Our team built a tool that allows users to run, compare, and analyze three different types of machine learning models: centralized, federated, and personalized. The tool’s web app allows a user to select the number of devices to be used, the number of clusters (groups of devices that contain similar data), and the type of machine learning model to train. After configuring the machine learning task on the web app, this information is sent to a server device and several client devices as specified by the user.

We explored the use of machine learning to verify properties of data structures. We aimed to understand the learnability of data structure properties using off-the-shelf machine learning models and potentially increase efficiency of conventional software testing with the trained models. This project included generating a dataset of graphs with varying properties, building machine learning models to be trained and tested with these graphs, and exporting the trained models into a JUnit test suite. In the end, we were able to show that certain graph properties are easily

The application is utilized by people and companies looking   to convert from less environmentally friendly energy   sources to solar power. The product that was created is   meant to provide a way for these prospective solar panel   users to find out whether or not installing a solar panel at a   given location is viable and to help them optimize the location to select for installation.

Nanotechnology has taken engineering to new scales, but how has time resolution changed with these scales? Testing in nanoscale has recently become much more time resolvable through the use of high-speed high-resolution data acquisition systems. Our project uses the National Instruments PX1e-1071 data acquisition system to significantly speed up measurements taken with a scanning probe microscope. To start, a scanning probe microscope moves a probe across the surface of a test chip to map its topological or electrical characteristics.

Created robot that can deliver parts autonomously in a factory.

Students are tasked to design and build a low-cost system capable of measuring the permittivity (dielectric constant) and loss tangent of radio frequency materials across the 1-40GHz range. Students will investigate a range of different approaches, with an eye towards cost, spectral bandwidth, and ease of measurement. They will downselect to a preferred approach and then build the system. They will be expected to develop calibration techniques and data acquisition techniques, as well as processes for extracting the desired data from the measured quantities.

We have developed a system to monitor and control water usage in residential homes. Our focus was centered around encouraging water conservation by giving users the ability to view their regular water usage and restrict their water flow as desired. Our system works by installing a node, which includes a valve and sensor, to the user’s desired appliance(s) in the home. The node is then connected to an Android application where it displays the usage data and gives the user control of the water flow through each node.

Many of the wearable devices today aim at tracking physical activities such as step count and distance walked. With our project, we aim at expanding the information that is tracked to include social interactions. With these social interactions, we aim at reporting the social life of individuals to determine the types of conversations individuals are having. We developed this device using machine learning and audio capture on an embedded system to fit on an individual’s wrist.

The Wireless Sleep Monitor (WSM) is a non-contact device used for measuring a user's heart rate and respiration rate over a period of sleep. Using a millimeter-wave radar on the XWR1843 EVM Board from Texas Instruments, the WSM collects signals from body displacement during sleep and records the processed data. Heart rate and respiration rate data is tracked and stored for user display through the WSM web application.

Stool and urine can say a lot about a person’s health. Our project, Loggo, is a toilet-mounted Raspberry Pi with a camera, microphone, and pressure-sensitive mat. The device periodically photographs the contents of the toilet bowl and collects audio samples. We use a neural network based on ResNet50 to classify each image according to the type of stool in the bowl, if any. We also use another model to categorize the audio as urine, a flush, or silence.