Model uncertainty and risk in machine learning
- Michael Branicki, University of Edinburgh
- Goncalo Dos Reis, University of Edinburgh
- Blanka Horvath, Imperial College London
- Christa Cuchiero, University of Vienna
This workshop will focus on the interplay between model uncertainty and risk in machine learning and address the main developments towards understanding why deep learning works. The workshop will look at how to analyse learning efficiency of deep learning algorithms, convergence and their robustness under input perturbations using tools from the theory of dynamical systems, mean-field games and ODEs. Identifying the crucial ingredients of a systematic and rigorous framework, and optimization methods for designing deep learning architectures with high approximation capacity and efficient training rates, critically study their performance, and predicting their behaviour under different conditions with provable guarantees on the estimates.
The workshop will engage with industry partners to inform a dialogue towards developing comprehensive risk management frameworks with ML-based algorithms with an outlook for long-term collaborative and impactful research.
Bringing together international and UK experts in the area of random and stochastic dynamical systems, machine learning and deep learning, and mean-field theory, this workshop will foster a dialogue between these groups and provide a forum for collaborations and cross-fertilization.