Developed a zero-shot learning model using dynamic ensemble of LLMs and contrastive learning, achieving a 4.78% improvement in intent categorization accuracy, outperforming state-of-the-art models on multiple datasets in ambiguous setting environment.

This study evaluates LLMs like Llama-3 and Mistral as autonomous research generators, achieving a 27% boost in creativity and 42% improvement in user satisfaction, setting new benchmarks for ideation in research.

This study explores the use of synthetic plant images from generative models to improve rare species identification, achieving a 29% boost in zero-shot and 31% in few-shot learning performance, setting new benchmarks in botanical classification.

Empath-Obscura is an ensemble model for emotion detection in obfuscated faces, combining YOLO V5/V8 with Poster++ and a novel SPIGA-based augmentation technique, achieving a 13.18% performance improvement on low-quality facial images.

This study introduces IMPEARL, a reinforcement learning agent for optimizing home appliance scheduling, achieving a 37.5% reduction in billing costs and a 66.5% increase in user satisfaction by considering uncertainties and user preferences.

This study extends IMPEARL strategy to a multi-agent setting for smart grid scheduling, balancing user satisfaction in adversarial and cooperative environments, and achieving improved cost reduction and efficiency.

Analysis of predictive capabilities of different time-series models, including ARIMA and RNNs, for NIFTY 500 and S&P 500 indices, efficient RNNs outperform traditional methods, especially in capturing trends and the causality between global markets.