t,p: Language Generation with Wang Ling

brief: Many language generation and understanding tasks require the production of arbitrary functions. I will present a language generation approach, where the text is generated by executing a program that is induced automatically from the model. Crucially, this approach allows arbitrary functions to be incorporated in the generation process, granting additional generalization capabilities for the model. I will describe the inference challenges that emerge from marginalizing over all possible programs that can generate the observed text, and present two scenarios where the model can be applied. Firstly, I show that the model can be used to generate code from natural language descriptions. Secondly, I will use the model to solve and explain algebraic math problems.

bio: Wang Ling is a researcher from Deepmind working on applying machine learning methods to natural language processing. He graduated in 2015 from Carnegie Mellon University and Instituto Superior Tecnico. His recent work focuses on building models that generate programs from natural language.

----

Title legend: the "t,p" gives you an idea of the event. Here is what it means:

t: theory

p: practice, applied talk

x: experience

h: hackathon

m: informal meet&greet

r: repeat