سیویلیکا را در شبکه های اجتماعی دنبال نمایید.

Enhancing Spatial Pooler Performance in Hierarchical TemporalMemory Algorithm through Sparsification Analysis: An Information TheoryPerspective

Publish Year: 1403
Type: Conference paper
Language: English
View: 58

This Paper With 11 Page And PDF Format Ready To Download

Export:

Link to this Paper:

Document National Code:

CECCONF23_002

Index date: 19 August 2024

Enhancing Spatial Pooler Performance in Hierarchical TemporalMemory Algorithm through Sparsification Analysis: An Information TheoryPerspective abstract

Hierarchical Temporal Memory (HTM) is an unsupervised machine learning algorithm inspired byneocortical computational principles. The Spatial Pooler (SP), a core component of HTM, converts binaryinput into sparse distributed representations. This study examines SP's sparsification through aninformation theory perspective, demonstrating that increased sparsity enhances SP's performance.Comparative analyses using Gaussian and non-Gaussian (e.g., Cauchy distribution) data distributionsreveal that sparsity levels significantly impact SP's output, as assessed by the Cramer–Rao lower bound.Our findings highlight the critical role of sparsity in optimizing SP's performance and offer insights forthe design and optimization of HTM algorithms

Enhancing Spatial Pooler Performance in Hierarchical TemporalMemory Algorithm through Sparsification Analysis: An Information TheoryPerspective Keywords:

Spatial Pooler (SP) , Hierarchical Temporal Memory (HTM) , Sparsity , Fishery informationmatrix (FIM) , Cramer-Rao Lower Bound (CRLB)

Enhancing Spatial Pooler Performance in Hierarchical TemporalMemory Algorithm through Sparsification Analysis: An Information TheoryPerspective authors

Shiva Sanati

Dept. of Computer Engineering, Ferdowsi University of Mashhad, Mashhad, Iran

Modjtaba Rouhani

Dept. of Computer Engineering, Ferdowsi University of Mashhad, Mashhad, Iran

Ghosheh Abed Hodtani

Dept. of Electrical Engineering, Ferdowsi University of Mashhad, Mashhad, Iran