2:00pm: Gianluca Bontempi (ULB) - "Machine learning for multivariate time series: from forecasting to causal inference"
Conventional approaches in times series literature are restricted to low-dimension series, linear methods and short horizons. Big data revolution is instead shifting the focus to problems (e.g. issued from the IoT technology) characterized by very large dimension, nonlinearity and long forecasting horizon. The presentation will discuss a number of settings where machine learning approaches may be used to deal with time series forecasting and causal understanding. The first part will focus on machine learning strategies for one-step-ahead and multi-step-ahead forecasting both in univariate and multivariate tasks. In particular we will discuss MIMO strategies for multi-step-ahead forecasting of univariate time series and DFML, a machine learning version of the Dynamic Factor Model (DFM), a successful forecasting methodology well-known in econometrics. The DFML strategy is based on a out-of-sample selection of the nonlinear forecaster, the number of latent components and the multi-step-ahead strategy. While accurate forecasting may be obtained by learning associative dependencies between different time instants and time series, an open challenge is how to discriminate between associative dependencies and effective causal relationships. This is particularly challenging in large-variate temporal settings (e.g. spatio-temporal time series) where the multivariate nature of interactions induces a significant correlation between most of the variables. The second part of the presentation will discuss how supervised classification techniques may be used to identify causal dependencies in time series once a proper set of context-dependent descriptors are introduced. The approach, called D2C (Dependency to Causality) performs three steps to predict the existence of a directed causal link between two variables in a multivariate setting: (i) it estimates the Markov Blankets of the two variables of interest and ranks its components in terms of their causal nature, (ii) it computes a number of asymmetric descriptors and (iii) it learns a classifier (e.g. a Random Forest) returning the probability of a causal link given the descriptors value.
3:00pm: Coffee break
3:15pm: Tom Lenaerts (ULB) - "Cooperative AI, game theoretical research into socially beneficial AI"
The advent of complex high-quality autonomous AI systems raises a number of questions on how these systems should decide and act when released into the wild without (or with minimal) human supervision. To avoid disasters, either exogenous (regulations) or endogenous (design) solutions are being proposed. Essentially both solutions want to ensure that the use of AI in business and society is able to align both individual and social preferences and norms, without causing a negative disruption.
In this seminar, I will show that evolutionary game theory, a theoretical framework for studying multi-agent interactions and learning, and related behavioural experiments can help achieve this ambition. Central to this ambition is our ongoing work on studying mechanisms that may influence a collective of agents to prefer cooperation in the context of competitive situations, where individual and collective preferences are not be aligned. I will start this seminar with a short introduction to evolutionary game theory and then zoom in on some cases where we will use both simulations and human experiments to understand how cooperation or coordination can be improved.
What sets EGT apart from other learning paradigms to explore this cooperative AI question is that it immediately combines both individual and societal effects in one framework. It is this duality that needs to be incorporated into the study of cooperative AI systems in order to achieve intelligent systems that are aligned with both an individual’s preferences and norms as well as those of society as a whole.
4:20pm: Jérémie Fays (Technology Transfer Office of ULiège) - "Business model open-source"