Pivot-Assisted Consensus for Enhancing LLMs in Low Resource Languages • Developed Pivot-Assisted Consensus (PAC), an innovative method improving Large Language Models (LLMs) for Low Resource Languages (LRLs), addressing linguistic data scarcity and enhancing task accuracy. • Conducted pivotal research integrating cultural and linguistic insights, significantly advancing NLP capabilities and model inclusivity. • Engineered Efficient prompts and Adhered to a zero-shot learning framework, ensuring the model's adaptability and effectiveness without requiring language-specific training. • Executed comprehensive experiments utilizing an array of datasets like MGSM, XQUAD, & XNLI focusing on improvements in translation accuracy, sentiment analysis, and reasoning tasks evidencing the success of the our approach in elevating LLM efficiency. (Paper Currently Under-Review) Tech Stack: Python, TensorFlow, Hugging Face Transformers, Pandas, NumPy, NLTK, Git, Jupyter Notebooks.

더보기

• Spearheaded the personalization segment of a pioneering stress detection study, enhancing model accuracy by integrating user-specific data and personality traits. • Developed a 'Similar Model User' strategy, employing K-means clustering based on the Big Five personality traits and XGBoost models, optimizing for user-specific stress detection. • Advanced the project with an innovative multitask learning approach using Gradient Boost-ing Tree Multitask Learning, significantly increasing AUC to 54.1% and establishing a new benchmark in the field. • Contributed to the broader project goals by setting reproducible baselines and advocating for the importance of large,

더보기