18.4 MB

Ctrl+K

1 contributor

Claim 6: verify the GW gradient to measured order 2.00 across three CNT families at n=150, and run gradient descent that monotonically reduces the loss at n up to 300

c37ab20 verified about 15 hours ago

code
Claim 6: verify the GW gradient to measured order 2.00 across three CNT families at n=150, and run gradient descent that monotonically reduces the loss at n up to 300 about 15 hours ago
outputs
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
packaged_replay
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
pages
Claim 6: verify the GW gradient to measured order 2.00 across three CNT families at n=150, and run gradient descent that monotonically reduces the loss at n up to 300 about 15 hours ago
repro_code
Add entropic Gromov-Wasserstein experiments 1 day ago
source
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
.gitattributes
1.88 kB
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
BUNDLE_SHA256SUMS.txt
6.33 kB
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
README.md
665 Bytes
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
SOURCE_PIN.txt
639 Bytes
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
bucket-icon.svg
413 Bytes
Publish independent ICML 2026 reproduction: Optimal Unconstrained Self-Distillation 7 days ago
build_manifest.py
952 Bytes
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
claims.json
1.19 kB
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
index.html
1.85 kB
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
logbook.css
29.6 kB
Publish independent ICML 2026 reproduction: Optimal Unconstrained Self-Distillation 7 days ago
logbook.js
77.3 kB
Publish independent ICML 2026 reproduction: Optimal Unconstrained Self-Distillation 7 days ago
logbook.json
5.13 kB
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
official_claims.json
1.19 kB
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
outputs_gw2_results.json
2.73 kB
Claim 1: measure real wall-time/peak-heap scaling (slopes 1.85/1.95) instead of counting operations. Claim 5: test SGW non-negativity on CNT costs (tree metrics, circle geodesics) with a corrected isometric control about 16 hours ago
outputs_gw5_results.json
7.55 kB
Claim 3: verify Theorem 3.4's reduction via the Krein GW-embedding identity (quartic contraction vs Hilbert-Schmidt form) to 1.7e-14 at n=180 on three CNT families, for arbitrary couplings about 16 hours ago
outputs_gw6_results.json
3.34 kB
Claim 4: test descent from a random initialisation at n=400/200 iters (28-93% decrease); monotone in 8/9 cells, symmetric-instance exception attributed to entropic instability about 15 hours ago
outputs_gw7_results.json
2.78 kB
Claim 6: verify the GW gradient to measured order 2.00 across three CNT families at n=150, and run gradient descent that monotonically reduces the loss at n up to 300 about 15 hours ago
poster_embed.html
1.62 kB
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
reproduce.py
19.6 kB
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
requirements.txt
32 Bytes
Repurpose zero-loss duplicate for Gromov-Wasserstein scale reproduction 4 days ago
trackio-logo-light.png
30 kB
Publish independent ICML 2026 reproduction: Optimal Unconstrained Self-Distillation 7 days ago
trackio-logo.png
55.6 kB
Publish independent ICML 2026 reproduction: Optimal Unconstrained Self-Distillation 7 days ago
trackio-wordmark-dark.png
89.8 kB
Publish independent ICML 2026 reproduction: Optimal Unconstrained Self-Distillation 7 days ago