Alyssa Johnson

I’m a Ph.D. candidate and HEPCAT Fellow / Sloan Scholar working in experimental physics, optical instrumentation, and high-dimensional data analysis. Outside of the lab, I work as a quantitative developer building ML systems for signal detection, forecasting, and alpha-driven financial modeling. Drawing on my background in signal processing, anomaly detection, and large-scale data pipelines developed through instrumentation and astrophysics research, I apply a physics-driven approach to financial modeling and market analysis. My projects span end-to-end quantitative trading systems, NLP-based sentiment analysis, ensemble modeling, and automated backtesting to generate and evaluate alpha signals from market data.

I earned my B.S. in Physicsfrom California State Polytechnic University, Humboldt, as well as an MFA and BFA in Painting & Drawing from The Ohio State University and Temple University, respectively. A highlight of my undergraduate career was completing a systems engineering internship at NASA Goddard Space Flight Center under the NASA Sally Ride Fellowship and the Space Generation Advisory Council (SGAC) Scholarship. As a NASA intern I supported the Laser Communications Relay Demonstration (LCRD) mission by developing experiments that demonstrated user data services for the satellite’s optical communications system. My current research is conducted with Dr. Shelley Wright in UCSD’s Optical & Infrared Laboratory, where I am supported by the HEPCAT Fellowship to work on the assembly, integration, and validation of PANOSETI telescope modules using SiPM detectors for nanosecond-scale optical transient searches.

Projects

In-Database ML Forecasting with MySQL HeatWave

Benchmarked MySQL HeatWave AutoML against XGBoost, LightGBM, LSTM, and GRU on time series forecasting — evaluating accuracy and inference latency across in-database and external compute.

Quant / ML · Signal Processing

Stellar Density × Detection Probability FoM for PANOSETI

Computed a sky-wide detection Figure of Merit across ~3M Galactic directions, combining multi-component stellar density, 3D dust extinction, and a detection probability kernel to prioritize PANOSETI telescope pointings.

Physics

Active Servo-Leveling Tilt Analysis

First-author publication in J. Undergrad. Rep. Phys. (2023) — tilt-control calibration tools for the HSU short-range gravity experiment, with quadratic regression on voltage-tilt sensor data.

Physics · Signal Processing paper

Solar UV Irradiance Spectroscopy – MSU REU

• Analyzed UARS SOLSTICE satellite data to characterize solar UV irradiance across C IV (1548 Å) and Ne VIII (770 Å) transition region and coronal spectral lines.
• Developed IDL pipelines to ingest multi-date solar irradiance spectra, producing wavelength-irradiance plots and time-series animations.
• Demonstrates experience in satellite remote sensing data reduction and spectral analysis of high-temperature solar atmospheric emission.

Physics website

Planck × ACTPol Harmonic-Space Map Analysis

• Analyzed Planck HFI sky maps (353–857 GHz) and ACTPol data using HEALPix, performing coordinate transformations between ICRS and Galactic frames.
• Implemented spherical harmonic filtering via healpy map2alm/alm2map pipeline, applying Gaussian window functions in harmonic space to full-sky CMB maps.
• Demonstrates foundational proficiency in multi-frequency CMB map analysis, harmonic-space signal processing, and mm-wave dataset cross-matching.

Physics

Cross-Correlation of Planck Dust Maps with ACT SZ Clusters

• Performed a stacking analysis of Planck 353 GHz dust emission at ACT DR5 SZ cluster positions to probe dust filament structure in cluster environments.
• Processed full-sky HEALPix maps (nside=2048) using pymaster and healpy, cross-matching cluster catalogs with Planck Legacy Archive data via astropy.
• Demonstrates experience in CMB × large-scale structure cross-correlation, catalog-based stacking pipelines, and multi-survey mm-wave data analysis.

Physics

In-Database ML Forecasting with MySQL HeatWave

Quant / ML · Signal Processing

×

Multi-model benchmark — accuracy metrics, inference latency, and normalized performance heatmap across ARIMA, XGBoost, LightGBM, LSTM, and GRU

Training loss (MSE) over 50 epochs — LSTM vs. GRU

• Built an end-to-end forecasting pipeline inside MySQL HeatWave — data ingestion, feature engineering, model training, and inference without leaving the database.
• Benchmarked HeatWave AutoML against XGBoost, LightGBM, LSTM, and GRU on the same time series targets using a shared evaluation harness; evaluated MAE, RMSE, MAPE, directional accuracy, and inference latency.
• Gradient boosting models (XGBoost, LightGBM) achieved lowest error metrics; GRU offered the best accuracy-to-latency tradeoff among deep learning architectures. ARIMA served as the statistical baseline.
• Demonstrated that in-database ML with HeatWave AutoML is viable for rapid prototyping, with competitive accuracy and significantly lower operational overhead versus external Python pipelines.

GitHub →

Stellar Density × Detection Probability FoM for PANOSETI

Physics

×

Top to bottom: integrated stellar density (10⁶ stars sr⁻¹ to 0.5 kpc), V-band dust extinction A_V (mag), and detection FoM F_deep — full Galactic sky in Mollweide projection

• Computed a detection-probability-weighted FoM integrating multi-component stellar density (thin/thick disk, bulge, halo) against 3D dust extinction from Bayestar19 maps.
• Pipeline processes ~3M HEALPix directions using vectorized NumPy operations — full run completes in under 10 minutes on a laptop.
• Output sky maps directly inform PANOSETI telescope pointing strategy for maximizing SETI detection probability.

GitHub →