Welcome to todays Meeting

Deep Dream Group

Agenda

  1. Text Classification Tasks!
  2. Possible ICCIT!
  3. Fair Data Generation Project!
  4. Explainable Audio Classification!

Text Classification Task

  1. Finished part of Introduction
  2. Finished abstract
  3. Finished all the experiments
  4. Need to write all the remaining parts

Fair Bengali Handwritten Character Generation

Presented by: Md Fahim Sikder

Deep Dream Group

Motivation

  1. Data Fairness is needed!
  2. Trustworthy AI!
  3. Not much research has been done in Bengali!

Research Problem

  1. How can we create an Image dataset for Bengali domain?
  2. How we can generate fair generative models?

Data Fairness

  1. Real world data is full of bias
    1. COMPAS case in the USA

Data Fairness

Bangla Sensitive Dataset

Bangla Sensitive Dataset

Generative Models

  1. GAN-based
  2. Diffusion-based

Possible Contribution

  1. Bangla Sensitive datasets
  2. Generative model
  3. Interpretability / Explainability

See zotero library.

Explainable Audio Classification

Idea

  1. Train an audio classifier (preferably cnn/transformers based)
  2. Explain the results
  3. Compare different explainable techniques

Possible Contribution

  1. Unique model
  2. Explainability

Datasets

  1. https://www.kaggle.com/datasets/ejlok1/toronto-emotional-speech-set-tess
  2. https://github.com/Jakobovski/free-spoken-digit-dataset/tree/master

See zotero library.