Writing Programming Code with Deep Learning
Neural networks have recently shown major progress with text understanding and text generation, partially because small errors in results are still passable to human eye. On the other hand, generation of highly structured output, such as programs, still poses a challenge. Small perturbations of the program code lead to drastically different results. NEAR is focused on teaching machines to write programs from natural language descriptions. Illia will present how NEAR directly addresses challenges of code generation, by using novel architecture combined with methods from more classical program synthesis. Additionally, he will talk about product discovery process around this new technology.
Illia Polosukhin is co-founder and CTO of NEAR.AI, company that develops a technology to autonomously build software. Prior to NEAR, he was an Engineering Manager at Google Research, leading a team of Deep Learning Researchers working on Natural Language Understanding projects. Also he was a major contributor to TensorFlow.