Deep episodic value iteration : a theory of sample efficient learning for machine learning and cognitive (neuro) science
Abstract/Contents
- Abstract
- This thesis aims to show that, with some extensions, deep learning is computationally capable of addressing rapid adaptation, can account for behavioral phenomena surrounding this ability, and is parsimonious with existing theories regarding the functional role of the relevant neuroanatomy. The main contribution of this thesis is an algorithm that can account for planning from limited data, which we call Deep Episodic Value Iteration (DEVI). DEVI uses a deep neural network to learn a similarity metric, which is applied recursively for planning over episodic memories. By training this network end-to-end based on planning performance, we posit that DEVI should be capable of meta reinforcement learning, even in high dimensional state spaces. We evaluate DEVI in comparison with traditional deep learning techniques as well as other approaches to meta learning. We close by showing how DEVI can be seen as a computational account of schema formation and schema-consistent learning.
Description
Type of resource | text |
---|---|
Form | electronic; electronic resource; remote |
Extent | 1 online resource. |
Publication date | 2018 |
Issuance | monographic |
Language | English |
Creators/Contributors
Associated with | Hansen, Steven Stenberg |
---|---|
Associated with | Stanford University, Department of Psychology. |
Primary advisor | McClelland, James L |
Thesis advisor | McClelland, James L |
Thesis advisor | Goodman, Noah |
Thesis advisor | Poldrack, Russell A |
Advisor | Goodman, Noah |
Advisor | Poldrack, Russell A |
Subjects
Genre | Theses |
---|
Bibliographic information
Statement of responsibility | Steven Stenberg Hansen. |
---|---|
Note | Submitted to the Department of Psychology. |
Thesis | Thesis (Ph.D.)--Stanford University, 2018. |
Location | electronic resource |
Access conditions
- Copyright
- © 2018 by Steven Stenberg Hansen
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...