Shrink Your Virtual Analog Model Neural Networks - Optimizing Architecture for Audio Processing

Watch a 16-minute conference talk from ADCxGather 2024 exploring innovative approaches to optimizing neural network architectures for virtual analog modeling. Learn how to move beyond trial-and-error methods in determining network size by examining geometric structures and symmetries in model complexity. Discover a systematic framework for creating efficient neural networks that accurately replicate analog circuit behaviors while minimizing computational resources. Gain insights into avoiding oversized networks that merely memorize training data, and understand how to develop architectures specifically tailored for function approximation in audio signal processing. Presented by Christopher Clarke, a PhD researcher specializing in low-latency audio processing and AI/ML technologies, this talk offers practical solutions for implementing efficient virtual analog models suitable for real-time applications.