Title : Decoding the Quantum Frontier: AI-Driven Error Correction
Date: August 2 2026 Sunday 14:00 - 16:00 EDT
Abstract:
The path to fault-tolerant quantum computing is fundamentally gated by our ability to perform real-time Quantum Error Correction (QEC). While surface codes provide a robust geometric framework for topological protection, the classical control layer currently faces a scaling crisis. Traditional decoding algorithms, such as Minimum-Weight Perfect Matching (MWPM), are increasingly insufficient for large-scale lattices, struggling with both the exponential growth of computational complexity and the nuanced, correlated noise patterns inherent in modern hardware.
In this talk, Dr. Bagherzadeh and Samira will present a paradigm shift in the QEC stack: replacing rigid classical decoders with adaptive, AI-driven neural architectures. We demonstrate how the 2D lattice geometry of surface codes can be effectively treated as a dynamic "image" problem, allowing for the application of Vision Transformer (ViT) and sequence-based Transformer models to decode syndrome signals. By leveraging self-attention mechanisms, these neural decoders can identify non-local error correlations that traditional algorithms miss, significantly improving logical error rates at scale.
Beyond the algorithmic advantages, we explore the engineering realities of implementing these models within the cryogenic control loop. We discuss techniques such as model distillation, quantization, and FPGA-based co-processor integration, aimed at achieving the sub-microsecond latency required for real-time error correction. We conclude by framing the future of quantum computing not merely as a quest for more physical qubits, but as an optimization challenge for the classical "classical brain" that keeps those qubits alive. Attendees will gain an understanding of how integrating LLM-inspired architectures into the quantum stack is the essential, missing component for transitioning from noisy intermed