. Such experimental optimisation remains an integral part of the Scientific Method. This time-honoured method needs a step change to accelerate scientific innovation because this iterative method quickly hits limits.
To speed-up innovation, it is imperative to expand the capability of experimental optimisation and improve its efficiency. This talk will demonstrate how sample efficient AI can be used to deliver this acceleration in experimental design. I will discuss how the methods can be applied widely, focusing on health and advanced manufacturing particularly in settings where data is scare and experimentation is expensive. In healthcare, I show how these methods can accelerate the design of clinical/health trials to efficiently determine the optimal strategy. In advanced manufacturing, I will show how it can be applied broadly from inventing new materials and alloys to accelerating industrial processes.
The second part of the talk will focus on the new machine learning innovations that have been formulated and solved to advance experimental design. These include incorporating experimental design constraints such as process constraints, transferring knowledge from pervious experiments or experimenter “hunches”, and high dimensional Bayesian optimisation so that the number of experimental control variables can be increased.
Svetha Venkatesh is an Alfred Deakin Professor and a Co- Director of Applied Artificial Intelligence Institute (A2I2) at Deakin University. She was elected a Fellow of the International Association of Pattern Recognition in 2004 for contributions to formulation and extraction of semantics in multimedia data, a Fellow of the Australian Academy of Technological Sciences and Engineering in 2006, and a Fellow of the Australian Academy of Science in 2021 for ground-breaking research and contributions that have had clear impact. In 2017, Professor Venkatesh was appointed an Australian Laureate Fellow, the highest individual award the Australian Research Council can bestow.
Professor Venkatesh and her team have tackled a wide range of problems of societal significance, including the critical areas of autism, security and aged care. The outcomes have impacted the community and evolved into publications, patents, tools and spin-off companies.
Indian Statistical Institute, India
Artificial Intelligence Techniques for Making Biological Discoveries: Some Case Studies
Artificial Intelligence techniques are finding increasingly innovative applications in life sciences for making novel discoveries and for gaining deeper insights into various processes of life. In this talk we will first present an overview of molecular biology and bioinformatics. We will then discuss case studies in three areas, namely single cell RNA-seq data analytics, multiobjective optimization in drug design and graph based approaches in drug interaction studies.