Cartilage is a basic foundation that stabilizes the integrity of the human body with its unique structure that withstands significant stress from the body. The loss of cartilage due to health disorders has been a prevalent health issue with no suitable substitute for cartilage. The high water concentration, mechanical properties, and biocompatibility provide the hydrogel with an advantage in replacing lost cartilage. This paper looks at the unique structures of hydrogels prepared via microphase separation, which allows for the polymer components to bond through hydrogen bonding; leaving behind a highly porous structure containing mainly water but maintaining high mechanical strength. The objective is to analyze two methods of microphase separation and determine the best method in creating a hydrogel similar to the mechanical properties of cartilage. Cononsolovency is a method of decreasing or eliminating the solubility of a macromolecule by mixing two solvents; the first method for this paper works similarly but using only one solvent, DMSO. UV polymerization then prompts microphase separation to create a rigid but porous structure within the hydrogel that enhances both its mechanical properties and swelling rate for effective water transport. The second method uses the Hofmeister effect where hydrogels that underwent a freeze-thaw cycle are submerged in a sodium solution to induce microphase separation. The hydrogels using the Hofmeister effect have a modulus between 1 - 15 MPa matching that of biological cartilage (0.3 - 20 MPa) concluding that this method creates optimal hydrogels for cartilage replacement.

Question answering (QA) on spontaneous speech performs worse when the speech signal originates from an underrepresented dialect, such as African-American Language (AAL). We experimented with various methods ranging from prompt engineering to fine-tuning. However, major problems arose with prompt engineering, such as generating irrelevant or false information and ignoring given prompts. Therefore, we adopted fine-tuning as a more effective and reliable method. To decrease transcription errors that negatively impact information retrieval by QA models, we use the Corpus of Regional African American Language (CORAAL) database to train and evaluate Whisper. We then input the transcript generated by Whisper to the QA model DeBERTa, and compare its performance with large language models (LLMs) such as LLaMA 2. By utilizing Meta’s LLaMA 2 and fine-tuning Open AI’s Whisper ASR system, preliminary results show improved QA performance compared to conventional question answering models.

Since the inception of biosensors, two design principles—conventional highly conductive brittle materials, and soft and compliant materials—have reigned supreme. The former has mechanical mismatch with tissues, and the latter doesn't have electrical properties comparable to conventional materials. Recent developments have created a third type of biosensor, combining the existing designs. The design choice of the Interconnected & Integrated Bioelectronics Lab consists of a stretchable, highly conductive, and strain-insensitive three-layer model. The layers consist of gold, which is brittle and conductive, anisotropic conductive film (ACF) containing silver particles that allow vertical electrical conduction, and silver paste. When stretched, the gold cracks and forms islands that require at least one ACF silver particle underneath to maintain electrical conductivity. A Crack Propagation Model, using Python, was developed to improve the crack mechanism, by controlling the optimal propagation paths. Additionally, an Image Processing Model was needed to analyze microscopic photos of ACF and locate the coordinates of the silver, which were then used as the input in the Crack Propagation Model. These models enhanced the design of soft and strain-insensitive biosensors and will be applied to predict optimal drug dosing using machine learning algorithm. The Auto-Regressive Integrated Moving Average (ARIMA) model is a popular time series forecasting model used to predict future values based on past data patterns. Through training the ARIMA model with prerecorded datasets, the pharmacokinetic curve can be predicted in early stages after drug injection. This biosensor will contribute to the advancement of biotechnology and accurate drug personalization.

Given the substantial variance in eye anatomy among individuals, conventional eye-tracking calibration often requires personalized approaches. Yet, there is no implicit calibration method to achieve accurate calibration without explicit user involvement. Since eye-tracking technology plays a pivotal role in numerous fields, ranging from understanding human behavior and user experience to medical research and assistive technology, we aim to improve the previous software by fine-tuning eye-tracking systems without the need for explicit user involvement of specific calibration tasks and decreasing the calibration drift. We hypothesize that it is possible to calibrate eye-tracking software using implicit methods effectively. To achieve this conclusion, we utilized saliency mapping to identify the object of focus in order to track the eye’s attention accurately. We also incorporated the Starburst Method for strategically placing visual cues on the screen, enabling accurate gaze prediction. Moreover, a sophisticated machine learning model was implemented to predict both eye movement trajectories and the corresponding probability distribution of the viewer’s focus within the image. The creation of the software was completed with only slight errors in machine learning where the software would crash. With additional time to debug the code, the software will run in the future on different types of computers.