Generative Artificial Intelligence (GenAI)

PhD: Variational Autoencoder and Diffusion Models for Plasmonic Design (Accelerated Design)

PhD, Chapter 9: Developed an AI-powered pipeline for plasmonic NFT design, generating 2000+/h optimized structures - automating a process that traditionally requires years of expertise. Achieved 20% higher power efficiency vs expert-designed structures, with results validated through advanced electromagnetic simulations.

Generative Adversarial Network (GAN )

ex: generating images and creating an art 

Trade-Off: Fidelity vs Accuracy

# unconditional generator from StyleGAN

np.random.RandomState(100)

batch_size = 4

truncation = 0.95 # condition: increase diversity/ decrease fidelity

# unconditional generator from StyleGAN


np.random.RandomState(100)

batch_size = 4

truncation = 0.05 # condition: decrease diversity/ increase fidelity

Text to image Generative Adversarial Network (GAN)

# this is an implementation of CLIP (Contrastive Language-Image Pre-Training)(paper)(2021) with Taming Transformers for High-Resolution Image Synthesis (paper)(2020)


alpha = 1 # the importance of the <include> input

include = ['A horse on the plane'] # place for your input


# Note: This illustrates the concept; its unpolished appearance is due to insufficient training.

Generating Magnetic Domain Pattern Images Using GANs

The aim is to extract parameters from images and/or generate new images of magnetization patterns using micromagnetic simulation data. 

Here, I show images of simulated magnetization patterns generated using Generative Adversarial Networks (GANs). This shows GANs' capability to generate magnetic domain pattern images.  

Midjourney AI (Illustrations for Neuromorphic Computing) 

I've been working with Modjourney AI to create some illustrations on neuromorphic computing. If you're interested, feel free to download them!