PDF version: PhilTravis_cv_2025_v6.pdf (91 kB)

 

Education

  • Ph.D., plasma physics, 2025, University of California, Los Angeles
  • Master of Science, Physics, 2018, University of California, Los Angeles
  • Bachelor of Science, Physics, 2017, University of Illinois at Urbana-Champaign

 

Projects

Optimization of ICF simulations using JAX (2025 - present)

  • Goals: learn the fundamentals of simulating plasmas; optimize ICF plasma instabilities
  • Utilizing the ADEPT differentiable simulation framework
  • Optimized beam parameters to reduce growth rate of the two-stream instability

Evidence of interchange modes in LAPD mirrors (2024 - 2025)

  • Goal: destabilize the curvature-induced interchange instability in the LAPD
  • Achieved β ~ 1% — an order of magnitude higher than previous LAPD mirror studies
  • Analysis included Thomson scattering, Langmuir probes, a fast framing camera, and other diagnostics
  • Observed low-m, low-frequency, high-amplitude mode consistent with the interchange instability

Reconstructing diagnostic signals using (generative) energy-based models (2021 - 2025)

  • Goal: learn correlations between diagnostics and machine state information via generative modeling
  • Trained (generative) energy-based models (EBMs) using transformer-like inputs heads and CNNs
  • Reconstructed missing diagnostics from the learned distribution via conditional sampling

Inferring trends in the LAPD mirrors using machine learning (2024 - 2025)

  • Goal: infer trends from a model trained on randomly varied LAPD actuator states
  • Built neural network model to predict ion saturation current given a mirror machine actuator state
  • Inferred trends in discharge current, mirror configurations, and gas puffing durations
  • Used model to infer state required to optimize axial variation in a mirror cell
  • Made public the code, models, data, poster, and writeup: doi:10.5281/zenodo.15007853

Upgrading LAPD diagnostic pipelines for machine learning applications (2021 - 2023)

  • Goal: collect as much data as possible from every shot on the Large Plasma Device for ML purposes
  • Constructed a new LabVIEW- and python-based system that ties into the existing acquisition system
  • Aggregated machine state information and auxiliary diagnostics for every discharge
  • Recorded over 29 million shots — largest known magnetized plasma dataset by shot count

Gradient-based optimization of a 0d mirror machine reactor (2022 - 2023)

  • In collaboration with individuals at WHAM. Funder: ORFEAS competition
  • Goal: develop and optimize a 0D mirror machine reactor model for minimal cost
  • Built a 0d mirror machine model in SymPy and JAX
  • Optimized estimated dollar cost, Q, and other cost functions with respect to particular machine and physics parameters, including stability constraints
  • Demonstrated superiority of the tandem mirror approach over simple mirrors

Automating Langmuir sweep analysis using neural networks (2019 - 2020)

  • Goal: automate the analysis of noisy or otherwise non-ideal Langmuir probe sweeps
  • Aggregated Langmuir sweep data and built an NN-, attention-based fitting routine in TensorFlow
  • Built a surrogate model for the theoretical Langmuir sweep curve
  • Fitting routine was robust to strong drift wave turbulence

Study of turbulence and transport in magnetic mirror geometries in the LAPD (2017 - 2019)

  • Goal: explore and determine the effect of a magnetic mirror geometry on the turbulence spectrum
  • Utilized the flexible magnetic geometry of the LAPD to create single and multiple mirrors.
  • Analyzed ion saturation current, floating potential, and magnetic fluctuation probe measurements using cross-correlation techniques in Python
  • Saw no mirror-driven instabilities, but observed modification of the turbulence spectrum and decreased cross-field particle flux with increased mirror ratio

 

Advising

Advised four UCLA physics undergraduates:

  • Juri Alhuthali (Summer 2024 - Fall 2024)
    • Supervised Bayesian inference of swept Langmuir probe traces
    • Used PyStan (a Bayesian inference package) and Pyro (a probabilistic computing package)
  • Jessica Gonzalez (Summer 2023 - Summer 2024)
    • Supervised analysis of time-series monochromator signals of three Helium neutral lines
    • Analyzed time traces using ColRadPy, a collisional-radiative solver
    • Compared results with measurements from the LAPD Thomson scattering system
  • Tyler Hadsell (Spring 2022 - Spring 2024)
    • Supervised python-based analysis of Phantom v7.3 fast framing camera footage of the LAPD
    • Supervised development of clustering routines to quantify dataset diversity and extract trends
  • Kian Orr (Summer 2021 - Spring 2022)
    • Supervised construction and deployment of software to control and read from an Ocean Insight HR4000 spectrometer
    • Integrated spectra into the existing machine state and auxiliary diagnostics system
    • Advised analysis of recorded spectral lines using ColRadPy to deduce electron temperature and densities in the LAPD

 

Publications

  • P. Travis, J. Bortnik, T. Carter, “Machine-learned trends in mirror configurations in the Large Plasma Device” Physics of Plasmas 32, 082106 (2025). doi:10.1063/5.0270755
  • P. Travis, T. Carter, “Turbulence and transport in mirror geometries in the Large Plasma Device” Journal of Plasma Physics 91 (2025). doi:10.1017/S0022377825000029
  • Qian, Yuchen, et al. “Design of the Lanthanum hexaboride based plasma source for the large plasma device at UCLA.” Review of Scientific Instruments 94, 085104 (2023). doi:10.1063/5.0152216

 

Selected presentations

  • An open dataset from the Large Plasma Device for machine learning and profile prediction (Poster)
    • Phil Travis, Troy Carter — APS Division of Plasma Physics 2024, Atlanta, GA
  • Predicting profiles in LAPD mirror configurations (Poster)
    • Phil Travis, Tom Look, Lukas Rovige, Chris Niemann, Pat Pribyl, Troy Carter — APS Division of Plasma Physics 2023, Denver, CO
  • Developing a generative ML model for LAPD trend inference and profile prediction (Poster)
    • Phil Travis, Steve Vincena, Patrick Pribyl, Troy Carter — APS Division of Plasma Physics 2022, Spokane, WA
  • Developing a generative ML model for profile evolution prediction on the LAPD (Poster)
    • Phil Travis, Steve Vincena, Patrick Pribyl, Troy Carter — Transport Task Force 2022, Santa Rosa, CA
  • Upgrading LAPD diagnostic pipelines for training generative ML models (Poster)
  • Generating synthetic LAPD discharges using energy-based models (EBMs) (Contributed talk)
    • Phil Travis, Steve Vincena, Troy Carter — APS Division of Plasma Physics 2021, Pittsburgh, PA
  • Autosweep: automated Langmuir sweep analysis (Contributed talk)
    • Phil Travis — APS Division of Plasma Physics 2020, Cyberspace
  • Automated Langmuir sweep analysis using machine learning (Poster)
    • Phil Travis and Troy Carter — APS Division of Plasma Physics 2019, Ft. Lauderdale, FL
  • Study of turbulence and transport in magnetic mirror geometries in the LAPD (Poster)
    • Phil Travis and Troy Carter — Transport Task Force 2018, San Diego, CA
  • Dependence of edge profiles and stability on neutral beam power in NSTX (Poster)
    • P. Travis, G. Canal, T. Osborne, R. Maingi, S. Sabbagh, and the NSTX-U team — APS Division of Plasma Physics 2016, San Jose, CA

 

Skills

Computing

  • Proficient in Python, C/C++, LabView, LaTeX
  • Experience with CUDA, OpenMP, MPI
  • Experience with MATLAB/Octave
  • Proficient in PyTorch; experience with TensorFlow, JAX
  • Proficient in Linux (Ubuntu and Debian)
  • Proficient in handling TB-scale data aggregation and storage (sockets/networking, HDF5 and binary files)

Diagnostics and analysis

  • Proficient in analysis of: magnetic flux (bdot), electrostatic (Langmuir) probes, non-collective Thomson scattering, heterodyne interferometry
  • Experience with analyzing spectroscopic data and visible-light diagnostics
  • Proficient in using cross-correlation and spectral techniques

Simulation

  • Experience with Vlasov and fluid solvers
  • Experience optimizing through simulators

Machine learning and statistics

  • Proficient in using neural networks, generative modeling, gradient-based optimization
  • Experience with Bayesian inference, MCMC methods, uncertainty quantification
  • Experience with classical ML techniques (regression forests, clustering, support vector machines, gaussian processes, and so on)

Laboratory techniques

  • Proficient in using general lab electronics (oscilloscopes, function generators, etc…)
  • Experience with operating a 300 mJ laser and supporting equipment
  • Proficient in using probe biasing and signal conditioning equipment
  • Proficient in interfacing with general lab hardware and DAQ devices