Technical Skills

  • C/C++, Java, Python, MATLAB, R, Android Development, Web (HTML/CSS/JS), LaTeX
  • Deep Learning Frameworks: Tensorflow, PyTorch, and OpenAI Gym

Selected Projects

Multi-Human Assisted Learning for Machine Agents using EEG, BCI DL/RL Python Tensorflow (Current)
with Prof. Raghupathy Sivakumar and Prof. Fekri, Georgia Tech,

  • Research, design and develop an interesting solution paradigm allowing humans to assist RL algorithms without burdening human-in-the-loop through EEG-based brain waves
  • Demonstration of the impact of our approach in improving state-of-the-art RL algorithms (e.g., DQN) by developing multiple Atari-like discrete-grid based games in OpenAI Gym
  • Experimentally showed that error-potentials can be learned in a zero-shot manner (with AUC > 0.8), and achieves a training acceleration of 2.25x while making 75.56% less queries

[arXiV preprint] [Paper-RL4G'20] [Poster-RL4G'20] [Paper-WearSys'20] [Code-WearSys'20] [Code-MazeGame]

Low-Power Command Detection for BCI Wearables, BCI ML Python (Aug'16-Aug'18)
with Prof. Raghupathy Sivakumar, Georgia Tech,

  • Proposed a wakeup command detection design and detection strategy enabling always-on BCI wearables to run on low-power mode achieving 2.7x improvement in battery life
  • Proposed BLINK, an algorithm to self-learn and detect eye-blinks in user brainwaves with 98% and low false-positive rate without requiring any user-training

[Data] [Paper-CHI'20] [Code-CHI'20] [Paper-ICC'20] [Paper-Allerton'19] [Code-Allerton'19]

THINK: Turning Thoughts into Action, BCI ML Python (Jan'15 - July'15)
with Prof. Raghupathy Sivakumar, Georgia Tech,

  • Developed THINK, a general purpose platform to communicate by mere imagination
  • Explored signal processing and detection of mu-waves, specifically in non-invasive domain (EEG)
  • Achieved counter-intuitive results for system accuracy (81.2\%), think rate and form-factor

[Paper-Hotwireless'15]

Skin Lesion Analysis towards Melanoma Detection, ML/DL Python Tensorflow Caffe Keras (Spring'16)
with Nandita Damaraju, Sahbi Chaieb and Prof. Zsolt Kira (Academic Course Project: Deep Learning)

  • Automated skin cancer detection by proposing CNN based architectures for skin lesion segmentation, feature extraction and classification
  • Won 2nd prize for the classification (81.3%) and feature extraction in ISBI 2016 Challenge
[Code] [Poster] [Report]

VisualAIDS: An Interactive visualization of HIV/AIDS data, d3 Javascript (Fall'17)
with Prof. Alex Endert, Georgia Tech,

  • Developed an interactive visualization designed in d3/js to investigate and explore HIV/AIDS data for various countries over time.

[Interface]

High-Dimensional Samping: A Machine Learning Approach, DL Python Tensorflow Summer'17
with Bhavya Kailkhura, J.J. Thiagarajan, Peer-Timo Bremer, Lawrence Livermore National Laboratory

  • Estimating Pair-Correlation Function with Neural Nets
  • Generating desired samples with GAN architectures

Video Action Classification using Deep Stateful Networks, ML/DL Python Tensorflow Summer'16
with Rob Liston and Dan Tan, Cisco Systems Inc.,

  • Designed deep neural nets using LSTMs in Tensorflow, for action recognition in video clips
  • Performed experiments to leverage temporal coherence in sequence of video frames
  • Quantified the comparison between stateful and stateless models in UCF-101 dataset

General Game Playing Agent, AI Java Spring'13
with Ujjawal Singh and Prof. Amitabha Bhattacharya (Academic Course Project: Artificial Intelligence)

  • Developed an artificial gaming agent, capable of playing any game without human intervention
  • Implemented state-of-the-art work published by CadiaPlayer (3 times winner in General Game Playing (GGP) competition in AAAI conference) which involves putting Upper Confidence Bound (UCB) in Monte-Carlo Tree Search (MCTS)
  • Selected amongst top 5 projects, to compete on a global scale in GGP competition in AAAI
[Presentation] [Poster] [Report]

OCR based Mathematical Equation Recovery, ML/CV Matlab Andoird App Spring'15
with Long Yun, Ma Meng, Wang Hongyang, and Prof. Mark Davenport (Academic Course Project: Statistical Signal Processing)

  • Developed a mathematical equation recognition system, capable of converting scanned mathematical equations to machine readable format along with its LaTeX code and solution
  • Implemented character segmentation and equation recovery algorithms for complex mathematical equations involving 2-D information (ex. superscript, summation, fraction etc.)
  • Achieved 98.92% accuracy in case of 60 symbols including operators and greek characters
[Code] [Poster] [Report]

Modelling the Rehearsal Effect of Humans, ML/DL Neuroscience Python Tensorflow Fall'16
with Stevens Christopher and Prof. Santosh Vempala (Academic Course Project: Computation and Brain)

  • Demonstrated the notion of forgetting and rehearsal in humans in the realm of neural networks
  • Successfully simulated Ebbinghaus forgetting curve and learning curve, and explored various rehearsal properties by building a Recurrent Neural Net in Tensorflow
[Code] [Poster] [Report]

NDroidPad: Android Application, Andoird App Winter'12
with Nikhil Gupta, (Electronics Club, IIT Kanpur)

  • Developed an Android application to use smartphone as wireless controller/joypad for Laptop
  • Implemented motion-sensor and multi-touch features, providing Xbox Interface to application
  • Won 1st Prize in Techfest’12; The app crossed 1000+ downloads within a week on Google Play

Graphic Driver on ATMEGA Microcontroller Platform, C Summer'11
with Mridul Verma, (Electronics Club, IIT Kanpur)

  • Developed Graphic Driver features on ATMEGA16, to project on external CRT/LCD monitor
  • Programmed VGA protocol using SRAM (as buffer memory) and SPI for high data rate
  • Created VGA library for text/image display, designed ping-pong demonstrating library usage