The New Frontiers in Learning, Control, and Dynamical Systems is a new workshop currently under submission to ICLR 2023. The tentative goals of the workshop include:
- Discuss the interaction between control theory and deep learning. In particular, how can deep learning be used to solve stochastic control problems (Forward-Backward SDE, PDE, Mean-Field Games) and how can stochastic control methods be applied to deep learning (reinforcement learning, generative modeling, Schrodinger bridge, DNN optimization).
- Explore the scalability and interpretability of deep methods inspired from control theory and dynamical systems. Which approaches scale and which do not? Can control-theoretic algorithms meet industrial empirical standards? How is interpretability defined and how useful it can be?
- Present applications of deep (stochastic) control methods in robotics, medical imaging, protein modeling, scientific computing, and more generally to mean-field problems and multi-agent systems in economics and population modeling.
We invite researcher in ML, control and dynamical systems to submit their latest works to the workshop. Proper topics include but not limited to (see Call for paper for more details):
- Optimal transport in ML applications
- Stochastic analysis and stochastic optimal control
- Dynamical probabilistic inference e.g. MCMC, Variational Inference
- Neural ODE/SDE/PDE, e.g. diffusion Model