I present here a sample of my projects and creations. Contact me on linkedin.
Exploring some cutting edge machine learning models and physical applications.
Diffusion models:
Quantum information:
At the University of Innsbruck: Institute for Theoretical Physics. Code available in Repository
Abstract: Machine learning has experienced an uninterrupted growth over the last few years. Advancements in research and computation hardware allow the application of neural networks in a diverse spectrum. Motivated by remarkable results in generating indistinguishable fake images of humans and landscapes (e.g., Deep-fakes) physical problems are getting more attention. In this thesis, we look at generative adversarial networks (GAN) which are able to learn the hidden distribution of a training set sampled from a physical system. Based on conditional GANs we show the Ising model can be learned and reproduced at unseen thermodynamic control parameters. We propose the SpinGAN architecture that catches the phase transition of the 2D Ising model with great accuracy. The deterministic neural network allows us to evolve single distinct Ising states through the temperature range. Utilizing the exact differentiation of the network we present a new GAN fidelity to identify critical temperatures without knowledge of the physical system.
Keywords: GANs, deep learning, Monte-Carlo, Ising-model
Custom raytracing render engine (using NVIDIA OPTIX™) for displaying photo realistic real time simulations.
In cooperation with Swarovski.
Abstract: Some crystal glassware of Swarovski is marked by means of laser engraving. The quality control of the engravings is currently done manually by random sampling. This manual quality assurance does not meet the Swarovski quality standards any more. In addition, it does not prevent the delivery of inferior goods to customers. The manual verification of the engravings should be replaced with an automated image processing system. The design of a concept and development is part of the project. The intention of the project is a finished prototype which fully automatically verifies the engraving of a manually inserted crystal using image processing software. In this way, a detection of inferior crystals according to clearly defined quality standards is possible. The software is able to get images from the camera and execute analyzing algorithms.
Keywords: image processing,
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
