ALICE: Active Learning with Contrastive Natural Language Explanations

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 4380–4391, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 4380 ALICE: Active Learning with Contrastive Natural Language Explanations Weixin Liang Stanford University wxliang@stanford.edu James Zou1 Stanford University jamesz@stanford.edu Zhou Yu1 University of California, Davis joyu@ucdavis.edu Abstract Training a supervised neural network classi- fier typically requires many annotated training samples. Collecting and annotating a large number of data points are costly and some- times even infeasible. Traditional annotation process uses a low-bandwidth human-machine communication interface: classification labels, each of which only provides a few bits of information. We propose with Contrastive ExplanationsActive Learning (ALICE), an expert-in-the-loop training framework that uti- lizes contrastive natural language explanations to improve data efficiency in learning. AL- ICE learns to first use active learning to se- lect the most informative pairs of label classes to elicit contrastive natural language explana- tions from experts. Then it extracts knowl- edge from these explanations using a seman- tic parser. Finally, it incorporates the extracted knowledge through dynamically changing the learning model’s structure. We applied ALICE in two visual recognition tasks, bird species classification and social relationship classifica- tion. We found by incorporating contrastive explanations, our models outperform baseline models that are trained with 40-100% more training data. We found that adding 1 expla- nation leads to similar performance gain as adding 13-30 labeled training data points. 1 Introduction The de-facto supervised neural network training paradigm requires a large dataset with annotations. It is time-consuming, difficult and sometimes even infeasible to collect a large number of data-points due to task nature. A typical example task is med- ical diagnosis. In addition, annotating datasets also is costly, especially in domains where ex- perts are difficult to recruit. In a traditional an- 1Co-supervised project. Ring Billed Gull A B C Which image is not a Ring-Billed Gull? Figure 1: An example task that would benefit from learning with natural language explanation. The top-left corner shows an example image of a ring-billed gull. In the other three images (A), (B), (C), which one is not a ring-billed gull but a California gull? Given the natural language explanation “Ring-billed gull has a bill with a black ring near the tip while California gull has a red spot near the tip of lower mandible”, it would be easier to find that (A) is the correct choice. notation process, the human-machine communi- cation bandwidth is narrow. Each label provides log C bits per sample for a C-class classification problem. However, humans don’t solely rely on such low bandwidth communication to learn. They instead learn through natural language communi- cation, which grounds on abstract concepts and knowledge. Psychologists and philosophers have long posited natural language explanations as cen- tral, organizing elements to human learning and reasoning (Chin-Parker and Cantelon, 2017; Lom- brozo, 2006; Smith, 2003). Following this intu- ition, we explore methods to incorporate natural language explanations in learning paradigms to im- prove learning algorithm’s data efficiency. Let’s take a bird species classification task …

Content truncated to 500 words