Christos Boutsidis
CurrentlyI lead the Graph Analytics team within the Surveillance Analytics Group at Goldman Sachs. We are building the ‘‘Goldman Sachs Social Graph”. Our work involves combining multiple sources of structured (e.g., trades, transactions) and unstructured (e.g., text, voice) data into a single, highly heterogeneous, social network. In addition, our group develops scalable solutions (Hadoop, MapReduce, HBase) for several graph mining problems such as vertex centrality (e.g., pagerank), vertex similarity, (shortest) paths, edge outlier detection, graph sparsification, community detection, and graph embeddings, to name a few. ResearchMy research focuses on solving important problems in numerical linear algebra and machine learning; specifically, I am interested in developing fast approximation algorithms for such problems. Those algorithms are often used in place of the more traditional exact algorithms when one wants to trade the accuracy of the solution with the running time. Additionally, I am exploring solutions to those problems in different models of computation such as distributed, streaming, and online. The main tool that I am using to design new algorithms is ‘‘sketching”, which offers powerful randomized and/or sampling techniques for matrices. Besides the design and the theoretical analysis of efficient approximation algorithms, I am interested to understand what are the limits of each model (lower bounds) and how the algorithms perform in practice on large matrices. In particular, I have worked extensively in the following Linear Algebraic / Machine Learning problems (more details): Experiences
