A Deep Learning model that takes as input an image created by averaging two random samples from CIFAR-10 dataset and is tasked with predicting the categories of the two components using Convolutional Neural Network.

A pipeline to detect and remove biased reasoning steps of Chain-of-Thought in reasoning LLMs to reducing stereotype propagation without compromising accuracy, benchmarked model performance across bias score and accuracy on BBQ (Bias Benchmark for QA) and MBBQ (Multilingual BBQ) datasets.

This project focuses on generating high-quality synthetic data for machine learning applications using state-of-the-art language models for Docker commands. The project leverages Llama-3.1-405b for content generation and Nemotron-4-340b-reward for quality assessment.

Fine-tune mDeBERTaV3-base model for 5 different languages in 3 different settings (Multilingual, Monolingual, and zero-shot) for a binary classification task in order to detect the Subjectivity or Objectivity of the news sentences a part of this challenge. Our results placed us among the top 5 teams in almost all languages.

A graph analysis of the top AI/ML papers in the passed year