Utilizing Google Trends and past respiratory disease trends, we created a multi-ensemble LSTM model that leveraged state COVID-19 data in order to predict disease trends for other respiratory diseases such as RSV and Influenza.

Key results and takeaways is that with this new approach, we are able to create an accurate model of the disease trends despite having low data volume for those diseases when it comes to predictions. Other methods that we analyzed were unable to achieve similar results that our novel model has. This is potentially effective approach for future predictions in emerging and novel widespread respiratory diseases.

My own role included creating and training the individual LSTM models and the ensemble LSTM model in Keras and cleaning and creating usable data through pandas.

Created a physical smart orthotic knee brace device with various sensors that can help detect asymmetrical gaits through sensing and provide insights for those with mobility issues and helping with rehabilitation over time. 

Created a signal processing-based technique through data analysis and other exploration methods to detect asymmetrical gaits on device. Also did user testing to see the effectiveness of the device.

The project also provides a mobile application that can display the data to the user.

Key results and takeaways include discovering that we could detect overall mobility issues with asymmetrical gaits through one knee by detecting the gait phases utilizing gyroscope sensor and flex sensor data.

My role included building the model, sensor data analysis through pandas and scipy, data collection, and user testing. 

Created a novel compression algorithm with LSTM models for edge devices, such as smartphones, smart devices, etc, that can maintain its accuracy and improve the speed of model updates on the network. 

Modified a compression algorithm called Knowledge Augmentation and effectively applied it to the LSTM model with minimal loss. 

Key takeaways are that we found that compared to other known and state-of-the-art compression algorithms, it had the least amount of loss in accuracy and the smallest compression size. We also found that the algorithm could be applied to other RNN-based models. 

My role involved creating the LSTM model that predicted sentiment analysis as well as creating and applying our LSTM-based compression algorithm onto the model. I utilized PyTorch in Python for this project.

Created a robotic cat toy with Bluetooth controls and an autonomous mode for at-home cat entertainment. Has multiple features such as physical movement of the overall system, controllable cat toy wand, and a food dispenser to keep the cat engaged with the cat toy.

The robotic cat toy utilizes the Mbed OS and development board that is programmable with C/C++. In order to have multiple features work at the same time, this at-home cat entertainment utilizes multithreading to allow flexible functioning for the user to utilize. 

It also has a custom design chassis to help accommodate all included features and improve its durability. 

My role included the entirety of programming the embedded systems, which include the Bluetooth control, multithreading the multiple sensors (i.e. ultrasonic sensor) and motors that are required for it to work in C/C++.