Coder Social home page Coder Social logo
Aditya Kunar photo

adityakunar Goto Github PK

followers: 3.0 following: 0.0 repos: 14.0 gists: 1.0

Name: Aditya Kunar

Type: User

Company: Generatrix- A.I based data synthesizing platform

Bio: I recently obtained a master's degree in Data Science from Delft University. I am passionate about solving problems using deep learning techniques.

Location: Delft, Netherlands

Blog: adityakunar

  • 👋 Hi, I’m @adityakunar. A M.Sc Data Scientist from TU Delft

  • 🌱 I’m currently working towards bettering synthetic tabular data generation methods via GANs, read more here->Blog

  • 💞️ I’m looking to collaborate on investigating and evaluating privacy and memorization risks associated with training differential private GANs. Read more about my work on differential privacy here->Blog

  • 👀 I’m passionate about A.I, machine learning and deep learning. I've worked on the following projects:
    - Fast and Effective Image Classification using MobileNet+SSD: To achieve a fast and effective image classifier, studied the integration of the MobileNet architecture into the Single Shot Detector (SSD) framework in pytorch, check out our work here -> Blog
    - Processing GDelt Dataset in Apache Spark and AWS: Analysed how to optimally extract insights from the GDELT dataset using Apache Spark and AWS, read more here->Blog
    - A Reproduction of Dropout for Deep Neural Networks: For effectively preventing overfitting in deep neural networks, studied dropout as a regularizer for deep neural networks using pytorch, read more about our work-> Blog
    - Cyber Data Analytics: Executed effective class imbalance learning and anomoly detection for static and time-series data using python
    - Multivariate Data Analysis: Worked on MAP estimation for classifying healthy and diseases samples in ECG data, as well as assignments related to Bayesian regression, generalised linear models, Gibbs sampling and Gaussian process classification
    - Data Visualization in Tableau: Created visually appealing, user‑friendly dashboards to let end‑users explore open data about crime and policing in the UK easily and effectively
    - Research Methodology for Data Science: Worked on linear/logistic regression models and scientifically tested hypotheses in R; examined optimizing dimensionality reduction techniques like PCA for high dimensional datasets in MATLAB
    - Natural Language Processing: Built a fake news detector using python by implementing the techniques of the paper, "Emergent: a novel data-set for stance classification"

  • 📫 You may reach me at [email protected]

Aditya Kunar's Projects

computer-vision-seminar icon computer-vision-seminar

Studied the use of MobileNet as the primary backbone architecture in the SSD framework. Evaluated performance across multiple dimensions such as computational efficiency and model accuracy by carrying out experiments on multiple hardware settings and different sizes of objects. We crucially showed that the MobileNet architecture serves as an effective backbone that provides a decent balance between performance and efficiency especially when running only using CPUs and for objects of small sizes.

cyber-data-analytics icon cyber-data-analytics

Worked on the following topics:- Dealing with class imbalance using SMOTE while detecting credit fraud using machine learning models; anomaly detection for time-series using ARMA and PCA; detecting anomalies in network traffic data and hashing for counting occurrences of unique events in large streams of data; adversarial machine learning to deal with adversarially modified samples using dfgsm

deep-learning icon deep-learning

Examined the effects of dropout as a regularizer to prevent overfitting for deep neural networks. Our main aim was to reproduce the findings of the original paper. We found that our reproduction efforts couldn't lead us to similar conclusions reported in the paper and that more detailed investigations are needed to evaluate the effects of dropout on deep neural networks.

infovis-final-project icon infovis-final-project

Visualized open data about crime and policing in England, Wales and Northern Ireland to reveal patterns in criminal activity across the UK utilizing tableau. Created visually appealing, user‑friendly dashboards to let end‑users explore the data easily and effectively. Applied theoretical concepts of visualising geospatial, multi-dimensional categorical and continuous variables in the data.

multivariate-data-analysis icon multivariate-data-analysis

This repo contains my work on MAP estimation for classifying healthy and diseases samples in ECG data, as well as assignments related to Bayesian regression, generalised linear models, Gibbs sampling and Gaussian process classification.

natural-language-processing icon natural-language-processing

Implemented the techniques of the paper "Emergent: a novel data-set for stance classification" used for identifying whether news topics are true or false. In executing this project we learnt about how to extract useful feature representations from textual descriptions which allow for numeric representations of news headlines (e.g., word2vec). These features were then used for training machine learning models that are subsequently used for generating predictions.

research-methodology-for-data-science icon research-methodology-for-data-science

Explored statistical modelling using linear/logistic regression models and performed statistical hypothesis testing in R; learnt how to optimize the use of PCA for high dimensional datasets using Nyström approximation and snapshot methods and compared its use to other dimensionality reduction techniques such as MDS in MATLAB.

sdgym icon sdgym

Benchmarking synthetic data generation methods.

supercomputing-for-big-data icon supercomputing-for-big-data

Gained an understanding of central concepts related to efficiently using Apache Spark and the Amazon Web Services (AWS) in a practical context by analyzing a large open data set and identifying ways of processing it. The data set we used was the GDELT 2.0 Global Knowledge Graph (GKG), which indexes persons, organizations, companies, locations, themes, and even emotions from live news reports in print, broadcast and internet sources all over the world. We used this data to construct a histogram of the topics that are most popular on a given day giving us some interesting insights into the most important themes in recent history.

wikipassageqa icon wikipassageqa

Reproduction of "WikiPassageQA: A Benchmark Collection for Research on Non-factoid Answer Passage Retrieval" by Daniel Cohen, Liu Yang, and W. Croft (SIGIR18)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.