Multi-Tier Hierarchical Knowledge Transfer Learning: LLM-Powered Mother Machine Knowledge Distillation for Continual Learning on Resource-Constrained Child Machines MSc. Thesis

A novel hierarchical knowledge transfer learning framework for distributed intelligent systems, featuring a powerful mother machine equipped with an LLM that serves as a central knowledge repository across multiple domains.

WhisperWave: Synergizing Parameter Efficient Fine-Tuning using LoRA Technique in Transformer Models for Out-Of-Distribution Bangla Automated Speech Recognition (Accepted in IEEE CSDE23)

Fine-tuned pre-trained multilingual Whisper LargeV2 with Low-Rank Adaptation Parameter Efficient Fine-Tuning Technique on a low specification computer and compared the performance with other transformer-based models like wav2vec2 and conformer.

Flagging Sexism on Social Media Leveraging Hidden Layers of Transformers as Word Representations and Facilitating Model Interpret-ability through XAI Techniques (Accepted in IEEE CSDE23)

Calibrated different layers of pre-trained BERT-based transformer models to obtain higher scores on the SemEval-2023 Task 10 Dataset while explaining their insights using Lime and SHAP.

Machine Learning-Based Prediction of Rice Leaf Nutrient Contents Across Growth Stages Using UAV Data (Submitted to Journal Smart Agricultural Technology)

Analyzed multi-spectral images from UAV drones to calculate 39 different vegetation indexes and canopy reflectance from a completely new dataset sponsored by BRRI to predict Nitrogen, Phosphorus, Sulphur and Potassium Contents in Rice leaves at Different Growth Stages.