Projects
Semantic Influence Score
Statistics and Data Analytics
A novel metric was developed to quantify the quality of a published scientific article.
The developed metric follows a log-normal distribution and counters the limitations of the raw citation count which adheres to an exponential distribution. Findings have been published in Springer.
Raga Identification in Real-Time
Audio processing and Deep Learning
Raga is a feature of Indian Classical Music, that characterises a tune. A deep learning model with Temporal Convolutional Networks was developed to detect Ragas from a piece of Hindustani Classical Music in Real-Time by using pitch based features. 87% test accuracy was obtained.
Intrusion detection in cloud containers
Cloud Computing and Machine Learning
Attacks on Kubernetes pods are detected in reconnaissance stage by detecting anomalous system calls among system call traces. Self Organising Maps, One-Class SVMs and LSTM auto encoders have been used with a semi supervised learning approach.
Neural net trained using Eagle Strategy
Metaheuristic Optimization
Neural network was trained on Eagle Strategy employing Levy flight for global search and Simulated Annealing for local search. The training time was reduced 3 times while maintaining the same accuracy.
HTML Code generation from screenshots
Deep Learning
HTML Code was generated from screenshots using encoder CNN to learn features, encoder RNN to learn the language and decoder RNN to generate code. LSTM and GRU cells were compared to conclude that GRU's learn faster. Obtained BLEU score of 78%.
Binary classification of stars and quasars
Machine Learning
The classes being linearly and non linearly inseparable, Stars and quasars were classified using telescopic photometry data using Decision Trees. The model was encoded in Python from scratch to obtain an accuracy of 91% across 4 datasets.