Circulating tumor cells, also known as CTC, are found in the blood of patients with cancer that has left primary or metastatic tumors and entered the peripheral blood. Throughout the years, scientists have been trying to identify the presence of CTC cells to improve patient care, such as enabling early cancer detection, determining patient prognosis, and directing longitudinal treatments. In addition, CTC count has also been seen to be an important prognostic factor for patients diagnosed with metastatic cancer. Many efforts have been made to develop better technologies that identify, characterize and isolate the extremely rare subpopulation of CTCs. However, identifying CTCs has proven to be a challenging problem as there are low concentrations of CTCs existing in a patient's peripheral blood; as such, there are approximately 1-10 CTCs found in 1 billion blood cells. Currently, the traditional way of counting CTCs is via a device called the labyrinth (made up of PDMS), immunostaining then imaging is very costly and time-consuming. This research aims to present a method that will be an accurate and easy solution to identifying CTCs. The approach for identifying CTCs uses a machine-learning algorithm to classify cells detected in microscopic images of patient blood samples, which will contain both CTCs and white blood cells. Through experiments, the machine learning model can accurately classify and count CTCs without needing immunostaining and tedious labor processes. After more patient data is collected in the upcoming months, the machine learning model can be further improved.

Information privacy is an essential need for every person. As we begin improving technology and gaining information, securing it has become exponentially more of a challenging task. With the increased use of the Internet of Things (IoT), i.e., the network of interconnected devices, objects, and people, protecting users’ information privacy is becoming even more critical. This research focuses on detecting and categorizing botnet attacks in IoT systems using machine learning. A botnet attack is a large-scale network attack that may come in the form of malware infected IoT devices or malware carried out in a simple spam email. Our objective here is to utilize supervised machine learning algorithms to aid in the detection and categorization of botnet attacks. The targeted botnet attacks in this research are Scan, Theft, Denial of Service, and Distributed Denial of Service. Detecting and categorizing such attacks can lead to more efficient ways of defense against them and, thus, better secured IoT systems.

As computing applications become increasingly complex and widespread, the demand for powerful and efficient computer designs has significantly increased. However, simply increasing raw processor speed renders major diminishing returns. The need to implement heterogeneous computing techniques – that is, utilizing specialized hardware that is optimized for the application at hand – is apparent. By using multiple computer architectures and hardware accelerators such as scalar processors, vector processors, and/or domain specific systems on a chip (DSSoCs) in a computer system, we can achieve performance gains beyond those that are possible with raw increases in processor speed. While execution time and power consumption characteristics are improved, this comes at the cost of a greater design complexity, requiring additional effort in hardware accelerator integration, resource management, and application development. These issues are addressed at the pre-silicon design stage in the Compiler-Integrated Extensible DSSoC Runtime (CEDR) framework. CEDR provides/combines many design features that help a hardware system designer conduct the cost-benefit analysis for different hardware implementations. The present study focuses on utilizing CEDR to implement multiple computer applications, such as a computer vision lane detection algorithm and a 5G protocol stack. This allows for analyzing how various hardware configurations affect the performance of the application. Power consumption, execution time, and scheduling characteristics are taken into consideration during the pre-silicon design stage to determine a cost-benefit analysis of implementing DSSoC hardware accelerators for these applications.

Abraham Ochoa July 6th, 2022 Assignment: UROC Abstract Sustainable Propulsion Systems Abstract Currently the aviation industry is responsible for 3% of global carbon dioxide emissions. This is expected to increase by 4-5% per year. With the increasing climate regulations, the aviation industry needs to lean towards an aircraft propulsion system that is both efficient and sustainable. To address this, an investigation into the comparison of pure-electric to conventional propulsion systems is presented. A pure-electric propulsion system offers the potential for a decrease in emissions. However, current technological capabilities decrease the viability of these propulsion systems. A mass-energy balance has been conducted comparing a conventional Cessna 172 and a pure-electric Cessna 172. The results found that the maximum range with the current battery specific energy density of 200 Wh/kg is 285 kilometers (154 nautical miles). In order to reach this maximum range, a pure-electric Cessna 172 would have a maximum take off weight of 67,134 kg. A conventional Cessna 172 has a maximum takeoff weight of 1,157 kg and a maximum range of 1,185 kilometers (640 nautical miles) with 53 gallons of aviation fuel. For a pure-electric Cessna 172 to match the performance of a conventional Cessna 172, the battery specific energy density would have to be increased to 1,164 Wh/kg. These results are discussed and compared to previous work. Findings indicate that with current technology a pure-electric propulsion system is most viable for small aircrafts whose mission profile consists of a short distance travel.