blog-img
blog-img
blog-img
blog-img
blog-img
blog-img
blog-img
blog-img
blog-img
blog-img
blog-img
blog-img
blog-img
blog-img

Music Genre Classification using Convolutional Recurrent Neural Networks
CS725 Machine Learning | Course Project

  • Developed a music genre classifier based on a ISMIR 2016 paper, using convolutional recurrent neural networks with mel-spectrogram of the audio as input feature vector.
  • Achieved 0.84 top-3 accuracy on Marsyas dataset.
  • Brain Tumor Segmentation using Convolutional Neural Networks
    Data Science Team | Fractal Analytics

  • Developed an AI tool to automatically detect, segment and classify tumors in MRIs, and suggest corresponding diagnosis methods using deep learning techniques as convolutional neural networks achieving 64% segmentation accuracy.
  • Utilized image data pre-processing techniques like symmetry-subtraction, localized image quantization and deployed 3Dand Triplanar Convolutiontal Neural Networksin Keras on Theano to automatically segment the tumors.
  • Seizure detection from EEG Signals using Convolutional Bi-RNNs
    EE779 Advanced Topics in Signal Processing | Course Project

  • Implemented a ICLR 2016 paper with improvements and modifications to extract robust spatio-temporal features as image representations of the FFT of the polar projected EEG signals and trained a recurrent convolutional neural network to achieve 0.62 AUC score.
  • Carrom-RL - Reinforcement Learning based agent to play carrom.
    CS747 Intelligent and Learning Agents | Course Project

  • Developed an reinformcement learning agent to play carrom in single and multi-player settings.
  • Explored and implemented reinforcement learning various techniques as deep Q-learning (DQN), deep deterministic policy gradients (DDPG) and geometric approaches.
  • Convolutional Neural Networks for Image Classification
    CS 663 Digital Image Processing | Course Project

  • Implemented a Convolutional Neural Network in MATLAB from scratch for object based image classification using several feature extraction techniques and data precprocessing using ZCA whitening.
  • Achieved competiitve results on CIFAR-10, STL-10 and MNIST datasets
  • .

    PCA Based Face Recognition and Reconstruction
    CS 663 Digital Image Processing | Course Work

    CUDA Library for High Performance Computer Vision.
    ME766 High Performance Scientific Computing | Course Project

  • Implemented popular Image Processing and Computer Vision algorithms to CUDA kernels for improved execution times.
  • Achieved speed gain around 3x to 6x over non-GPU accelerated code for Adaptive Histogram Equalization, Gaussion Noise Filters, S.T. Corner Detection etc.
  • Digital Image Processing Techniques
    CS 663 Digital Image Processing | Course Work

    8-bit Pipelined & Multicycle RISC Processor Design
    EE309 Microprocessors | Course Work

  • Designed 8-bit Multicycle and Pipelined Processors based on Little Computer Architecture from scratch and synthesized the same on Altera DE0 Nano FPGA.
  • Used Branch Predictors, Priority Encoders and Hazard Detection Units to reduce latencies and significantly reduce the branch hazards.
  • Synthesizeable Arbitrary Precision Floating and FixedPoint Library
    Architectures for Computation Acceleration Team, Advanced Digital Sciences Center

  • Developed bit-width independent Fixed Point Integer and Arbitrary Precision Floating Point C++ libraries compatible with High Level Synthesis HLS Tools to accelerate and optimize hardware descriptions
  • Implemented modified versions of arithmetic operations of fixed point integers and floating points to achieve comparable results with fixed point libraries in industry level HLS tools such as Xilinx Vivado and Calypto Catapult.
  • Presently the only library with fixed points compatible across all major HLS tools with competitive performance and precision.
  • Matsya - Autonomous Underwater Vehicle IIT-Bombay
    IIT Bombay

  • Developing and enhancing the Software & Control System for AUV (Autonomous UnderWater Vehicle) Matsya 3.0 which included fetching data from visual, inertial, acoustic & depth sensors planted on the vehicle and analysing the data to perform real-life tasks using Robotic Operating System (ROS).
  • SemiFinalists at RoboSub 2013, 2014.
  • Capacited Vehicle Routing Problem with Time Windows
    Quantitative Methods Lab | IIM Ahmedabad

  • Developed mathematical models for k-period symmetric capacited travelling salesman problem with time windows using novel subtour elimination constraints and Branch & Bound techniques as a part of an Industrial Supply Chain Optimization Project by Britannia Industries Ltd.
  • Implemented the model as a combinatorial optimization problem in IBM-ILOG CPLEX solver to produce quick integer programming solutions.
  • Vishynoid - Physical Chess Playing Robot
    Institute Technical Summer Project | IIT Bombay

  • One of the Top 10 Summer Projects out of about 100 projects in 2014 selected for display at Science EXPO 2014.
  • Developing and optimizing the Artificial Intelligence of the bot using α-β pruning to minimize the moves-search tree and used NegaMax function to analyse the score at every node.
  • Used edge and color detection to detect human moves, implementing 2-axis robotic grippers to move the chess pieces,with all threads running on Beagleboard Black.
  • TechFest 2k15 Android Application | IIT-Bombay
    Techfest 2015 | IIT Bombay

  • Made a first of its kind and highly appreciated event experience enhancement app that provided a list of events happening at the current time of the day in a visually categorized on a dynamically updated map and navigation to events during the festival from users location.
  • Rated 4.9/5.0 and over 5k downloads.
  • Google Drive Schedules and Cache Optimization to minimize app updates and reduce network dependence respectively
  • R-Based Statistical Computing Interface.

  • Developed a R-based Django web interface for statisticians to upload data in standard formats and providing statistical insights into the data and interactive and exportable visualizations.
  • Implemented R query optimization using the OpenCPU Architecture to make the interface independent of the request load and used d3 and highcharts for interactive visualization.

  • IIT-Bombay ASC Visualization Toolkit.
    ME766 Software Development | Course Project

  • Django based app to display ASC IIT-Bombay data in a interactive and popular formats with features of automatically generating Google Calendar compatible iCalender files and HTML timetables of student schedules.
  • Developed an API in python which uses browser emulation tool selenium to bypass the security measures of ASC website to extract data.
  • Music Genre Classification using Convolutional Recurrent Neural Networks
    CS725 Machine Learning | Course Project

  • Developed a music genre classifier based on a ISMIR 2016 paper, using convolutional recurrent neural networks with mel-spectrogram of the audio as input feature vector.
  • Achieved 0.84 top-3 accuracy on Marsyas dataset.
  • Brain Tumor Segmentation using Convolutional Neural Networks
    Data Science Team | Fractal Analytics

  • Developed an AI tool to automatically detect, segment and classify tumors in MRIs, and suggest corresponding diagnosis methods using deep learning techniques as convolutional neural networks achieving 64% segmentation accuracy.
  • Utilized image data pre-processing techniques like symmetry-subtraction, localized image quantization and deployed 3Dand Triplanar Convolutiontal Neural Networksin Keras on Theano to automatically segment the tumors.
  • Seizure detection from EEG Signals using Convolutional Bi-RNNs
    EE779 Advanced Topics in Signal Processing | Course Project

  • Implemented a ICLR 2016 paper with improvements and modifications to extract robust spatio-temporal features as image representations of the FFT of the polar projected EEG signals and trained a recurrent convolutional neural network to achieve 0.62 AUC score.
  • Carrom-RL - Reinforcement Learning based agent to play carrom.
    CS747 Intelligent and Learning Agents | Course Project

  • Developed an reinformcement learning agent to play carrom in single and multi-player settings.
  • Explored and implemented reinforcement learning various techniques as deep Q-learning (DQN), deep deterministic policy gradients (DDPG) and geometric approaches.
  • Convolutional Neural Networks for Image Classification
    CS 663 Digital Image Processing | Course Project

  • Implemented a Convolutional Neural Network in MATLAB from scratch for object based image classification using several feature extraction techniques and data precprocessing using ZCA whitening.
  • Achieved competiitve results on CIFAR-10, STL-10 and MNIST datasets
  • .

    Capacited Vehicle Routing Problem with Time Windows
    Quantitative Methods Lab | IIM Ahmedabad

  • Developed mathematical models for k-period symmetric capacited travelling salesman problem with time windows using novel subtour elimination constraints and Branch & Bound techniques as a part of an Industrial Supply Chain Optimization Project by Britannia Industries Ltd.
  • Implemented the model as a combinatorial optimization problem in IBM-ILOG CPLEX solver to produce quick integer programming solutions.
  • PCA Based Face Recognition and Reconstruction
    CS 663 Digital Image Processing | Course Work

    CUDA Library for High Performance Computer Vision.
    ME766 High Performance Scientific Computing | Course Project

  • Implemented popular Image Processing and Computer Vision algorithms to CUDA kernels for improved execution times.
  • Achieved speed gain around 3x to 6x over non-GPU accelerated code for Adaptive Histogram Equalization, Gaussion Noise Filters, S.T. Corner Detection etc.
  • Digital Image Processing Techniques
    CS 663 Digital Image Processing | Course Work

    Matsya - Autonomous Underwater Vehicle IIT-Bombay
    IIT Bombay

  • Developing and enhancing the Software & Control System for AUV (Autonomous UnderWater Vehicle) Matsya 3.0 which included fetching data from visual, inertial, acoustic & depth sensors planted on the vehicle and analysing the data to perform real-life tasks using Robotic Operating System (ROS).
  • SemiFinalists at RoboSub 2013, 2014.
  • Music Genre Classification using Convolutional Recurrent Neural Networks
    CS725 Machine Learning | Course Project

  • Developed a music genre classifier based on a ISMIR 2016 paper, using convolutional recurrent neural networks with mel-spectrogram of the audio as input feature vector.
  • Achieved 0.84 top-3 accuracy on Marsyas dataset.
  • Seizure detection from EEG Signals using Convolutional Bi-RNNs
    EE779 Advanced Topics in Signal Processing | Course Project

  • Implemented a ICLR 2016 paper with improvements and modifications to extract robust spatio-temporal features as image representations of the FFT of the polar projected EEG signals and trained a recurrent convolutional neural network to achieve 0.62 AUC score.
  • 8-bit Pipelined & Multicycle RISC Processor Design
    EE309 Microprocessors | Course Work

  • Designed 8-bit Multicycle and Pipelined Processors based on Little Computer Architecture from scratch and synthesized the same on Altera DE0 Nano FPGA.
  • Used Branch Predictors, Priority Encoders and Hazard Detection Units to reduce latencies and significantly reduce the branch hazards.
  • Synthesizeable Arbitrary Precision Floating and FixedPoint Library
    Architectures for Computation Acceleration Team, Advanced Digital Sciences Center

  • Developed bit-width independent Fixed Point Integer and Arbitrary Precision Floating Point C++ libraries compatible with High Level Synthesis HLS Tools to accelerate and optimize hardware descriptions
  • Implemented modified versions of arithmetic operations of fixed point integers and floating points to achieve comparable results with fixed point libraries in industry level HLS tools such as Xilinx Vivado and Calypto Catapult.
  • Presently the only library with fixed points compatible across all major HLS tools with competitive performance and precision.
  • Vishynoid - Physical Chess Playing Robot
    Institute Technical Summer Project | IIT Bombay

  • One of the Top 10 Summer Projects out of about 100 projects in 2014 selected for display at Science EXPO 2014.
  • Developing and optimizing the Artificial Intelligence of the bot using α-β pruning to minimize the moves-search tree and used NegaMax function to analyse the score at every node.
  • Used edge and color detection to detect human moves, implementing 2-axis robotic grippers to move the chess pieces,with all threads running on Beagleboard Black.
  • TechFest 2k15 Android Application | IIT-Bombay
    Techfest 2015 | IIT Bombay

  • Made a first of its kind and highly appreciated event experience enhancement app that provided a list of events happening at the current time of the day in a visually categorized on a dynamically updated map and navigation to events during the festival from users location.
  • Rated 4.9/5.0 and over 5k downloads.
  • Google Drive Schedules and Cache Optimization to minimize app updates and reduce network dependence respectively
  • R-Based Statistical Computing Interface.

  • Developed a R-based Django web interface for statisticians to upload data in standard formats and providing statistical insights into the data and interactive and exportable visualizations.
  • Implemented R query optimization using the OpenCPU Architecture to make the interface independent of the request load and used d3 and highcharts for interactive visualization.

  • IIT-Bombay ASC Visualization Toolkit.
    ME766 Software Development | Course Project

  • Django based app to display ASC IIT-Bombay data in a interactive and popular formats with features of automatically generating Google Calendar compatible iCalender files and HTML timetables of student schedules.
  • Developed an API in python which uses browser emulation tool selenium to bypass the security measures of ASC website to extract data.