PhD candidate in Physics at MIT (Binary Star Astrophysics group, MIT Kavli Institute). I use GPUs and machine learning to search hundreds of millions of stellar light curves for ultracompact binaries — white dwarf pairs that orbit in minutes, and among the loudest guaranteed sources for LISA. Along the way I build the tests that catch a model learning the survey instead of the sky.
👉 Start here: emmachickles.github.io
Featured work, papers, and interactive demos live there. A couple worth clicking directly:
- 🎨 Paint a star — on a real survey cadence, a light curve pins down only 19 numbers out of 4,608 map pixels. See what the prior fills in.
- 🔭 Embedding explorer — poke at what a self-supervised encoder actually learned from 26k stars.
inversebench-timedomain | Two inverse problems contributed to InverseBench (ICLR 2025). Measures what a learned prior invents, rather than caveating it. 🏆 Best Visualization, IAIFI 2026 |
period-diagnostic | Template Matching, Not Time Learning (ICML 2026, AI4Physics). Encoders look like they read stellar periods; mostly they recognize the class and recall its typical period |
ztf-pocket · ztf-embedding-demo | Dependency-free browser tools for inspecting learned representations |
astrotools · ztf-example | Analysis library and tutorial notebooks for survey light curves |
blender_binaries | Physically faithful renders of interacting binaries — the mesh is the surface producing the model light curve |
- Chickles, E. & Burdge, K., Template Matching, Not Time Learning — AI4Physics Workshop, ICML (2026) · paper · poster
- Chickles, E., et al., An eclipsing 8.56-minute orbital period mass-transferring binary — ApJ (2026)
- Chickles, E., et al., A gravitational-wave–detectable Type Ia supernova progenitor — ApJ (2025)
📧 echickle@mit.edu · 🌐 emmachickles.github.io · 🔭 ORCID