This project presents a Monte Carlo Tree Search model for optimizing guessing strategies in Wordle, achieving a 99.0% success rate with an average of 3.96 guesses, surpassing the baseline entropy-based method's 98.6% success rate and 4.06 guesses. Further refinements will focus on enhancing decision-making efficiency and adapting the model to broader applications.
This paper presents a Language Model for predicting patent approval outcomes, achieving a 64.37% accuracy, surpassing the current state-of-the-art model's 57.96%. Despite this improvement, further enhancements are needed to overcome performance plateaus. Future efforts will focus on refining model capabilities and addressing new challenges.
This study improves rural road development in Mozambique by predicting travel speeds using high-resolution satellite imagery and GPS data, surpassing conventional assessments. The methodology provides valuable insights for targeting road improvements and enhancing agricultural productivity in impoverished areas.
This study explores lip reading, the transcription of text from visual information alone. We developed three models: a 3D-CNN for detecting speech with 77.63% accuracy, a CNN-LSTM for predicting spoken words with 52.19% accuracy, and an Encoder-Decoder Transformer for entire sentence prediction, which needs further improvement. The results show promise in advancing lip reading technology, though sentence prediction remains challenging.
This project enhances sports player tracking by integrating Model-Agnostic Meta-Learning with the OSNet model within the Deep-ExpansionIoU framework. This novel approach improves tracking precision and efficiency, achieving higher accuracy compared to the state-of-the-art Deep-EIoU model, particularly in complex sports scenarios. The results demonstrate notable improvements in key tracking metrics, setting a new benchmark in sports analytics.
This study uses machine learning to enhance Yelp's user experience by predicting restaurant star ratings from reviews, recommending restaurants based on user ratings, and leveraging OpenAI's text embeddings for improved recommendations and semantic search. The approach aims to streamline the process of finding optimal matches on the platform.
"I See You" (ICU) is an object detection program focusing on basketball players within the boundary of the court. After evaluating multiple models, Yolov8 stood out as the best with a 96% accuracy rate. This project is the first step in a larger initiative aimed at enhancing the scouting process for basketball coaches.