You can also find my articles on Google Scholar
Published in (Submitted), 2021
In this work, we propose the use of graph neural networks to centrally control AMoD systems. We argue that graph neural networks exhibit a number of desirable properties, and propose an actor-critic formulation as a general approach to learn proactive, scalable, and transferable rebalancing policies.
Published in (Submitted), 2020
In this work, we show how (i) exploiting correlations between multiple signals (i.e. multi-output) and (ii) allow for input-dependent noise (i.e. heteroscedasticity) can fundamentally improve censored data estimation.
Published in (Submitted), 2020
Spatio-Temporal density estimation is one the open challenges in machine learning and statistical sciences. In this work we introduce Recurrent Flow Netwroks, a novel generative neural architecture to model complex data distributions. We show how these models can effectively model and predict future urban mobility distributions.
Published in Transportation Research Part C: Emerging Technologies, 2020
How can we estimate the true (i.e. latent) demand of mobility from incomplete historical observations? This paper approaches this problem by introducing a novel Gaussian process architecture to overcome the censored nature of the demand process.