This presentation investigates the learning process of artificial intelligence by training a model to play the game of Go using an AlphaZero-type algorithm. Through evaluation of 12 Go models, the authors reveal the split personality many exhibit, much like the famous Schreiber book Sybil. The best models appear indistinguishable from human players in the early stages of the game before devolving into self-destructive tendencies in the endgame. Possible remedies for this behavior are explored through modifying training data generation, hyperparameter tuning, and optimizing neural network input dimensions.

In the dynamic environment of venues with large seating capacities, efficient management of seating occupancy emerges as a critical challenge. Traditional manual monitoring methods are often cumbersome and prone to inaccuracies, hindering optimal seat allocation and event management. Addressing this issue, our senior design project introduces an AI-based solution tailored to revolutionize real-time seating availability reporting for event organizers.
This project aims to provide a comprehensive tool that enables event organizers to track seating occupancy in real-time, facilitating the identification of peak attendance periods and enabling data-driven decision-making. By harnessing the power of artificial intelligence, our system offers a detailed analysis of seating patterns, thereby enhancing the efficiency of event operations and optimizing resource allocation. The ultimate goal is to improve the event experience for both organizers and attendees by ensuring a seamless flow of information regarding seating availability, leading to more effective management of large-scale events. Through this initiative, we endeavor to set a new standard in venue management, where technology and data converge to create smarter, more responsive event environments.

Efficient teacher-student interaction analysis is essential for educators to enhance teaching quality. Traditional manual review methods are excessively time-consuming and can yield subpar feedback. ClassifAI offers a streamlined solution for educators to gain insights without sacrificing work hours, utilizing the OpenAI Whisper model for transcription and a fine-tuned Gemma model for question categorization.

ClassifAI is advancing existing tools by addressing four key improvements: transitioning to local hosting for cost savings and data security, integrating the WhisperX model for improved transcription accuracy, automating Costa's Three Levels of Thinking question classification via Google's Gemma, and upgrading the web interface for better user experience.

ClassifAI's architecture comprises a user-friendly web server with ExpressJS and React, a local MongoDB database, a fine-tuned Gemma model for question categorization, and WhisperX for speech-to-text. ClassifAI offers speech recognition, diarization, question categorization, and analysis, delivering enhanced performance. Educators easily upload their teaching audio/video on our platform via a file or YouTube, which is then processed by our GPU server for transcription and analysis. The resulting transcript, graphs, and metrics are accessible for review and can be exported in various formats.

Speech impairment ranks among the world's most prevalent disabilities, affecting over 430 million adults. Despite its widespread impact, many existing video-conferencing applications lack a comprehensive end-to-end solution for this challenge. In response, we present a holistic approach to translate American Sign Language to subtitles in real time by leveraging advancements in Google Mediapipe, Transformer models, and web technologies. In March 2024, Google released the largest dataset for the problem domain with over 180 GB in size, containing ASL gesture sequences represented as Mediapipe numeric values. Our methodology begins with the implementation and training of a Transformer model using preprocessed Google dataset, followed by the establishment of a back-end server equipped with the trained model. This server handles video input preprocessing and real-time inference, communicating with client services as a REST endpoint. To demonstrate the practicality of our approach, we developed a video conferencing application utilizing the AgoraRTC SDK, which communicates with our back-end server to transcribe user gestures to text in real time, displaying them on the receiving end. Through this end-to-end system, we enable video calls enhanced by the real-time transcription of fingerspelled gestures with low latency and high accuracy, effectively bridging the communication gap for individuals with speech disabilities. With a growing imperative for AI applications engineered for human well-being, our project seeks to promote the integration of AI in applications designed to enhance human wellness, thus bringing the broader awareness and adoption of this endeavor.

Wood is a fundamental material in various industries, from construction to furniture making. Understanding its mechanical behavior is crucial for optimizing its use and ensuring structural integrity. This study investigates six different wood types under flexural loading, offering insights into their performance in real-world applications. By analyzing key parameters such as density, flexural strength, and stiffness, this research aims to provide valuable data for informed material selection and design optimization. The wood types under scrutiny comprise white oak, birch, bamboo, maple, pine, and walnut with two contrasting grain configurations.

Key parameters: Density, Flexural Strength, Flexural Stiffness