In this postdoc, we plan to focus on computer vision tasks where existing deep learning methods require lots of labeled samples to work well. Acquiring labeled samples is time-consuming and often impractical. Thus, we investigate three different classes of methods to alleviate the label scarcity problem: active learning, weakly-supervised learning, and few-shot learning. In active learning, the goal is to label the most important samples to maximize the performance of the model while reducing labeling costs. In weakly supervised learning, the goal is to train models using weak labels.
There is currently an abundance of research in the community in terms of capturing X-ray data around COVID-19 analysis to help diagnosis for radiologists. However, at the same time there are staffing shortages that lead either to long diagnosis wait times and potential misdiagnosis. Our research is focused around rapid deployment of AI and developing of a system to deploy AI technologies and algorithms using existing infrastructure within the hospital.
With respect to large-area display applications, it is desirable to have not only the active layers but also the electrodes in the OLEDs that can be formed by solution fabrication process. To address the manufacturing challenges of high-performance OLEDs, several scalable techniques such as doctor blading, ink-jet printing, and ultrasonic spray coating have been developed or employed.
Fourien Inc. is developing a diagnostic medical instrument for rapid and low-cost detection of COVID-19 virus using saliva samples from early-stage infections. The custom-built vibrometer instrument will use micro-sensors to detect the RNA of the virus. There is a need to develop optimized and sensitive modules of optics, electronics and mechanics that are critical for the instrument to work. The intern will use her prior academic experience to bring creative solutions to industry problems.
In light of the COVID-19 pandemic, there has been an increase in sexual assault incidents. Research shows that when there are disasters and economic meltdowns, there is usually a spike in sexual assaults. The world is experiencing both a pandemic and economic downturn. With self-isolation, social distancing, and the stay-at-home orders, the use of technology to provide support to survivors is now more critical than ever before. However, the use of technology could be a double-edged sword. Using technology to support survivors could sometimes increase the risk for survivors.
A study of Covid-19 patients revealed that about five percent of patients with the worst effects of the infection require a ventilator to push air into the lungs, take over the body’s breathing process, and offer the best chance of survival. To have a precise control, this research program aims to develop a Power electronics drivers for high-speed motors to control pressure loop precisely for maximum performance of ventilation capacities.
Many of the current greenhouse cultivation processes can be labor intensive, unable to accurately capture all information on a plant, and hard to manage as grower’s operation scale. By proposing a new method of collecting and analyzing data in these greenhouse using computer vision and machine learning, interns will try to improve the efficiencies of these processes. This proposed system aims to collect valuable information such as plant dimensions and fruit sizes that was previously very inefficient for human labour to do, and to predicts most optimal growing environment with this data.
The research described in this project proposes a very realistic and Quantum Computing resilient key distribution networking protocol to enable highly secure and efficient information flow. On one hand, the protocol provides unconditional one-time-pad based encryption; on the other hand, it is based on information-theoretic concepts that can be cost-effectively implemented today to establish Quantum Key Infrastructure (QKI) that somewhat resembles already existing Public Key Infrastructure (PKI). The PKI to a large extent could be used to enable QKI.
Bell Media receives content from different providers, including content it produces in-house. There are standards for tagging audio tracks with metadata however many facilities (including Bell) do not adhere to these standards. Currently Bell uses a manual approach to classify unlabeled audio tracks, which is inefficient, and time consuming for massive digital media that Bell has and receives. Bell is developing a single ingest pipeline to accelerate the labeling and processing of media files it receives.