Login
Talks & Seminars
Title: Learning Functions over Sets via Permutation Adversarial Networks
Mr. Chirag Pabbaraju, Microsoft Research India
Date & Time: July 5, 2019 16:00
Venue: Department of Computer Science and Engineering, Room No. 109, 01st Floor, New CSE/CC Building
Abstract:
Inputs to several ML applications are naturally structured as sets. Traditionally though, ML algorithms assume inputs to be vector-valued and are not invariant to permutations of the input vector, thus being ill-suited in general to handle set-valued inputs. In this talk, we consider the problem of learning functions over sets that are invariant to permutations of input set items. Recent approaches in this domain suggest techniques of pooling individual element embeddings of the set in order to obtain a permutation invariant representation of the set. However, such techniques can necessitate extremely large embedding sizes for challenging functions, and also miss out on capturing relations amongst set elements. To address this challenge, we present a novel architecture called Set Permutation Adversarial Networks (SPAN) that simultaneously learns the target function as well as adversarial or worst-case permutations for each input set. We show results and comparisons with state-of-the-art methods on a variety of tasks from the domains of statistics, graph functions and linear algebra. We also present a case study of how learning set functions can help extract powerful features for real-world recommendation systems.
Speaker Profile:
Chirag Pabbaraju is a Research Fellow in the Machine Learning and Optimization Group at Microsoft Research India. Previously, he completed his B.E. in Computer Science from BITS Pilani Goa Campus in 2018. For his Bachelor’s thesis, he worked on developing ML algorithms for IoT devices as part of the EdgeML project at MSR India. He is presently working on the meeting slot recommendation system of Outlook’s Calendar. He will be joining the Machine Learning Department at Carnegie Mellon University as a Masters student in August 2019.
List of Talks

Webmail

Username:
Password:
Faculty CSE IT
Forgot Password
    [+] Sitemap     Feedback