Bio
Lorenzo Rimella started his studies with a Bachelor's Degree in Mathematics for Finance and Insurance at the University of Torino (UNITO). After earning his bachelor's degree he took a more mathematical path, starting a Master's Degree in Stochastics and Data Science at the University of Torino (UNITO). Along with his Master's Degree he decided to be part of the "honor program" at Collegio Carlo Alberto which enhanced his knowledge in Statistics and Economics applications and gave him a Master's in Statistics and Applied Mathematics. He is now a PhD student in Statistics at the University of Bristol under the supervision of Nick Whiteley and an Enrichment student at The Alan Turing Institute.
He is particularly impressed by how quickly the AI community is improving as well as by how easily any computer can learn a human task. This has encouraged him to believe that automation is often the best way to do things. However, he also believes that machines need the correct guide in order to go in the right direction. This belief motivates him to deepen his knowledge in Data Science and AI. Outside the academic world he likes training in the gym and hanging out with friends. He is also a board-game freak and loves spending the weekend socialising by playing games.
Research interests
Hidden Markov models (HMMs) are statistical models that explain patterns in observed data in terms of the evolution over time of a process which is not observed directly. The main challenge is generally to infer the conditional probabilities of the hidden states given the observations, which are technically called: filtering distribution (if it is the conditional distribution over the data up to the current time step) and smoothing distribution (if it is the conditional distribution over all the data).
HMMs are extremely flexible and they have found multiple applications across the Machine Learning community. Unfortunately their prohibitive computational cost because of their high-dimension has significantly limited their use. Lorenzo tried to overcome this issue by proposing a low-cost approximated filtering and smoothing algorithm, which is also proven to be a "good" approximation. He applied his findings to model the flow of people in the London's underground network and he would like to open new collaborations in the Turing's community in this field. He has recently grown an interest in Bayesian neural network and wishes to find a connection between HMMs and their training techniques.