Srinivas S. Kulkarni

Real Time Voice Activity Detection Using ConvNet

Advisor : Dr. Nasser Kehtarnavaz • 2016 — Present

Efficiently detects Voice and Noise signals. Using tensorflow, to create model and test it on dataset. Training on ConvNet 13 layer architecture . Implementing the trained model on smartphone. Real time constraint such as time complexity are taken care. Working on graph quantization, for memory constraint. Currently accuracy is 90%, still working on feature engineering to improve.

University Of Texas At Dallas

MS, Electrical Engineering • 2015 — May-2017

Sri Jayachamrajendra College Of Engineering, Mysuru. India

BS, Electronics and Computer Engineering • 2009 — 2013

Time-Series Regression using LSTM Recurrent Neural Network

TensorFlow • Spring — 2017

Generated own dataset (ex: cosine values). Implemented using Dynamic RNN with cell size of 10. Used Tensorboard to visualize the graph structure, weights and loss function. Working on LSTM to improve results.

Credit Card Approve or Reject

Scikit-learn • Spring — 2017

Used Pandas to fill in missing values and standardise all the continuous inputs. Categorical features were binarized and labels were converted to one hot encoding. Used SVM with kernel to fit the classifier, as samples were limited to train (700 samples). Accuracy on test data was 91%

Stock Predictions

Scikit-Learn • Fall— 2016

Supervised machine learning technique on stock prices of two months. Used Support Vector Regression algorithm (scikit-learn) to predict the missing values. Major part spent on receiving and cleaning data.

Titanic and Forest Cover Prediction (Kaggle Competition)

Python • Fall — 2016

As part of Kaggle competition. Used Python language for implementation(Scikit learn). Using Pandas Input samples were cleaned and missing values were filled in. Implemented using RandomForests classifier for both projects. Code is available in github and can be ran using Anaconda Jupyter Notebook.

Webscraping and Data Analysis on Social Security Website

Python regular expression • Fall — 2016

Baby names from Social Security website for period of 10 years was scraped using regular expressions (Python). Arranged and plotted most/least occuring name in each year. Results were interesting. This is part of Google Python class project.

MNIST digit recognition

TensorFlow • Fall — 2016

MNIST is collection of handwritten digits' images with 40*40 dimension. Using them as dataset, ran ConvNet to classify each digit, Used Tensorflow to implement the Architecture. Accuracy : 99%

Recommendation System Using MovieLens Dataset

Python • Summer — 2016

Implemented Item-Item Collaborative filtering algorithm. Used Cosine similarities to recommend movies. Pandas (data extraction) and Scikit-learn (algorithm and performance metrics) module to obtain ROI curve.

Gender Classification Based on Heights and Weights information

Scikit-Learn • Summer — 2016

GridSearchCV : Cross Validation. PCA : Dimensionality Reduction. Lambda Function : "Just because I know how to use it" :P KNN : Classification Algorithm. This Project was mainly for understanding How Cross Validation helps choosing best hyper parameter (Cross Validation), in our case number of neighbors.

Implementation of Signal Processing Algorithm on a smartphone

MATLAB, C • Spring — 2016

Used C language to implement on ARM processor. (MathCoder). Developed and optimized fixed point and floating point coding. Implemented Doppler Effect, DTMF Decoder, Real time filtering (FIR), Adaptive filtering and FFT using MATLAB.

Delphi Automotive Systems

Software Engineer • Aug, 2013 — July 2015

Programmed and tested Electronic Control Module (Name: DCM 2.5) for Hyundai Motors. Worked on CAN communication protocol and feature like Fuel Injection in Daimler Project. Extensively used Polyspace (Static testing) and RTRT (Dynamic testing) tools for unit testing embedded systems.

• Manged team of two engineers on Daimler car project for a month
• Added extra efforts, worked on weekends during time critical project
• Active member in organising fun events (team outing)

Programming Languages

Python, MATLAB, C++, C

Operating Systems

Linux, Windows, Basics of Android

Libraries and Tools

Tensorflow, Matconvnet , Numpy, SciPy, Matplotlib, Scikit-Learn, Pandas , PostgreSQL, PySpark (MapReduce), Anaconda (jupyter notebook), MongodB, Android Studio, RTRT, Polyspace

• Support Vector Machine
• Neural Networks
• Random Decision Forests (Ensemble Methods)
• Linear and Logistic Regression
• Support Vector Regression
• Fast Fourier Transform
• K-Nearest Neighbors

Deep Learning

MapReduce

MongodB

Data Science with Python

Convolutional Neural Network for Visual Recognition

Machine Learning

MATLAB

• Spotlight on various tools for machine learning
• Tips and Tricks For Building ConvNet
• Support Vector Machine