Active Information Acquisition

We propose a general model that sequentially and dynamically acquire useful information to solve a task under the Learning to Search framework. By focusing on most prominent parts of each instance, our method obtains promising results on sentiment analysis and image recognition. The model also learns to give harder instances more attention without explicitly being trained to do so.

• Type: Journal
• Year: 2016
• Where: CORR
• Authors: He He, Paul Mineiro, Nikos Karampatziakis
• Keywords: CORR 2016, Education