M.S. in Machine Learning: from particle simulations to real-world solutions
Jonny Keane '23, '24, a data scientist who holds a B.S. in Computer Science and M.S. in Machine Learning from MSOE, explored the potential of GPU programming to create video simulations for an independent study in his graduate program. He focused on the simulation of a million particles, each with simple behaviors like following chemical trails or making random decisions when no trail exists.
These particles collectively form intricate, web-like structures, and their behavior culminates in a final shape. For this project, Keane chose to have them form the shape of the MSOE logo. At first glance, the animation might seem like little more than a static logo with minor movements, but Keane’s work delves much deeper. By harnessing the power of Rosie (MSOE’s NVIDIA GPU-powered supercomputer), he simulated millions of particles simultaneously, creating complex and ornate patterns that emerge from these simple behaviors.
Unlike traditional CPU-based programming, which handles tasks one at a time, GPUs can manage thousands of calculations at once, allowing for simulations on a much larger scale. This is relevant to real-world applications like deep learning and AI, where millions of small operations, when executed simultaneously, can lead to extraordinary results. One example of how this type of GPU programming is applied is in simulating realistic worlds, like hyper-realistic video games.
Keane likens the particles in his simulation to cells in the human body or individual operations in a deep learning algorithm, each contributing to a much larger system. This parallel processing approach could be applied in various industries, including healthcare, where simulating complex biological systems like the brain would be vastly more efficient with GPU power.
One of the key takeaways from his education, Keane says, is the importance of thinking about problems at scale. In a world where computing power is distributed across multiple systems, understanding how to architect solutions that can handle large datasets and complex tasks is essential. This is something Keane brings to his work every day, whether he’s tackling exploratory projects or improving existing technologies.
In his role as a data scientist at Direct Supply, Keane applies the skills he honed through his undergraduate and graduate studies to solve real-world problems. Working on a variety of projects—from improving search engine algorithms to exploring how AI can revolutionize senior living—Keane’s problem-solving mindset and deep technical knowledge allow him to break down complex issues and develop impactful solutions.
Keane’s work doesn’t just focus on coding and data manipulation. It also involves critical thinking about how to balance accuracy and value in AI systems. For example, with AI-driven solutions, he must constantly evaluate how to handle mistakes and communicate the limitations of the technology, ensuring that customers still derive value even if the system isn’t 100% perfect.
Looking back at his time as a student, he appreciates how his experiences using cutting-edge tools like Rosie the Supercomputer—even as an undergrad—set him apart from others in his field.