School of IT students crush AI competition sponsored by Deakin Community Bank
A focus on enhancing discipline-specific skills in computer vision have seen School of Information Technology students competing to develop the best algorithmic models in the 2021 Deakin Simpsons Computer Vision Competition. The competition focused student engagement activities on practical skills development in a fun experience, with the goal of strengthening graduate employment outcomes.
Developed by Dr Mohamed Reda Bouadjenek and supported by Professor Peter Eklund and Dr Sunil Aryal, this sophisticated competition invited undergraduate and postgraduate IT students to develop and submit an artificial intelligence (AI) model using machine learning and deep learning to recognise individual Simpsons characters. Students were supplied with a dataset of 19,548 labelled images to ‘train’ their model and were encouraged to tune their hyperparameters to tweak accuracy. Students were allowed to work on their own or collaborate in teams with the competition’s aim to improve students’ AI and computer vision modelling, problem solving and critical thinking skills.
Chosen based on a thorough technical review and the highest accuracy scores, the top teams/individuals were awarded prizes by the competition sponsor. Student Ngoc Hung Huynh came in first place, followed by Peter Allen (second place), Bevan Fairleigh (third place) and Team Doh! Data (Josh Anthony, Roger Middenway and Melissa Mony) in fourth place.
Dr Mohamed Reda Bouadjenek is a lecturer in applied artificial intelligence, in the School of Information Technology.
‘I was really impressed by some students who have demonstrated a competitive spirit and a lot of creativity in developing innovative computer vision models during the 2021 Deakin Simpsons Challenge,’ Dr Bouadjenek explains.
‘This goes well beyond the scope of the technical AI material we teach in our units. For instance, some students have used and combined technically advanced AI models to achieve high accuracy performance, which shows that our students have pushed their boundaries toward goals of excellence, have learned how to be fully independent since we did not offer technical support to them, and have demonstrated to be ready to work in a professional environment.
‘This clearly demonstrates the quality of our pedagogy and our courses within the School of Information Technology. This is in a way our trademark at Deakin University.’
This highly successful student engagement experience had a total of 76 participants with 386 submissions. Student submissions were evaluated for their accuracy with students allowed to continually refine their models until the end of the competition, with some submitting their models upwards of 40+ times.
Competition was fierce throughout the trimester. A competitive leader board would dynamically update per new submission, with the lead sometimes changing multiple times each week and helping to promote friendly competition. Watch students’ progress through the leader board visualisation.
Professor Peter Eklund is the Associate Head of School, Development of Deakin’s School of IT.
‘The Simpson’s character recognition competition was the crowning achievement for student engagement during the pandemic’, Professor Peter Eklund said.
‘Reda, Sunil and I were so pleased that not only postgraduate students, specifically trained in deep machine learning, but also undergraduate students were willing and able to compete.’
Tom FitzGerald is the Chair of the Community Bank at Deakin University, the sponsor of this competition.
‘Our Community Bank – Deakin University Strengthening Potential Fund was established to support important initiatives like this one, which benefits our students, staff, alumni and wider Deakin University community. We are proud to support this initiative and encourage other students, researchers and members of the Deakin community to engage with the Fund,’ Mr FitzGerald said.