Quantum machine learning with superconducting circuits
summary
Experimental quantum neuromorphic system uses superconducting resonators on a single niobium chip cooled in a dilution refrigerator, reading photon numbers with a coupled qubit and FPGA. Using bosonic oscillators instead of qubits yields exponentially many basis “neurons”: one oscillator (states 0–4, time‑multiplexed to 20 neurons) classifies sine vs square with 99.7% accuracy, and a four‑oscillator device with trained internal coupling reaches 100% accuracy after ~200 epochs while reducing trainable parameters (442 vs 578). Training is hybrid classical/analog with analytic gradients; measurement overhead and short coherence times remain key limits.
AI-written summary · about
chapters (14)
- 0:00Quantum neuromorphic computing
- 1:10Neuromorphic computing inspiration
- 3:39Quantum physics for scalability
- 4:00Qubits and state space
- 5:26Quantum circuits for ML
- 8:01Qubit neural network limits
- 9:05Quantum oscillators as neurons
- 11:51Superconducting quantum hardware
- 13:18Quantum reservoir computing
- 16:15Quantum measurement challenges
- 17:37Training weights in quantum systems
- 19:48Quantum neuromorphic advantages
- 20:51Future beyond backpropagation
- 21:15Q&A
related talks
44mQuantum machine learning: Prospects and challenges
49mNew Computer Architectures : Explore Quantum Computers & SyNAPSE neuromorphic chips
53mThe quantum computers are coming
01h 08mPutting Quantum Supremacy on a Rigorous Footing
20mQuantum (Inspired) Computing at Fujitsu
31mQuantum neural networks
48mHow to Build a Quantum Supercomputer
54mThe Present and Possible Future of Quantum Computing