06-04, 15:30–15:55 (Asia/Jerusalem), Hall 2 (PyData)
In this talk I will describe an end-to-end solution to a text classification problem using publicly available frameworks. I will focus on the practicalities of getting a Deep Learning-based text classification model up and running.
2018 has been declared by many as the "ImageNet moment" of NLP. Novel attention and transformer-based Neural Network (NN) architectures significantly improved state-of-the-art performance in many tasks. NLP-oriented transfer learning techniques claim to make text classification easy by adapting pre-trained models, trained on huge corpora, to proprietary datasets with only a very small number of labels. Models that previously required significant computational power over vast periods of time can now be trained in several hours on standard CPUs.
But with all of these models and frameworks to choose from, how does one make sense of it all? Where to begin?
In this talk I will describe an end-to-end solution to a text classification problem. I will demonstrate how to employ the available classification methods, evaluating their performance and also (arguably more importantly) their ease of use. I will highlight common pitfalls, explaining what it takes to get a Deep Learning-based text classification model up and running.
Inbal completed her military service at one of the technological units of 8200. She holds a B.Sc in physics and electrical engineering from the Technion and an M.Sc in computer science from the Weizmann Institute of Science, where she focused on statistical machine learning with applications in computer vision. Later she completed a two-year research position at Tokyo University. In the past two years she has been working at Gong.io as a data scientist where she builds models to solve a wide range of problems in NLP/NLU.