Dr Nagender Aneja

Dr Nagender Aneja

Research Projects

Published in School of Digital Science, 2022

This post covers projects available for FYP/Research students. Students may also contact with their project proposals.

Projects Ongoing

  • Adversarial Machine Learning

    • Aim: To provide defense against adversarial attacks on machine learning

    • Duration: Jan 2022 – Dec 2024
    • Funding: UBD FIC Research Grant ~ BND 32K
    • Collaborators: SDS and FIT
    • Output: One Scopus-indexed publication
      • Defense against adversarial attacks on deep convolutional neural networks through nonlocal denoising, paper

  • Medical Imaging using Deep Learning

    • Aim: Early detection of diseases using Deep Learning
    • Duration: Ongoing
    • Demonstration: http://s3.sds.ubd.edu.bn
    • Output: Two Scopus-indexed Publications
      • Proposed methodology for Early Detection of Lung Cancer with low-dose CT Scan using Machine Learning, paper
      • Transfer learning for cancer diagnosis in histopathological images, paper
      • Semi-Supervised Learning for Cancer Detection of Lymph Node Metastases, paper

  • Fake News Detection using Machine Learning

    • Aim: To detect fake news based on linguistic features using machine learning
    • Duration: 2019 - 2021
    • Output: Two Scopus-indexed publications
      • Predictive linguistic cues for fake news: a societal artificial intelligence problem, paper
      • Detecting Fake News with Machine Learning, paper

FYP for BDSc

  • Adversarial Machine Learning
    • The project’s goal is to investigate various vulnerabilities in Deep Learning Algorithms and examine currently published solutions in order to minimize the threats. Specifically, the research will investigate the vulnerabilities of Convolutional Neural Networks in image categorization, which has immediate implications in face recognition technologies, health applications, and self-driving or autonomous cars, among other areas. We will develop adversarial examples for robustness in image classification training. MNIST, CIFAR-10, and CIFAR-100 datasets will be used in this study.
    • Skills: Python, PyTorch
  • Text Classification
    • Text Classification is used to classify customer feedback, emails for spam filter, sentiment analysis of social media posts for a topic. In this project we will implement DistilBERT, https://arxiv.org/pdf/1910.01108.pdf which is smaller in size and efficient but gives performance comparable to BERT. We will use two datasets one that is publicaly avaialble dataset and other downloaded from twitter.
  • Image Classification with Vision Transformers
    • To accomplish this task, we will divide a picture into patches and provide the sequence of linear embeddings of these patches as input to a Transformer. This model will be trained in a supervised manner using image classification datasets MNIST, CIFAR-10, and CIFAR-100. A comparison between vision transformer models and convolutional neural networks models will be made as well. We will implement the paper “an image is worth 16x16 words: transformers for image recognition at scale”, https://openreview.net/pdf?id=YicbFdNTTy
    • Skills: Python, PyTorch
  • Model Compression
    • Deep learning approaches depend on over-parametrized models that are difficult to implement. Model compression strategies that are both efficient and accurate are needed in order to decrease memory, energy, and hardware usage while maintaining accuracy. Model compression will aid in the deployment of lightweight models on the device and the protection of privacy by conducting private on-device computation. To examine the differences between over-parametrized and under-parametrized networks, the technique of pruning is being applied. This research will look at model pruning for popular CNN architectures, which will be examined in depth.
    • Skill: Python, PyTorch
  • DCGAN
    • We will train a generative adversarial network (GAN) to generate new celebrities after showing it pictures of many real celebrities. DCGAN was proposed by “Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks”, https://arxiv.org/pdf/1511.06434.pdf
    • Skills: Python, PyTorch
  • Melanoma Classification
    • The objective of the project is to train a machine learning model that can identify skin cancer in particular Melanoma. The project also includes deploying the model as a web app.
      • Dataset: https://www.kaggle.com/c/siim-isic-melanoma-classification/data [Any other similar dataset may be used for a similar medical imaging task]
      • Skills: Python, PyTorch
  • Conversational AI - Chatbot