Résumé

Research Assistant

University of New South Wales

Build a benchmark for sentiment and sarcasm classification for under-represented dialects of English.

• Develop better data sampling strategies to make the existing datasets more challenging for LLMs.
• Design and execute experiments using all classes of LLMs, which involve in-context learning and parameter-efficient fine-tuning.

Casual Academic

University of New South Wales

Lead tutorials for the course–COMP9444: Deep Learning and Neural Networks.

• Mentor student groups through their end-to-end artificial intelligence projects.
• Achieve a broad agreement of 95% from feedback provided by 20 students.

Masters

University of New South Wales, Sydney

Information Technology

• Principles of Programming
• Data Science and Engineering
• Big Data Management
• Data Analytics for Graphs
• Computer Vision

Predicting the Target Word of Game-playing Conversations using a Low-Rank Dialect Adapter for Decoder Model

arXiv

Propose a novel adapter-based architecture to make pre-trained decoder models (Mistral and Gemma) robust to other dialects of English.

Evaluating Dialect Robustness of Language Models via Conversation Understanding

arXiv

Evaluate open-source (Llama, Mistral, and Gemma) and closed-source (GPT-3.5 and 4 Turbo) models on masked target word prediction of conversations from dialogue games.

Comparative Analysis of Abstractive Summarisation Techniques

Evaluate the effectiveness of static and context-aware embeddings in dialogue summarisation, using custom Seq2Seq, BART-base, and FLAN-T5 models.

• Large Language Models
• Dialogue Understanding