<p dir="ltr">Brain oscillations are crucial for several cognitive functions, such as memory, attention, perception, and communication between brain regions. In particular, visual familiarity is known to induce oscillations in the theta frequency band (4-8 Hz). Today, most machine learning models lack this biologically inspired component. In the first aim of this study, we build a deep learning model that exhibits oscillations in the theta band using predictive coding theory. Predictive coding theory suggests that higher-order brain regions predict the activity of lower-order brain regions. The prediction errors are then propagated up the hierarchy and the predictions are updated. We were able to replicate the neural activities seen in a mouse training paradigm and observed that units of the neural network also exhibit oscillations in the theta band for a familiar stimulus. This offers the potential to bridge the gap between artificial neural networks and biological systems to build more biologically plausible neural networks. In the second objective of this study, we investigate the learning impairment of FragileX (FX) mice compared to Wild-Type (WT) mice when trained to discriminate Go and No-Go visual stimuli. Using recorded neural data, we build a classifier to perform stimuli and genotype classification. Both FX and WT mice achieved significantly higher accuracies in the theta oscillatory activity window compared to spontaneous activity in the V1 region. This suggests that the theta oscillations store essential information learned from visual discrimination. However, we observe that FX mice require more neuronal units to achieve performance similar to that of WT mice. In the hippocampus, the overall accuracy was lower for the classification of stimuli, indicating a more complex nature of information processing in the hippocampus. In addition, high classification accuracies for genotype decoding indicate that the neural responses to stimuli in V1 region are influenced by genetic variations. These differences are discernible even in individual trials when averaged across a few hundred neurons.</p><p dir="ltr">Classification using neural activities from Go trials yielded a higher accuracy than No-Go trials suggesting that these differences could be task-dependent and enhanced during certain tasks. We also observed that these variations are more notable during the theta oscillatory activity window demonstrating that genetic factors substantially influence memory encoding via theta oscillations.</p>

<div>There are millions of individuals in the world who currently experience limited mobility as a result of aging, stroke, injuries to the brain or spinal cord, and certain neurological diseases. Robotic Assistive Devices (RADs) have shown superiority in helping people with limited mobility by providing physical movement assistance. However, RADs currently existing on the market for people with limited mobility are still far from intelligent.</div><div><br></div><div>Learning control strategies are developed in this study to make a Cable-Driven Assistive Device (CDAD) intelligent in assisting a human joint (e.g., a knee joint, an ankle joint, or a wrist joint). CDADs are a type of RADs designed based on Cable-Driven Parallel Robots (CDPRs). A PID–FNN control strategy and DDPG-based strategies are proposed to allow a CDAD to learn physical human-robot interactions when controlling the pose of the human joint. Both pose-tracking and trajectory-tracking tasks are designed to evaluate the PID–FNN control strategy and the DDPG-based strategies through simulations. Simulations are conducted in the Gazebo simulator using an example CDAD with three degrees of freedom and four cables. Simulation results show that the proposed PID–FNN control strategy and DDPG-based strategies work in controlling a CDAD with proper learning.</div>

<p dir="ltr">Cardiac disease is one of the prominent reasons of deaths worldwide. The timely de-<br>tection of arrhythmias, one of the highly prevalent cardiac abnormalities, is very important<br>and promising for treatment. Electrocardiography (ECG) is well applied to probe the car-<br>diac dynamics, nevertheless, it is still challenging to robustly detect the arrhythmia with<br>automatic algorithms, especially when the noise may contaminate the signal to some extent.<br>In this research study, we have not only built and assessed different neural network models<br>to understand their capability in terms of ECE-based arrhythmia detection, but also com-<br>prehensively investigated the detection under different kinds of signal-to-noise ratio (SNR).<br>Both Long Short-Term Memory (LSTM) model and Multi-Layer Perception (MLP) model<br>have been developed in the study. Further, we have studied the necessity of fine-tuning<br>of the neural network models, which are pre-trained on other data and demonstrated that<br>it is very important to boost the performance when ECG is contaminated by noise. In<br>the experiments, the LSTM model achieves an accuracy of 99.0%, F1 score of 97.9%, and<br>high precision and recall, with the clean ECE signal. Further, in the high SNR scenario,<br>the LSTM maintains an attractive performance. With the low SNR scenario, though there<br>is some performance drop, the fine-tuning approach helps performance improvement criti-<br>cally. Overall, this study has built the neural network models, and investigated different<br>kinds of signal fidelity including clean, high-SNR, and low-SNR, towards robust arrhythmia<br>detection.</p>

<p dir="ltr">Understanding disease transmission is a complex and challenging task as it encompasses a wide range of intricate interactions involving pathogens, hosts, and the environment. Numerous factors, including genetics, behavior, immunity, social dynamics, and environmental conditions, contribute to the complexity. Furthermore, diseases exhibit significant variability in transmission patterns, including variations in the mode of transmission (e.g., respiratory, oral, touch-based, vector-borne), incubation period, and infectiousness. The dynamic nature of disease transmission compounds the existing challenges by introducing temporal variability and environmental variations, thereby intensifying the complexity of the study. Therefore, understanding disease transmission requires comprehensive research, integrated models, and a multidisciplinary approach to decipher the intricate web of interactions and factors involved. This dissertation aims to bridge the use and scalability gap between different levels of transmission models through the utilization of multi-paradigm modeling methods, incorporating varying levels of abstraction, to gain comprehensive insights into disease transmission. The first goal focuses on enhancing pandemic resiliency by analyzing the impact of varying parameters of heating, ventilation, and air conditioning (HVAC) on the dynamics of exhaled droplets and aerosols in the indoor environment using computational fluid dynamics (CFD) modeling. This goal operates at a micro-level of modeling, examining the detailed fluid dynamics and particle dispersion within indoor spaces. By simulating the movement of droplets under different HVAC configurations, this goal provides insights into the effectiveness of ventilation systems and optimizes parameter configurations in controlling disease transmission. The second goal of this dissertation is to aid organizations in evaluating potential policies to mitigate contact-caused risks in indoor spaces during a pandemic. This goal utilizes an ensemble of agent-based simulation (ABS) models, which operate at a higher level of abstraction. These models consider the behaviors and interactions of individuals within indoor environments, such as classrooms or meeting rooms, while incorporating physical distancing guidelines and seating policies. The third goal aims to improve pandemic prediction capabilities by developing a multivariate, spatiotemporal, deep-learning model that predicts COVID-19 hospitalization based on historical cases and evaluates the impact of state-level policy changes. This goal operates at the highest level of abstraction by utilizing deep learning techniques to analyze large-scale, publicly available data. The model captures temporal dependencies using long short-term memory (LSTM) networks and spatial dependencies using graph convolutional networks (GCN), graph attention networks (GAT), and graph transformer networks (GTN). By considering variables such as daily hospitalization and various policy changes, this approach provides a comprehensive framework for forecasting hospitalization cases and assessing policy impacts at the state level. This integrated, abstraction-based approach provides a more holistic understanding of disease transmission, allowing for the exploration of complex scenarios and the assessment of intervention impacts across different scales. This integrated architecture enables policymakers and public health professionals to develop targeted, effective strategies to mitigate the spread of diseases, allocate resources efficiently, and minimize the overall impact on public health. </p>