Distributed algorithms for prototype selection and multi-label classification

Speaker:
Dr. V. Susheela Devi
Principal Research Scientist
Dept. of CSA

Abstract

The Modified Condensed Nearest Neighbour (MCNN) algorithm is an algorithm for prototype selection which gives good performance but the time requirement is much higher than algorithms like Condensed Nearest Neighbour (CNN) algorithm. A distributed approach called Parallel MCNN (pMCNN) is proposed which cuts down the time required drastically. pMCNN has been used for prototype selection on large and streaming data and results are presented which shows good performance with saving in time.

The second part of the talk describes an algorithm for multilabel classification for discrete data. Two improvements are carried out to reduce the complexity. The first is feature selection which reduces both space and time complexity. The second is to use a distributed approach to this problem. Results show good performance as compared to standard techniques like ML-KNN (Multi label kNN). The use of feature selection and the distributed approach give saving in the time required which is helpful when the data sizes are large.

Speaker Bio:
V. Susheela Devi works in the Intelligent Systems group in the department. Her areas of interest include pattern recognition, machine learning and soft computing. She is the co-ordinator of the Pattern Analysis and Machine Intelligence lab. She has handled courses such as Pattern Recognition, Data Mining, Data Structures and Algorithms, Computational Methods of Optimization, Artificial Intelligence and Soft Computing.

Host Faculty: Prof. Sunil L Chandran & Prof. Shalabh Bhatnagar

Video Archive