About Joshua Logan Shunk
Joshua developed a new type of regularization technique, which is a kind of guide for an algorithm. He found that this regularization technique allowed neural networks that had been trained to recognize images to do so faster and with better accuracy than existing methods, even while using less data.
Neuron-Specific Dropout: A Deterministic Regularization Technique To Prevent Neural Networks From Overfitting and Reduce Dependence on Large Training SamplesView Poster
Joshua Logan Shunk, 18, of Gilbert, improved training of neural networks for his Regeneron Science Talent Search computer science project. During the COVID-19 pandemic, Joshua made a computer program that could look at chest x-rays and predict whether patients had COVID-19 or pneumonia. Joshua found that sometimes while training his program to find patterns, it would find some that were not really there. For example, it associated wider chests with pneumonia. This led Joshua to study “regularizers,” which are tools that gently guide algorithms to look for the right kinds of patterns. For his project, Joshua developed a variant of an existing regularizer, which he calls “neuron-specific dropout.” Joshua’s variant works especially well on networks that find patterns in images, allowing them to train faster and find patterns with greater accuracy while using less data.
Joshua attends Perry High School where he is the team captain of the Varsity Division I ice hockey team. He has won two all-academic awards and an all-state honor through the Arizona High School Hockey Association. The son of Susan and Benjamin Shunk, Joshua is planning to study computer science and cognitive neuroscience.
Beyond the Project
Joshua won a prestigious award at the 2022 Regeneron ISEF, which sent him to Sweden to attend the Nobel Prize Ceremony.
FUN FACTS: During the COVID-19 shutdowns, Joshua trained for a half-Ironman competition (a 1.2-mile swim, 56-mile bike and 13.1-mile run), during which he burned 2,500 extra calories per day.