Umnoon Binta Ali

Data Scientist at Astha.IT. I graduated with distinction in CSE from North South University, where I specialized in Artificial Intelligence.

My academic journey led me to discover the exciting world of Machine Learning and Deep Learning, which I explored through data-centric projects.

I worked as a Research Assistant at North South University. Focus of my research was to analyzed the limitations of the Bangla Language Model and proposed solutions to improve its performance.

Email  /  CV  /  Bio  /  Github

I studied the most unbiased and augmented dataset NumtaDB to classify images of isolated Bangla numerals. I experimented with the data by creating data augmentation using keras library and trained a Convolutional Neural Network model and finally compared the result with the state-of-theart ResNet34 model.

In this research, transfer learning was explored for speaker diarization on our own Bangla audio data. the approach experiments with different clustering algorithms and embedding techniques to reduce Diarization Error Rate, DER for noisy data.

In this research, I presented a contrastive learning strategy to for enhancing the quality and semantic consistency of synthetic images by paraphrasing the captions from the CUB dataset.

Here I used Global Outbreak Alert and Response Network (GOARN) data to predict the time of the highest spread of covid-19. The time-series visualization is done with Power BI.

I created a recurrent neural network or `RNN` and train it on a tweet emotion dataset from hugging Face, to learn and recognize emotions in tweets. The dataset has thousands of tweets each classified in one of 6 emotions. This is a multi-class classification problem in the natural language processing domain.

I built a basic data science webapp with streamlit analyzing motor vehicle collision in NYC. I used the following libraries: numpy, pandas, pydeck and plotly