A demo series · Emma Chickles · MIT Kavli

ZTF Light-Curve Representations

When must a model actually read physical time — and when is template-matching enough?

Self-supervised encoders look strong on light-curve classification. But a simple diagnostic shows they get there by matching class templates, not by reading each source's period off its raw, irregularly-sampled brightness — its absolute time. This series traces that thread: how we measure it, why even continuous-time models still don't read time, when the shortcut breaks, and what would fix it.

The one idea. High downstream R²/accuracy is not evidence of learned time. Period regression splits into a between-class piece (identify the class, predict its mean period = template matching) and a within-class piece (read the period off the individual source). Across eight encoders the within-class signal stays at ρ ≤ 0.26, while a classical Lomb–Scargle extractor reaches 0.78 on the same data. The signal is there; the encoders under-read it.

The series

0

Embedding Explorer

live

The map. UMAP of ~25k ZTF periodic variables, color by class / period / Gaia / anomaly score; click any source for its light curve, HR position, and nearest neighbors.

1

The Diagnostic

live

What makes an embedding "good"? Overall R² vs ρ_within per encoder — high R² with tiny ρ_within is template-matching; only Lomb–Scargle reads the per-source period. The operational definition the rest of the series uses.

2

The Externalization Ladder

live

Why doesn't continuous time help — and should we just feed a power spectrum? ρ_within barely moves as you add time to the input (none → continuous → +Δt); it jumps only when the search is done explicitly — and Lomb–Scargle is that power-spectrum-plus-peak-pick.

3

When Template-Matching Breaks

soon

Is reading time even important? Classification is fine → period regression breaks → quasiperiodic / period-changing sources break hard → time-structural anomalies become invisible.

4

Anomaly Lab

live

Global top outliers plus the most anomalous member of each variability class and each embedding region — a window on the shape anomalies the embedding can find (and a setup for the time anomalies it can't).

Building blocks (feed Chapter 2)

Period-aliasing explorer ↗

Why the same star has many "periods": the P/2 harmonic and its diurnal aliases, with a live phase-fold slider.

Period-search landscape ↗

Why gradient descent can't find a period: the dispersion landscape is flat with razor-thin spikes — period reading is a search, not a smooth readout.