About Me
Publications
- Siddhartha Arora, Mitesh M. Khapra, Harish Guruprasad Ramaswamy: On Knowledge Distillation from Complex Networks for Response Prediction. NAACL 2019 link
- Nikita Moghe, Siddhartha Arora, Suman Banerjee and Mitesh M. Khapra: Towards Exploiting Background Knowledge for Building Conversation Systems. EMNLP 2018 link
- Suman Banerjee, Nikita Moghe, Siddhartha Arora, Mitesh M. Khapra: A Dataset for Building Code-Mixed Goal Oriented Conversation Systems. COLING 2018 link
Academic Details
I have completed my B.Tech in Information Technology from G.L.Bajaj Institute Of Technology and Management, Greater Noida in 2012.
Work Experience
I worked as Assistant System Engineer at TATA CONSULTANCY SERVICES from May 2013 - Jan 2015. I worked as a Java developer and Automation tester in various applications of project The Home Depot .
Projects
-
Knowledge Distillation for Response Prediction Task
Designed a simpler deep neural network model for recently released Holl-E dataset. Then performed various knowledge distillation techniques from complex models like BiDAF (Bidirectional Attentional Flow) and QANet. Finally obtained an improved F1 score when trained using Student (Simpler model) - Teacher (BiDAF or QANet) model.
-
Sequential Generation and Span Prediction
Created a deep neural network model which learns whether to generate a word or copy a span at every time step of the decoder. Then formulated an optimization function that handles generation and span prediction task on Holl-E dataset.
-
Multilingual Dialogue Generation
Implemented Hierarchical Recurrent Encoder Decoder Model for multilingual (English and French) dialogue data. Experimented with language (English and French) aware and language agnostic variations of the baseline model.
-
Solving Online Travelling Salesman Problem (TSP) by Combinatorial
Implemented different versions of Exp (“exponential-weight algorithm for exploration and exploitation") algorithm to test under what conditions which algorithm performs better for Travelling Salesman Problem. Then provided insights into theoretical guarantees for Exp3semi and Exp4comb algorithms. Lastly performed regret analysis under semi-bandit and bandit feedback case of Travelling Salesman Problem.
-
Semi-supervised feature learning for collective classification on sparsely
Implemented semi-supervised stacked denoising auto-encoder to learn efficient representations for large attributed graphs. Then proposed graph regularization which shows improvement over the vanilla version of stacked denoising auto-encoder.
Contact:
sidarora[at]cse[dot]iitm[dot]ac[dot]in
sidarora1990[at]gmail[dot]com