Interpretability in Multimodal Deep Learning Purvanshi Mehta, Antoine Ledent, Marius Kloft
TU Kaiserslautern, Germany
Developed a ‘p-norm multiple deep-subnetwork learning methodology’ which takes a weighted combination of the embeddings developed from different networks, and induces sparsity in the combination to give us more interpretable results.
Presentation at Lulea Technical University, Sweden - April 2019 [Slides]
Poster presentation at Deep Learning Reinforcement Learning Summer School, Edmonton - July 2019 [Poster]
Bot and gender profiling using Tweets Kovács G., Balogh V., Mehta P., Shridhar K., Liwicki M.
Mindgarage, Kaiserslautern Germany
Took part in author profiling using semantic and syntactic features. Top 5% of the competition. Paper accepted in CLEF - 2019 [Paper] [Poster]
Arithmetic word problem solving Mehta, P., Mishra, P., Athavale, V., Shrivastava, M. and Sharma, D.,
IIIT Hyderabad, India
Worked on building a system which solves simple arithmetic problems . System uses a deep neural architechtures and natural language processing to predict operators between the numerical quantities with an accuracy of 88.81\% in a corpus of primary school questions. Demo paper was published in International Joint Conference on Natural Language Processing - IJCNLP, 2017
Presentation at IJCNLP - 2017 at Taipei, Taiwan [Paper]
Relation Extraction from Text Mehta, P., Jat, P.
Dhirubhai Ambani Institue of Information and Communication Technology
Worked on relation extraction from text using machine learning techniques(SVM’s) and distant supervision. My work involved understanding basics of natural language processing and information extraction. Paper review of three papers involving improving of distant supervision algorithm for automatically creating labeled training data, end to end relation extraction using distant supervision from external repositories and relation extraction on unlabelled corpora.
Presentation at DAIICT - 2016 [Paper] [Code]
You may reach me at purvanshi.mehta11 [at] gmail [dot] com
© 2019 Purvanshi Mehta