|
Building machines that learn and think like people| old_uid | 14261 |
|---|
| title | Building machines that learn and think like people |
|---|
| start_date | 2017/07/25 |
|---|
| schedule | 16h |
|---|
| online | no |
|---|
| details | External Seminar - Probabilistic models of language and cognition |
|---|
| summary | Machine learning systems, and reinforcement learning models
in neuroscience and AI, typically focus on problems of learning from
large data sets and weak prior knowledge. In contrast, much human
learning takes place very rapidly, from just one or a few relevant
examples or experiences, in the context of strong priors. What is
the form of the prior knowledge that makes such fast learning
possible? And how is that knowledge itself acquired? I will talk
about our recent work addressing these questions in a few areas drawn
from concept learning, learning linguistic rules, learning new skill
domains (such as video games), and learning to use tools and draw
pictures. Techniques from probabilistic programming, Bayesian program
synthesis and program induction, in some cases integrated with
insights from recent deep learning or classic reinforcement learning
approaches, provide a powerful toolkit for building cognitive models
as well as more human-like machine learning systems. |
|---|
| responsibles | <not specified> |
|---|
| |
|