Temporal Coherence in Energy-based Deep Learning Machines for Action Recognition

Adepu, R S (2014) Temporal Coherence in Energy-based Deep Learning Machines for Action Recognition. Masters thesis, Indian Institute of Technology, Hyderabad.

Preview

Text
CS11M03.pdf - Submitted Version
Download (901kB)

Abstract

Deep Learning, a sub-area of machine learning, has become a buzz word in recent days due to its great successes in many applications of machine learning, including speech processing, computer vision and natural language processing. Deep learning became famous in the initial days through the successful application of Convolutional Neural Networks as well as Energy-based Models -or Restricted Boltzmann Machines (RBMs) - on handwritten digit recognition. While the last decade has seen the growing use of convolution-based deep learning methods for image analysis, limited work has been done in adapting deep learning to video analysus. Existing methods have largely extended the ideas based on convolution applied to images into the video analysis setting. The primary deep learning approaches that have been proposed so far explicitly for video sequences are the 3D Convolutional Neural Networks and the Convolutional Gated RBM.

[error in script]

IITH Creators: