Brandon Pratt's Projects
Code dev for classifying fly behavior
Contained is the code used to analyze the interlimb coordination and walking kinematics of various Drosophila species
Code used for analysis or visualization in the manuscript entitled, "Snow flies self-amputate freezing limbs to sustain behavior at sub-zero temperatures"
In this final project, two others and I characterized the differences in efficiency between simple RNN, LSTM, and GRU neural networks in accurately predicting words given a sequence of preceding letters. We determined that GRU neural network generalizes better than the simple RNN and LSTM neural networks in predicting different target sequences. Moreover, the GRU neural network required less training epochs for predicting the sequence "hello world" perfectly as compared to the simple RNN and LSTM neural networks. However, the simple RNN and LSTM neural networks could predict the target sequence perfectly with fewer recurrent units than the GRU neural network. Overall, these results may provide insights into how the nervous system optimally encodes sequences of items.
This contains the code used to acquire video data, control the treadmill speed, and analyze fly walking
Lecture content for UW Software Engineering for Data Scientists
something
The goal of the final project for STAT 535 at UW was to conduct multi-class classification of the CIFAR10 dataset. I used a convolutional neural network with augmentation to do this and achieved a classification accuracy of 73% on the test set.
Supporting code for Pratt et al., 2024
in class exercise for CSE583 Autumn 2019