Stochastic Device

Stochastic devices, exhibiting inherent randomness in their behavior, are being actively investigated for their potential in building more efficient and robust neural networks. Current research focuses on developing accurate compact models of these devices, often employing machine learning techniques like Mixture Density Networks and Variational Autoencoders, to capture their stochastic dynamics and enable efficient simulations. This work aims to leverage the inherent randomness for probabilistic computing, improving the accuracy and uncertainty quantification of neural network outputs, and leading to more energy-efficient hardware implementations of artificial intelligence.