ICML 2026 Repro - Why Self-Training Helps and Hurts
Collection
Reproduction artifacts for ICML 2026 paper VnA5q5jXVz (arXiv 2602.14029): linear-regression theory + ResNet-50/CIFAR-10 self-distillation. • 3 items • Updated
config dict | iterations list | curves dict |
|---|---|---|
{
"n": 5000,
"K": 8,
"epochs": 25,
"etas": "0.4,0.6,0.8",
"repo": "Eishaan/selfdistill-cifar-repro"
} | [
0,
1,
2,
3,
4,
5,
6,
7,
8
] | {
"0.4": [
65.74000120162964,
65.27000069618225,
64.8499995470047,
65.3899997472763,
65.59999883174896,
66.7499989271164,
66.54999852180481,
67.01000034809113,
67.21000075340271
],
"0.6": [
80.50999939441681,
81.35000020265579,
81.22999966144562,
81.649999320507... |
No dataset card yet
- Downloads last month
- 105