"A Tensor Approach to Efficient Learning of Mixed Membership Community Models in Graphs"

Thu Jul 26, 2018 2:00 PM

Location: LTS Auditorium, 8080 Greenmead Drive

Furong Huang
Assistant Professor, Department of Computer Science and UMIACS

We will introduce a method-of-moments based approach for a latent variable modeling problem, community detection, in which we learn the latent communities that the actors/nodes in networks belong to. A new concept, “graph moments,” will be introduced for graph data, in analogy to the commonly used “data moments” for IID data. Our algorithm decomposes the graph moments to obtain a consistent estimator of the model parameters with guarantees under mild conditions.

The algorithm uses efficient sparse matrix computations and is suitable for large sparse data sets. We conduct optimization of multilinear operations and avoid directly forming the tensors, in order to save computational and storage costs. We demonstrate accuracy and computational efficiency on Facebook, Yelp and DBLP data sets. We compare our results to the state-of-the-art algorithms, such as the variational method, and report a gain of accuracy and a gain of several orders of magnitude in the running time.

Speaker Bio:
Furong Huang is an assistant professor of computer science with an appointment in UMIACS.

Huang’s research focuses on machine learning, high-dimensional statistics and distributed algorithms.

Applications of her research include developing fast detection algorithms to discover hidden and overlapping user communities in social networks, learning convolutional sparse coding models for understanding semantic meanings of sentences and object recognition in images, healthcare analytics by learning a hierarchy on human diseases for guiding doctors to identify potential diseases afflicting patients, and more.

Huang has made significant contributions in non-convex optimization for spectral methods and learning latent variable graphical models on distributed systems with large-scale data.

She was a keynote speaker at the Tensor Workshop at ICCV’17 and organized the Matrix Factorization Workshop at the 2017 Heidelberg Laureate Forum. Huang serves on the program committee for the SysML’18 conference.

She received the 2017 Adobe Research Award.