Bachelor in Computer Science and Engineering, University of Dhaka, Dhaka, Bangladesh. [Jan 2017 to Dec 2021]

Attended Courses:

Undergraduate Research: "Attention-Based Neural Network for Machine Translation"

During my undergraduate studies, I led a research project focused on developing an augmented neural network architecture using two stacked GRUs and an attention-based model to improve English-to-Bangla sentence translation. Our model achieved near-human level performance and was benchmarked using BLEU scores:

• 10.48 for 10k sentences with 50 epochs
• 17.68 for 45k sentences with 75 epochs

These results surpassed several previous models, including GRU with Hierarchical Back Translation (14.42), Phrase-based SMT + large LM (15.27), Attention-based NMT (15.57), and Transformer (18.12).

I curated datasets (including SUPara, the largest available Bengali-English parallel corpora), implemented and trained the model on Google Colab GPU, and evaluated the results using attention plots and BLEU score performance metrics.

Terms: Transformer Model, Gated Recurrent Unit(GRU), LSTM, Word Vector, Tensorflow, RNN