Research

Research

I'm doing my Master study on A.I. research, technically on neural networks, especially Multi Layer Perceptron (MLP) trained with back-propagation, Self Organizing Map (SOM), auto-encoder networks and Restricted Boltzmann Machine (RBM) based neural networks. The methods are implemented for recognition purposes, like face recognition and optical character recognition (OCR) for my case.

Currently I'm focusing on auto-encoder which helps to reduce the dimensionality of problems. It is believed should give a better precision compared to PCA, a non-lineaer statistical method which I also very familiar with. Furthermore I shall train the networks by initializing the network weights with RBM, factoring the weights leads to much faster convergence.

The experiments are carried out on Olivetti Research Laboratory (ORL) face database and MNIST hand-written character database, a subset of a larger set of database available from National Institute of Standards and Technology (NIST). ORL face database consists of 40 person images, 10 images per person, total 400 images. I use half of the images for training and the rest for testing. While MNIST character database consits of 60,000 characters for training and 10,000 characters for testing.

If you are in the same field or need further clarifications, welcome to drop me an email to have a detailed discussion for the good of developing and growing of knowledge.

International Conferences

  1. C. C. Tan, and C. Eswaran, "Reconstruction of Handwritten Digit Images Using Autoencoder Neural Networks", 21st IEEE Canadian Conference on Electrical and Computer Engineering, Niagara Falls, Ontario, Canada, May 2008, pp. 465-469.
  2. C. C. Tan, and C. Eswaran, "Performance Comparison of Three Types of Autoencoder Neural Networks", Second Asia International Conference on Modelling and Simulation 2008, Kuala Lumpur, Malaysia, May 2008, pp. 213-218.