Develop strategies to enhance NLI models' ability to understand lexical relationships, especially between antonyms and contradictions.

• Utilized data augmentation to enrich training data with antonyms and synonyms, improving lexical reasoning but affecting generalization.
• Employed adversarial training with the ANLI dataset, significantly boosting performance on lexical reasoning and generalization.

• Hugging Face (Transformers)

Design an AI agent for ice hockey within the SuperTuxKart ice hockey game using image-based strategies.

• Generated training data from simulated matches to train a Fully Convolutional Network (FCN) for puck detection.
• Developed two hand-tuned controller strategies: a simple controller for basic gameplay and an advanced controller utilizing complex tactics based on the puck's estimated 3-D position.

• Pytorch

Estimate home sale prices and provide real estate investment recommendations using machine learning and financial models.

• Forecasted future home prices using ARIMA, Prophet, and LSTM neural network models, focusing on the Austin, TX, housing market.
• Applied the Capital Asset Pricing Model (CAPM) to evaluate investment risks and returns.

• Machine Learning Models: Decision Trees, Neural Networks for home value estimation.
• Forecasting Models: ARIMA, Prophet, and LSTM for future price prediction.
• Financial Model: CAPM for investment evaluation.