How does Szymanski's work apply to AI today?
My work directly applies to training large-scale AI models, which are essentially parallel and distributed systems. The trade-offs I formalized—between computation, memory, and communication—are critical for scaling deep learning across thousands of GPUs. For example, my communication-avoiding algorithms reduce the bandwidth bottleneck in distributed gradient descent, speeding up training. In graph neural networks, my scalable graph processing methods enable learning on networks with billions of nodes, such as social media or biological interaction graphs. Additionally, my cybersecurity research informs adversarial robustness in AI, where detecting anomalous patterns in distributed systems parallels detecting attacks on neural networks.
Ask Boleslaw Szymanski the follow-up →