We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

eess.SY

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Electrical Engineering and Systems Science > Systems and Control

Title: Improved Attention Models for Memory Augmented Neural Network Adaptive Controllers

Abstract: We introduced a {\it working memory} augmented adaptive controller in our recent work. The controller uses attention to read from and write to the working memory. Attention allows the controller to read specific information that is relevant and update its working memory with information based on its relevance. The retrieved information is used to modify the final control input computed by the controller. We showed that this modification speeds up learning. In the above work, we used a soft-attention mechanism for the adaptive controller. Controllers that use soft attention or hard attention mechanisms are limited either because they can forget the information or fail to shift attention when the information they are reading becomes less relevant. We propose an attention mechanism that comprises of (i) a hard attention mechanism and additionally (ii) an attention reallocation mechanism. The attention reallocation enables the controller to reallocate attention to a different location when the relevance of the location it is reading from diminishes. The reallocation also ensures that the information stored in the memory before the shift in attention is retained which can be lost in both soft and hard attention mechanisms. We illustrate through detailed simulations of various scenarios for two link robot and three link robot arm systems we illustrate the effectiveness of the proposed attention mechanism.
Subjects: Systems and Control (eess.SY)
Cite as: arXiv:1910.01189 [eess.SY]
  (or arXiv:1910.01189v7 [eess.SY] for this version)

Submission history

From: Deepan Muthirayan [view email]
[v1] Wed, 2 Oct 2019 19:52:03 GMT (665kb,D)
[v2] Sun, 6 Oct 2019 08:16:14 GMT (665kb,D)
[v3] Tue, 10 Mar 2020 07:36:50 GMT (718kb,D)
[v4] Wed, 11 Mar 2020 00:56:24 GMT (718kb,D)
[v5] Thu, 12 Mar 2020 21:44:36 GMT (1914kb,D)
[v6] Mon, 16 Mar 2020 22:53:52 GMT (1916kb,D)
[v7] Thu, 19 Mar 2020 20:29:16 GMT (1914kb,D)

Link back to: arXiv, form interface, contact.