Jordan InturrisiMachine Learning | Electrical Engineering | Electronics Engineering
Portfolio
About
I am a Masters candidate at Deakin University where I do a mix of fundamental and applied research. My research focusing on deep learning, computer vision.
In the past I have applied deep learning to problems such as: predicting electricity consumption, predicting sporting outcomes, estimating the speed of a moving car based on dashcam video, as well as computer vision classification.
The Brownlow Medal is the AFL's highest individual honour and has been awarded since 1924 to the Fairest and Best player of each season. Voting is conducted by the field umpires immediately after each home-and-away match, with 3-votes awarded to the player believed to have been the best-on-ground, 2-votes for the second-best player and 1-vote for the third-best player on the day, in their opinion. Using data from 2500+ past matches with 50+ inputs, I trained a deep neural classifier to determine the probability that each player polls 3-, 2-, 1-, or 0-votes in each home-and-away match. The model's Top 5 ranked players account for 91% of all 3-votes, 80% of all 2-votes, and 66% of all 1-vote, whilst successfully predicting the player's receiving - 3-votes 57%, 2-votes 28%, 1-vote 20% of the time.
Consulting
Consulting services for Australia's largest renewable energy provider.
Forecasting electricity consumption
Forecasted total macro electricity demand at 30-minute and 24-hour prediction windows. Compared to the Australian Energy Market Operator (AEMO), reduced average error by 39%, standard deviation by 46%, and maximum error by 44% when predicting at 30-minute intervals. AEMO is responsible for Australia's largest gas & electricity markets and power systems. I further forecasted residential electricity demand also at 30-minute and 24-hour prediction windows. My model achieved 80-90% accuracy for 30-minute windows, 60-80% accuracy for 24-hour windows.
Computer Vision
Using popular computer vision benchmarks in MNIST, CIFAR-10, and CIFAR-100.
Project Work
Team member on a variety of projects including - Deakin Shell Eco-Marathon, Autonomous Ground Vehicle Competition (AGVC), and an industrial UAV research project.
comma.ai speed prediction
A coding challenge as part of the comma.ai interview process. Using the train.mp4 provided, I split the video into train (75%), validation (12.5%), and test (12.5%) sets. My model achieved >2 MSE on the test set, and >4.5 MSE on the validation set and >5 MSE on the test set.
