Some results on topic reconstruction and machine learned advice
23 gennaio 2020 alle 14:30
Sala Consiglio 8 Piano, via Celoria 18
Speaker: Alessandro Panconesi, Computer Science, Sapienza Università di Roma
Persona di riferimento: Nicolò Cesa-Bianchi
Iwill talk about two results in machine learning. The first concerns the classical problem of topic reconstruction. Where, roughly speaking, given a corpus of documents, one is given the task of automatically recover the topics the documents talk about. Latent Dirichlet Allocation (LDA) is a well-known paradigm for this problem. We show that LDA topic reconstruction is equivalent to the seemingly much easier task of reconstruction in the Single Topic Allocation model. The second is a result in the recent machine-learned-advice paradigm. Here, one assumes to have a powerful (machine learned) oracle. The goal is to leverage such an oracle to come up with an online algorithm with provable performance guarantees. We present results in this vein for the online facility location problem.
Joint work with Matteo Almanza, Flavio Chierichetti, Silvio Lattanzi, Giuseppe Re, Andrea Vattani
Alessandro Panconesi is full professor of computer science at Sapienza University of Rome. His main research interest is the study of algorithms, especially randomised. For his research he has received two Google Focused Awards, and faculty awards from IBM and Yahoo!. He is the co-recipient of the 2019 Edsger W. Dijkstra Prize in Distributed Computing and of the 1992 ACM Danny Lewin Award. Alessandro has a PhD in computer science from Cornell University and is a staunch supporter of AS Roma.