Quantized-memristor-enabled generative AI for predictive skin laser treatment imaging

초록

Generative artificial intelligence (AI) technologies can help to predict and visualize the results of skin laser treatments, improving patient consultations in cosmetic dermatology. However, existing generative adversarial networks (GANs) often struggle to produce diverse, realistic images due to issues such as mode collapse, primarily caused by low-entropy noise inputs from software-based pseudo-random number generators (PRNGs). Here, we introduce a conditional GAN (cGAN) framework empowered with a quantized-memristive true-random number generator (Q-mTRNG). The Q-mTRNG utilizes a Cu nanoscale filament formed inside a high-crosslinking-density polymer film to generate high-entropy, true random numbers via reliable quantum point contact. This enhanced entropy enables the cGAN to explore its latent space, generating diverse, realistic post-skin treatment images based on pre-treatment photographs.

키워드