BARC talk by Francesco Silvestri

Friday, 29 April 2022, Francesco Silvestri, Associate Professor at the University of Padova, Italy, will give a talk on "Algorithm design for emerging architectures".

Title:
Algorithm design for emerging architectures

Abstract:
To respond to the intense computational workloads of big data analytics and machine learning, novel hardware accelerators have been designed to improve the performance of key operations. In particular, two paradigms have emerged: 1) tensor architectures which include multiply-and-accumulate hardware units for accelerating tensor algebra; 2) processing-in-memory architectures which allow computational operations directly in memory. These new architectures pose a new algorithmic challenge: how can we exploit their computational power and efficiency to speed up a broad class of algorithms? In this talk, Francesco will present a computational model for tensor architectures and then propose some algorithms for accelerating some key primitives, like matrix multiplication, FFT, Johnson-Lindenstrauss dimensionality reduction, and similarity join. He will conclude with some open questions related to processing-in-memory architectures.

Francecso SilvestriBio:
Francesco is an associate professor in Computer Engineering at the Department of Information Engineering of the University of Padova. Previously, he was an assistant professor at the Department of Information Engineering of the University of Padova (2016-2019), post-doc in the Algorithm group of the IT University of Copenhagen (2015-2016), and post-doc at the Department of Information Engineering of the University of Padova (2009-2014). He has also held positions as part-time lecturer at the IT University of Copenhagen (2013-2014) and as visiting scholar at the Department of Computer Science, University of Texas at Austin (2006-2007). He received his Ph.D. in Computer Engineering from the University of Padova in 2009.
His research targets algorithms and data structures, with an emphasis on:

  • Big data: how can we efficiently extract information from huge amount of data?
  • High performance: how can we exploit modern hardware to crunch big data?