Research Work & Summary
- N. Sharma, K. Masalia, "Analysis on Complementary Count Min Sketch,"
(Independent Study Report.pdf)
- Proposed and analyzed a novel data structure called Complementary Count Min Sketch (CCMS) to support deletion in count min sketch.
- Compared the performance of CCMS with different sketches for feature selection in high dimensional data settings.
- Received comparable performance with Count Sketch under Power and Zipf's law distribution settings.
- N. Sharma, A. D. Dileep, and V. Thenkanidiyoor, "Text Classification using Hierarchical Sparse Representation Classifiers,"
16th IEEE International Conference on Machine Learning and Applications (ICMLA), Cancun, 2017, pp. 1015-1019.
(paper_2.pdf)
- Built a Hierarchical Sparse Representation Classifier (HSRC) and explored it for text classification.
- Explored weighted decomposition principal component analysis (WDPCA) technique to highlight the middle principal components to construct the dictionary that emphasizes discrimination among confusing classes.
- HSRC classified 83.30% of the documents correctly opposed to the 78.78% (using SRC) with WDPCA on 20 Newsgroup corpus.
- N. Sharma, A. Sharma, V. Thenkanidiyoor, and A. D. Dileep, "Text classification using combined sparse representation classifiers and support vector machines," 4th International Symposium on Computational and Business Intelligence (ISCBI),
Olten, 2016, pp. 181-185. (paper_1.pdf)
- Explored frequency based kernels such as Histogram intersection kernel, Χ2-kernel and Hellinger’s kernel for text classification using SVMs.
- Combined the SRC and SVM classifiers based on voting scheme. Obtained 81.83% accuracy using on 20 Newsgroup corpus.