革新知能統合研究センター

This event is hold on-line and in-person at the RIKEN AIP open space in Nihombashi from 11:00 – 13:00.

Zoom Link
Join Zoom Meeting
https://riken-jp.zoom.us/j/94304502970?pwd=VTBSZXJrQTZZanJuRjA0OSsxR21Sdz09

Meeting ID: 943 0450 2970
Passcode: et0Z54s0VX

Speaker
Samuel Kaski
Finnish Centre for Artificial Intelligence FCAI, Aalto University, and Centre for AI Fundamentals AI-FUN, University of Manchester

Title
Collaborative Machine Learning for Science

Abstract
I will discuss two intertwined problems:

1. Now that research in most empirical fields of science has become computational, in the sense that experiments are designed with simulations or insights derived from existing data, and even replaced with simulations, it is time to ask could we do research more efficiently with new tools, or even do research in new ways. The same questions arise on development work in industry. If we call the constellation of simulation tools a virtual laboratory, an important question is can we have better tools by combining strength across the different fields and developing tools usable across the different virtual laboratories. Many machine learning tools have this aim – they need to have domain-specific elements, such as the specific models which are different in, say, materials science and psychology, but tasks such as experimental design given the models are general-purpose operations.
2. Given common interfaces for the tools in the virtual laboratories in different fields, we can ask could the researchers be helped even more than the current tools are able to. I will discuss machine learning based ‘sidekick’ assistants, able to help other agents research their goals, even when they are not able to yet specify the goal explicitly, or it is evolving. Such assistants can help with tasks ranging from prior knowledge elicitation in modelling, at the simplest, to zero-shot assistance in design and decision making tasks, for instance in drug design. Ultimately they should be helpful for human domain experts in running experiments and solving research problems in simulation-based virtual laboratories. The assistants will be useful tools of domain experts who run such virtual laboratories, and serve as platforms for machine learning researchers to contribute to advancing research across a number of fields, each field running their virtual laboratories which combine field-specific models and domain-agnostic modelling and assistance tools.

Bio
Bio: Samuel Kaski is a professor of Computer Science at Aalto University
and professor of AI in The University of Manchester. He leads the
Finnish Center for Artificial Intelligence FCAI, ELLIS Unit Helsinki
and the ELISE EU Network of AI Excellence Centres. He received the
Turing AI World-Leading Researcher Fellowship in 2021. His field is
probabilistic machine learning, with applications in new kinds of
collaborative AI-assistants able to work well with humans in
modelling, design and decision tasks. Application domains include
computational biology and medicine, brain signal analysis, information
retrieval and user modelling. Prof. Kaski is an ELLIS Fellow, UKRI
Turing AI Fellow, and Turing Fellow of the Alan Turing Institute.

Host
Thomas Möllenhoff (Approximate Bayesian Inference Team)