Timezone: »

Information Bottleneck for Multi-Task LSTMs
Bradley Baker · Noah Lewis · Debbrata Kumar Saha · Md Abdur Rahaman · Sergey Plis · Vince Calhoun
Event URL: https://openreview.net/forum?id=hnqkGVw_jtA »

Neural networks, which have had a profound effect on how researchers , do so through a complex, nonlinear mathematical structure which can be difficult to interpret or understand. This is especially true for recurrent models, as their dynamic structure can be difficult to measure and analyze. However, interpretability is a key factor in understanding certain problems such as text and language analysis. In this paper, we present a novel introspection method for LSTMs trained to solve complex language problems, such as sentiment analysis. Inspired by Information Bottleneck theory, our method uses a state-of-the-art information theoretic framework to visualize shared information around labels, features, and between layers. We apply our approach on simulated data, and real sentiment analysis datasets, providing novel, information-theoretic insights into internal model dynamics.

Author Information

Bradley Baker (Georgia Institute of Technology)
Noah Lewis
Debbrata Kumar Saha (Georgia Institute of Technology)
Md Abdur Rahaman (Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS), Georgia Institute of Technology, Georgia State University, Emory University)
Sergey Plis (TReNDS center, GSU)
Vince Calhoun (Georgia State University)

More from the Same Authors