diff --git a/spectral_scheduling.md b/spectral_scheduling.md
new file mode 100644
index 000000000..5ac180ea9
--- /dev/null
+++ b/spectral_scheduling.md
@@ -0,0 +1,8 @@
+# Spectral Scheduling
+
+labels: experimental
+
+1. learn a singular basis over the data
+2. treat the eigenvalues as relative sampling weight, sample a basis dimension/component each batch
+3. sample a batch of data, using that datum's value in the sampled component as their relative likelihood (threshold this)
+4. incorporate a temperature term so we can sample disproportionately from the important components in early training, progressively increasing the weight of the tail probabilities.