As technology advances and society pivots to more renewable sources of energy, the need for a better way to store and deliver power is increasing. New materials are essential to improve the way we store energy. Finding the right material from the many possible combinations of composition and orientation is a lengthy process. Experimentally, it would take far too long to try and create these materials. Computational methods like density functional theory are faster than making the material, but are nonetheless too inefficient to offer a practical solution to fully search the space of potential materials. We seek to use information about known materials to train a machine learning model to help find materials with promising dielectric properties. We develop classifiers to initially categorize our unknown materials as metallic or insulating and stable or unstable. We also use regression to predict the dielectric constant and band gap of materials. With this information, we can create a more refined list of materials with the potential to serve as dielectrics in supercapacitors.

This research explores the role of emojis in sentiment analysis by leveraging TensorFlow and a dataset of 1,500 Reddit comments that include emojis. While traditional sentiment analysis models focus exclusively on text, this study aims to enhance a pretrained text-based sentiment analysis model to incorporate emoji data. The initial step involves employing a pretrained model from Hugging Face, which is adept at analyzing sentiment from textual data alone. Subsequently, the model will be fine-tuned using a specialized dataset to enable it to interpret and analyze the sentiments expressed through emojis in conjunction with text. This fine-tuning process involves training the model on the combined dataset, where emojis are integral to understanding the sentiment conveyed. The research will compare the performance of the fine-tuned model against the baseline model to determine the efficacy of including emoji data in sentiment analysis. By investigating how emojis influence sentiment detection, this study aims to provide deeper insights into the nuanced ways people express emotions in digital communication. The findings could enhance the accuracy of sentiment analysis tools and contribute to more comprehensive understanding of online interactions, where emojis play a crucial role in conveying sentiments that text alone might not fully capture.

Understanding large-scale structures is crucial to cosmology as it provides insights to structure formation and composition. Weak lensing, a phenomenon where gravity bends light around an object, causes the projected matter density of the Universe, denoted as kappa, to act as a lens that distorts the images of background galaxies (Bartelmann & Maturi, 2016). These distortions can be used to map out the matter density of the universe and gain insight into large scale structures. To achieve this, a statistical description of the matter density field, which can be approximated using the log-normal model. In this work, we characterize the cosmological dependence of the shift parameter in the log-normal approximation using numerical simulations and build an emulator to predict the shift parameter of the log-normal distribution at different distances as a function of cosmology. This emulator provides simulation tools for future weak lensing studies and possible insight to large scale structures.

Soficity is an important property of groups that allow for loosely approximating them with permutations of finite symmetric groups. There has yet to be a group found which fails to be sofic. The Higman group, popularized by its lack of finite quotients, has remained unknown if it is sofic. Our aim is to investigate the existence of a function that locally behaves like a modular exponent which has 4-cycles for almost all of its elements in order to prove or disprove the soficity of the Higman group. This presentation discusses soficity, highlights why the Higman group is unknown to be sofic, and explores the implications of finding such function using both dynamics and computation.