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


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


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

Computer Science > Information Theory

Title: Distributed Sequence Memory of Multidimensional Inputs in Recurrent Networks

Abstract: Recurrent neural networks (RNNs) have drawn interest from machine learning researchers because of their effectiveness at preserving past inputs for time-varying data processing tasks. To understand the success and limitations of RNNs, it is critical that we advance our analysis of their fundamental memory properties. We focus on echo state networks (ESNs), which are RNNs with simple memoryless nodes and random connectivity. In most existing analyses, the short-term memory (STM) capacity results conclude that the ESN network size must scale linearly with the input size for unstructured inputs. The main contribution of this paper is to provide general results characterizing the STM capacity for linear ESNs with multidimensional input streams when the inputs have common low-dimensional structure: sparsity in a basis or significant statistical dependence between inputs. In both cases, we show that the number of nodes in the network must scale linearly with the information rate and poly-logarithmically with the ambient input dimension. The analysis relies on advanced applications of random matrix theory and results in explicit non-asymptotic bounds on the recovery error. Taken together, this analysis provides a significant step forward in our understanding of the STM properties in RNNs.
Comments: 37 pages, 3 figures
Subjects: Information Theory (cs.IT); Machine Learning (stat.ML)
Journal reference: Journal of Machine Learning Research, 18:1-37 Jan. 2017
Cite as: arXiv:1605.08346 [cs.IT]
  (or arXiv:1605.08346v3 [cs.IT] for this version)

Submission history

From: Adam Charles [view email]
[v1] Thu, 26 May 2016 16:02:26 GMT (136kb,D)
[v2] Fri, 4 Nov 2016 13:14:31 GMT (143kb,D)
[v3] Fri, 27 Jan 2017 15:03:53 GMT (469kb,D)

Link back to: arXiv, form interface, contact.