About Me

Ph.D. Candidate in Physics
University of California, Santa Barbara

I am a physicist using machine learning to (i) emulate cosmological simulations, (ii) improve parameter estimation methods, and (iii) create foundational models for science.

As a part of the ENIGMA group (under the supervision of Dr. Joseph F. Hennawi), my research focuses on combining statistics, machine learning, and high-performance computing to enable, improve, and accelerate parameter estimation of cosmological models, particularly for studying the Epoch of Cosmic Reionization (EoR) using the Lyman-alpha forest. I am a recipient of the NASA FINESST Fellowship, which supports my research on inference pipelines for the EoR.

I am also currently a Research Scientist Intern at Lawrence Berkeley National Lab, working within the Computational Cosmology Center (Scientific Data Division) on creating a foundational machine learning model for hydrodynamical cosmological simulations.

Research Projects

My research spans multiple areas at the intersection of physics, machine learning, and high-performance computing.

Foundational Machine Learning Model

Creating a dataset of cosmological hydrodynamical simulations to model the effects of varying the reionization history on the Lyman-alpha forest, executed on the Frontier computer at ORNL and the Perlmutter computer at NERSC. Using Vector Quantization Generative Adversarial Networks (VQ-GANs) to tokenize multidimensional outputs from simulations and exploring the feasibility of using transformers to create a foundational machine learning model for the Lyman-alpha forest.

Simulation Based Inference

Created a dataset of mock observations of the power spectrum of the Lyman-alpha forest using the output from a hydrodynamical simulation and a stochastic forward model. Trained and tested the Balanced Neural Ratio Estimation (BNRE) algorithm on the dataset. Performed coverage tests and demonstrated the effectiveness of likelihood-free inference methods in improving the validity of parameter inference in the study of the Epoch of Reionization.

Neural Emulators and Hamiltonian Monte Carlo

Created a dataset of cosmological simulations to model the effects of varying the reionization history on the Lyman-alpha forest. Used deep neural networks to create surrogates for the mean power spectrum and covariance matrices of the Lyman-alpha forest. Used Hamiltonian Monte Carlo to further accelerate inference.

Radiative Transfer Simulations

Modified a radiative transfer simulation code to model the complex circumstellar structure around MWC-297, a high-luminosity young stellar object. Developed pipeline that creates synthetic visualizations of direct imaging from the simulation outputs.

Publications & Presentations

A selection of my recent work in machine learning for cosmology and astrophysics.

Publications

  • Reliable Parameter Inference for the Epoch of Reionization using Balanced Neural Ratio Estimation D. Gonzalez-Hernandez, M. Wolfson, J. F. Hennawi
    Accepted at the Machine Learning and the Physical Sciences Workshop (NeurIPS, 2025)
  • Using Neural Emulators and Hamiltonian Monte Carlo to constrain the Epoch of Reionization's History with the Lya Forest Power Spectrum D. Gonzalez-Hernandez, C. Doughty, M. Wolfson, J. F. Hennawi, J. Zhenyu
    Submitted to the Monthly Notices of the Royal Astronomical Society for review (MNRAS, 2025)
  • Neural Network Emulator to constrain the High-z IGM Thermal State from the Lyman-a Forest Flux Autocorrelation Function J. Zhenyu, M. Wolfson, J. F. Hennawi, D. Gonzalez-Hernandez
    Published in the Monthly Notices of the Royal Astronomical Society (MNRAS, 2025)
  • Accelerating Statistical Inference in Astrophysics with Neural Networks and Hamiltonian Monte Carlo D. Gonzalez-Hernandez, M. Wolfson, J. F. Hennawi
    Accepted at the Structured Probabilistic Inference & Generative Modeling Workshop (ICML, 2024)

Recent Presentations

  • Neural Simulation Based Inference for Physics - Lawrence Berkeley National Lab, High Energy Physics group (October 2025)
  • Using Neural Emulators to Accelerate Inference in Cosmology - UCSB AI CoP Spring Symposium (May 2025)
  • Emulating the EoR's History Dependent Power Spectra of the Lyman-a Forest - UCSB AstroLunch Seminar (April 2025)
  • Applying BNRE to better constrain the Epoch of Cosmic Reionization - SBI Hackathon, Tübingen, Germany (March 2025)

Contact

Feel free to reach out if you're interested in collaborating, have questions about my research, or would like to chat.