[22] Graham Neubig - Unsupervised Learning of Lexical Information
The Thesis Review
00:00
The Pros and Cons of Non-Parametric Bayesian Statistics in Neural Models
There's recently been a move to non-parametric models in neural models for language modeling or question answering as well. For example, there are models that basically process all of Wikipedia. And then you build a question answering model where you have a latent variable, which is one of the sentences in Wikipedia. So essentially what you have there is this is also a non- parametric model. It's a model where all of the data that you process at training time becomes part of your model.
Transcript
Play full episode
