linkin github github
profile card

Hi, πŸ‘‹πŸΌ

I'm Aditya codeSTACKr.com

Machine Learning PhD Researcher


I am a PhD student at the University of Florida. I completed my Masters in Computer Science from University of Florida in 2019.

My research interests include applying Machine Learning for acoustics and image segmentation. Currently, I am researching on emotion recognition from human speech and music. I like to write blogs on Machine Learning in my spare time. Check out my blog on Medium: @adityadutt.medium.com.



SKILLS   βš™οΈβ›πŸŒŸ



Languages : Python, SQL, MATLAB, Elixir, C/ C++, HTML, CSS

Tools/ Libraries : Keras, TensorFlow, Tensor Board, Pandas, Geopandas, NumPy,
Scikit-Learn, OpenCV, Matplotlib, Plotly, PyQt, Seaborn, Spacy, NLTK, Beautiful Soup, Jupyter Notebook, QGIS

Research Papers πŸŽ“πŸ’‘

  • Kumar, Sushil, et al. "Colour image segmentation with histogram and homogeneity histogram difference using evolutionary algorithms." International Journal of Machine Learning and Cybernetics 9.1 (2018): 163-183.
  • Aditya Dutt, Richa Dutt, Hema N. (2017). Development of Cost-Effective Substitute of Braille Tactile Display Operated by Linear Actuators: IEEE Xplore.

Academic Research πŸŽ“πŸ’‘

    Speech Emotion Recognition   Jan 2020 - Present
  • β€’ Researching on detecting different emotions from speech by analyzing spectrograms as a visual representation speech features such as the pause between each word, pitch, loudness, and more.
  • β€’ Testing an analytical blend of words with aggregate factors (pauses, pitch, etc.) to more accurately do sentiment analysis including which may alert to security breaches or health issues.
    Lexicon-Based Segmentation of Offline Cursive Handwriting   Jan 2019 - Dec 2020
  • Worked on a project to build a machine learning model that can recognize cursive handwriting, which is a more difficult task than recognizing printed characters. The objective was to develop algorithm that can understand handwriting and accurately convert handwritten text to printed documents. The application is useful for reading postal addresses, bank check amounts, and forms.
  • β€’ Implemented a combination of connected component analysis and distance transform to segment (or oversegment) word into possible characters. The small segments which are not characters can be used as a good dataset to test outlier detection algorithms.
  • β€’ Developed a robust peak detection algorithm to detect peaks pointed in all directions in binary images which correctly splits a word into possible characters. Used a dynamic programming algorithm and CNN to find a match score between each string in the lexicon (dictionary) and the segmented parts of word image.

Experience πŸ‘” πŸ’Ό

    Graduate Research Assistant
    ESSIE Department, University of Florida,  Jan 2021 - Present
  • βš™οΈ The goal of the project is to build a deep learning framework to translate different streams of data from individual sensor into a shared manifold space.
  • βš™οΈ Different sensors require different analysis models tuned for a particular task. Using this framework, embeddings from all frameworks can be mapped into a shared common space.
  • βš™οΈ Now, a single analysis model can be used which is sensor independent. It will help in saving development costs for separate analysis models for every sensor.
    Graduate Research Assistant
    IFAS, University of Florida,  Jan 2020 - Dec 2020
  • βš™οΈ Lead the construction of a custom SQL database to gather and analyze data from India’s National Sample Survey Office. Data is comprised of housing conditions, employment and agricultural assessment surveys of India.
  • βš™οΈ Implementing autoencoder and regression methods to estimate indicators of poverty and food security across economically depressed regions of India by analyzing regional patterns of land cover change.
  • βš™οΈ Predict future years socio-economic features using census data and poverty indicators.
    Graduate Teaching Assistant
    Computer Science Department, University of Florida,  Aug 2019 - Nov 2019
  • Course: Distributed Operating Systems
  • βš™οΈ Created course assignments that were approved by the supervising professor. Managed one Teaching Assistant and one Course Grader. Held weekly office hours, graded coursework, and championed student development.
  • βš™οΈ Collaborated in creating projects (and grading system) based on Twitter implementation and gossip algorithms in Elixir using the actor model to simulate the dissemination of information across a network using Honeycomb and 3D Torus.
  • βš™οΈ Received an end-of-semester evaluation score of β€œOutstanding” based on student feedback.
    IOT Intern
    IIT Delhi, India   Jun 2016 - Jul 2016
  • βš™οΈ Tasked with the challenge of controlling the rpm of a DC motor clock so that it synchronized with data from a GPS using Beagle Bone.
  • βš™οΈ Created a webpage using HTML, CSS, and JavaScript to monitor live data from microcontroller to detect and resolve time and clock errors.
  • βš™οΈ Gathered data from Arduino and Wi-Fi module (ESP8266) and stored it in the SQL database. Next, pushed data from database on the webpage allowing microcontroller hardware to be controlled remotely.

Projects πŸ‘¨πŸ½β€πŸ’»



Music Genre Classification
Used LSTM and CNN to classify music into progessive and non-progressive rock music.
Github
Object Tracking using SIFT Algorithm
Demonstration of sift algorithm to track objects and observing the effect of each parameter on performance.
Github
Bitcoin Simulator
Implementation of Bitcoin protocol to simulate bitcoin mining, wallet and transactions.
Github
Image segmentation Methods Demo
Demonstration of a few useful segmentation algorithms like Canny Edge Detector, Snakes, Active Contour and KNN.
Github
Gossip Algorithm
Implementation of gossip protocols for information dissemination in a network with different kinds of topologies.
Github
Cursive Handwriting Segmentation and Recognition Library
A Cursive Handwriting Segmentation and recognition Library. Multiple models are trained using Transfer Learning, One Shot Learning, CNN, and MLP. Click 'Demo' to see the model used for recognizing cursive chaarcters.
Github Demo
Cancer Cell Segmentation using H&E Slides
The aim of project was to segment cells in H&E slides. The inconsistent background, different contrast in different slides make the segmentation really challenging. I implemented combinations of different algorithm like thresholding methods, active contours, and U-Nets.
Demo
Bird Species Classification using birdcalls and songs
Built a deep learning model to classify 9 bird species by songs. Spectrograms were used as features. A Siamese network along with 1D dilated convolutions and Batch normalization layers were used. Accuracy on test set was around 97%.
Github Medium Article

Achievements πŸŽ–πŸŽ–

    βš™οΈ Robotics Competition: Robometry   IIT Roorkee Saharanpur Campus, 2014
  • Secued 1st Place. Our team built an autonomous line follower robot which completed all the tasks successfully.    
    βš™οΈ Autonomous Robotics Competition: Nightwing, 2014
  • Secued 3rd Place. The goal was to finish the race while following the line. Our team built an autonomous line follower robot.    

Volunteer

    βš™οΈ Organizing member of UCR Robotics club in JIIT, Noida, 2015
  • β€’ Designed the autonomous robotics competition track for line follower and managed track designing.    
  • β€’ Created event rules and scoring scheme.    
  • UCR Club JIIT