There are more than 1 million restaurants in the U.S. employing about 10 % of the U.S. workforce. Given the large number of restaurants and how often they fail, I created a model that can provide insights to investors and lenders and help them evaluate restaurants based on whether they are likely to close within the next 4 years. For more details read the relevant blogpost where I describe the process I followed to build this model and talk about my results. My code can be found on github.
An automated way of makeup removal would be useful to law enforcement or could be used as an added feature for dating apps. Here, I trained a convolutional neural net with an autoencoder architecture to remove lipstick from images of caucasian women. For more details check my github page.
Many of the microscopy images taken for scientific purposes (especially images of polymer particles) are not as complicated as real-life images (i.e. there is a presence of only two materials and a good contrast). The common edge detection algorithms (such as Canny) are "too good" for some scientific applications which means that the development of a simpler faster algorithm should be possible. I have developed an algorithm that creates a classification edge-detection dataset using the Canny edge detector. This dataset can then be used as input to train a machine learning algorithm to identify the edges of particles on similar images. This way the extraction of data from a large number of microscopy images becomes more feasible.
See an implementation example of this algorithm here that yielded 98 % pixel classification accuracy using linear logistic regression and only 0.5 % of the information contained in one image. The number of required basic operations was cut in half compared to the Canny algorithm.
I am currently working under the Princeton University incubator program to build "Chow Fleet". The company is delivering food from local restaurants that do not deliver to the Princeton campus in a lean experiment built to identify a scalable and repeatable business model for food delivery in low population density areas.