BarraCUDA: Bringing Electromagnetic Side Channel Into Play to Steal the Weights of Neural Networks from NVIDIA GPUs.
Péter HorváthLukasz ChmielewskiLeo WeissbartLejla BatinaYuval YaromPublished in: CoRR (2023)
Keyphrases
- neural network
- graphics processing units
- graphics hardware
- general purpose
- graphics processors
- gpu implementation
- artificial neural networks
- weighted sum
- fuzzy logic
- back propagation
- neural nets
- weight update
- hidden neurons
- parallel programming
- parallel processing
- pattern recognition
- linear combination
- genetic algorithm
- connection weights
- power analysis
- training process
- smart card
- parallel implementation
- relative importance
- countermeasures
- high frequency
- weighting scheme
- game playing
- fuzzy systems
- recurrent neural networks
- multi layer perceptron
- parallel architectures
- simulation software
- computer systems
- real time
- hidden units
- fault diagnosis
- feed forward
- fuzzy neural network
- network architecture
- multi layer
- neural network model
- multilayer perceptron