Timezone: »

A Lagrangian Duality Approach to Active Learning
Juan Elenter · Navid Naderializadeh · Alejandro Ribeiro

Wed Nov 30 02:00 PM -- 04:00 PM (PST) @ Hall J #509

We consider the pool-based active learning problem, where only a subset of the training data is labeled, and the goal is to query a batch of unlabeled samples to be labeled so as to maximally improve model performance. We formulate the problem using constrained learning, where a set of constraints bounds the performance of the model on labeled samples. Considering a primal-dual approach, we optimize the primal variables, corresponding to the model parameters, as well as the dual variables, corresponding to the constraints. As each dual variable indicates how significantly the perturbation of the respective constraint affects the optimal value of the objective function, we use it as a proxy of the informativeness of the corresponding training sample. Our approach, which we refer to as Active Learning via Lagrangian dualitY, or ALLY, leverages this fact to select a diverse set of unlabeled samples with the highest estimated dual variables as our query set. We demonstrate the benefits of our approach in a variety of classification and regression tasks and discuss its limitations depending on the capacity of the model used and the degree of redundancy in the dataset. We also examine the impact of the distribution shift induced by active sampling and show that ALLY can be used in a generative mode to create novel, maximally-informative samples.

Author Information

Juan Elenter (University of Pennsylvania)
Navid Naderializadeh (University of Pennsylvania)

Navid NaderiAlizadeh received the B.S. degree in electrical engineering from Sharif University of Technology, Tehran, Iran, in 2011, the M.S. degree in electrical and computer engineering from Cornell University, Ithaca, NY, USA, in 2014, and the Ph.D. degree in electrical engineering from the University of Southern California, Los Angeles, CA, USA, in 2016. After spending more than four years as a Research Scientist at Intel Labs and HRL Laboratories, he is currently a Postdoctoral Scholar in the Department of Electrical and Systems Engineering at the University of Pennsylvania. Navid's research interests include various aspects of machine learning, including multi-agent reinforcement learning, graph representation learning, and self-supervised learning, and the applications of information theory and machine learning for resource allocation in wireless communication networks. Navid ranked first in the Iranian Nationwide University entrance exam in 2007. He was a recipient of the Jacobs Scholarship in 2011. He was selected as a 2015-16 Ming Hsieh Institute Ph.D. Scholar. He was also a finalist in the Shannon Centennial Student Competition at Nokia Bell Labs in 2016.

Alejandro Ribeiro (University of Pennsylvania)

More from the Same Authors