A Radiation-Hardened, Fault-Aware Quantized Neural Network That Self-Heals

Key points

• Deep learning (DL) models are increasingly being deployed into radiation-challenging environments in spacecraft, particle accelerators, and, soon, space-based data centers.1
• Instantiated as a learned orchestration of numbers stored in persistent memory, models are susceptible to continual degradation as high-energy particles corrupt data.1
• Our work extends the well-established Fault-Aware Training (FAT) paradigm by adding an alignment penalty to concentrate undamaged activations and weights into a low-rank subspace, and we then reconstruct flagged corrupted coordinates from the uncorrupted survivors.1
• This introduces a new path towards designing resilient data-driven models, complementing established hardware redundancy, ECC, and FAT-based techniques.1

Sources (1)

• [1]A Radiation-Hardened, Fault-Aware Quantized Neural Network That Self-Heals
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 04:38 AM

  In this work we developed a runtime method for repairing the quantized weights of neural networks (NNs) after bit flips on commodity, commercial off-the-shelf (COTS) hardware, without storing extra replicas.

  Deep learning (DL) models are increasingly being deployed into radiation-challenging environments in spacecraft, particle accelerators, and, soon, space-based data centers.

Extractive summary: sentences quoted from the sources.