VALIDATINGPOD 01: Regional NLP & Indic LLMs
Cross-Lingual Pseudo-Labeling for Regional Sentiment Analysis
Objective & Abstract
Quantifying label noise transfer when bootstrapping regional Kannada sentiment classifiers from English instruction-tuned models with calibrated temperature bounds.
Investigated Methodologies
- Confidence calibration
- Temperature scaling
- LoRA PEFT fine-tuning
Datasets & Benchmarking Corpora
Regional Reviews Dataset
IndicSentiment Baseline
All training and evaluation datasets undergo ethics audit and licensing compliance per AIAC Research Council standards.