Events

Seminar “Some recent work on Convolutional Neural Networks (CNNs) for text classification”

INESC-ID

On June 22, at 2.30 p.m., will take place the seminar “Some recent work on Convolutional Neural Networks (CNNs) for text classification", presented by Byron Wallace.

On June 22, at 2.30 p.m., will take place the seminar “Some recent work on Convolutional Neural Networks (CNNs) for text classification”, presented by Byron Wallace (University of Texas at Austin).

Text classification is a fundamental natural language processing (NLP) task. Modern neural models that exploit (usually pre-trained) word embeddings have recently achieved impressive results on such tasks. Feed-forward Convolutional Neural Networks (CNNs), in particular, have emerged as a relatively simple yet powerful class of models for text classification, often outperforming more complex recurrent neural models such as Long Term Short Term networks (LSTMs). In this talk, I will review CNN architectures appropriate for text and discuss model design and hyper-parameter trade-offs. I will then introduce new variants of CNNs, including an architecture that jointly exploits multiple sets of embeddings and a model that capitalizes on “rationale-level” supervision, i.e., labels on sentences concerning their relevance to the classification task at hand. Finally, I will present recent work on “active learning” approaches for CNNs that aim to rapidly induce discriminative embeddings with as few labels as possible. I will present results with respect to diverse text classification tasks, ranging from verbal irony detection to biomedical text classification.

More information.