Apppublic

ProCreations/repro-near-optimal-and-efficient-first-order-algorithm-for-multi-task-learning-with-shared-linea

sourceHugging Faceupdated 3mo agoView on Hugging Face

0likes

App README

Reproduction bundle

Deterministic claim-complete audit of all six registered claims for OpenReview TnquAvyTtL and arXiv 2605.00473v1.

bash

PYTHONWARNINGS=error python3 reproduce.py --output-dir replay --source-dir source
python3 build_manifest.py --verify

The bundle covers the exact two-phase algorithm, regularizer gradients, eight finite TPGD runs, destructive controls, theorem-rate identities, sample requirements, dimension-independent contraction counts, and transfer-risk decomposition. The source's Phase-I prose typo is disclosed. Finite diagnostics do not replace the universal theorem proofs.