Experience

Max Planck Schools

• Germany
• Research Services
• 1 - 100 Employee

• Doctoral Student

  • 2020 - Present

  Max Planck School of Cognition Max Planck School of Cognition



Indian Institute of Technology, Delhi

• Higher Education
• 700 & Above Employee

• Undergraduate Research Student

  • Aug 2018 - 2020

  Title: Breast Cancer Detection in Mammograms using Deep Neural Networks. Advisor: Prof. Chetan Arora, IIT Delhi This work is in collaboration with AIIMS Delhi. Our aim is to develop object detectors which can accurately locate and classify benign and malignant lesions in Indian breast mammograms. Despite being a well researched problem, nearly no research is devoted to Indian breast types. In India we observe that breast cancer hits at a younger age of 40s which leads to difference in tissue structures and compositions. In India people tend to have denser tissue, which justifies the need for specific analysis. We first reproduce the state of the art techniques on foreign datasets following which we apply them on Indian data. We have designed an ensemble network for dealing with masses and calficifications separately and have used multi-scale object detectors to take into account the size variation prevalent among the lesions. Show less



Indraprastha Institute of Information Technology, Delhi

• India
• Higher Education
• 500 - 600 Employee

• Undergraduate Research Student

  • Aug 2018 - 2020

  Title: Data Sonification for Social SensitizationAdvisor: Prof. Timothy Scott Moyers, University of KentuckyData sonification is a novel means on using sound to represent data. I collected 12 year data on Crime Against Women in Indian States, and developed a sonification interface in Supercollider audio synthesis software. I implemented several functions to map data to Synths and sound sequences. The scream sounds selected were aimed to bring an impact to the rising crime rates. I also performed sound spatialization to add to the effect and conducted a survey to obtain user feedback on appropriateness of sound mappings and the sonification techniques adopted. Show less

• Undergraduate Research Student

  • Jan 2019 - Apr 2019

  Title: Protein-Ligand Scoring with Convolutional Neural NetworksAdvisor: Prof. G. P. S. Raghava, IIIT DelhiReproduced the paper Protein-Ligand Scoring with Convolutional Neural Networks by Matthew Ragoza et. al. The paper describes the use of protein-ligand binding data for deep learning to predict binding affinities and poses. The authors have developed a CNN which can enable automatic learning of key features of protein-ligand interactions and how they correlate to binding. A neural network is trained on 3D grid representations of protein-ligand structures generated through docking. They have proved that their CNN based scoring function outperforms the scoring function of AutoDock Vina which were also verified and confirmed by my reproduced results. Show less

• Teaching Assistant

  • Jan 2019 - Apr 2019

  Teaching Assistant for 2nd year course on Introduction to Quantitative Biology

• Undergraduate Research Student

  • Jan 2019 - Apr 2019

  Title: Universal EEG Encoder for Learning Diverse Intelligent TasksPublication: B. L. K. Jolly*, P. Aggarwal*, S. S. Nath*, V. Gupta*, M. S. Grover, R. R. Shah, IEEE Big Multimedia, Singapore, Sep 11-13 2019Advisor: Prof. Rajiv Ratn Shah, IIIT DelhiBrain Computer Interfaces (BCI) are being increasingly exploited for creating a direct communication between the human brain and an external agent. Electroencephalography (EEG) is one of the most commonly used signal acquisition techniques due to its non-invasive nature, and high temporal resolution. One of the major challenges in BCI studies is the individualistic analysis required for each task. Thus, task-specific feature extraction and classification are performed, which fails to generalize to other tasks with similar time-series EEG input data. Towards this end, we design a GRU-based universal deep learning encoding architecture to extract meaningful features from publicly available datasets for five diverse EEG-based classification tasks namely Emotion Detection, Digit Recognition, Object Recognition, Task Identification and Error Detection. The network can generate task and format independent embeddings and outperforms the state of the art EEGNet architecture on most experiments. Such a representation can enable efficient analysis across multiple datasets and eliminate the need to manually extract handcrafted features pertaining to every task. We also compare our results with CNN-based, and Autoencoder network, in turn performing local, spatial, temporal and unsupervised analysis on the data. Show less



University of Southern California

• United States
• Higher Education
• 700 & Above Employee

• IUSSTF Viterbi Research Intern

  • May 2019 - Jul 2019

  Title: Human decision making using reinforcement learning-based simulation and estimation Advisor: Prof. Shinyi Wu, Prof. Haomiao Jin, USC Worked individually with faculty on their research projects. Our aim was to develop a Reinforcement Learning Statistical model for modeling Human Decision Making through simulation and estimation. We developed an ensemble model-free and model-based network to accurately model instinctive and thought-driven processes. The code was compiled into a user-friendly interface with several parameters and options to control. We implemented RL learning techniques including Q-learning, SARSA, policy evaluation, policy improvement and estimation/optimization strategies such as local optimization, global optimization, constrained optimization, maximum likelihood estimation and cross entropy loss based estimation. Show less



TCS Research & Innovation

• Gurgaon, India

• Intern

  • Dec 2018 - Jan 2019

  Title: Enabling blind to read handwritten text Advisor: Ms. Ramya Hebbalaguppe, TCS Innovation labs Worked on a system for enabling fully and partially blind to read handwritten text. The system consists of an Android application which allows the user to scan/upload an image of handwritten text. The lines are decoded using an LSTM word level encoder-decoder. The detected text is forwarded into a speech engine where following which, a real scene is opened up where the text is rendered in using Augmented Reality. I integrated the speech module, worked on model compression using Tensorflow Lite and developed the augmented reality text rendering interface. Show less



Indian Institute of Technology Gandhinagar

• India
• Higher Education
• 700 & Above Employee

• SRIP Research Intern

  • May 2018 - Jul 2018

  Title: Brain Computer Interface Advisor: Prof. K. P. Miyapuram, IIT Gandhinagar 1) Worked on P300 Speller for ALS patients. The P300 Speller is a communication device which uses the Event Related Potential generated in the brain to detect the alphabet being spelled out. I worked on character recognition and error analysis by implementing several machine learning techniques on the data. 2) Worked on Emotion Detection. Data was collected at IIT Gandhinagar Dept of Cognitive Science. 9 emotional movie clips each representing a Navarasa was administered on 20 participants. We developed a novel merge-and-split clustering algorithm to observe the trends in the data. The results revealed that inter participant variation to all stimuli was significantly lower than cross participant responses to the same stimuli indicating that emotional responses could be a characteristic trait of every individual. Publication: S. S. Nath, D. Mukhopadhyay, K. P. Miyapuram, Emotive Stimuli-triggered Participant-based Clustering using a Novel Split-and-Merge Algorithm, YRS CODS-COMAD, Kolkata, Jan 3-5 2019. Show less



Indian Statistical Institute (ISI), New Delhi

• India
• Research Services
• 1 - 100 Employee

• Summer Student

  • May 2017 - Jul 2017

  Title: The Knapsack Problem and Its Algorithms Advisor: Prof S. K. Neogy, ISI Delhi The knapsack problem is a famous problem in optimization: "Given a set of items, each with a weight and a value, determine the number of each item to include in a collection so that the total weight is less than or equal to a given limit and the total value is as large as possible". I studied and coded greedy and DP algorithms for fractional knapsack problem where part of an item can be picked; and 0/1 knapsack where only the whole item can be picked. I also researched about the multidimentional knapsack which is an NP hard problem. Performed complexity analysis and used TORA integer programming software to test solutions. Show less