pith:M7LQKF3N
Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT
ConSFT preserves pre-trained capabilities in flow-matching VLAs by scaling learning signals to model confidence during fine-tuning.
arxiv:2605.08879 v2 · 2026-05-09 · cs.RO
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{M7LQKF3NPAOBGIVK7FBGNA3JSQ}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
The method outperforms vanilla SFT in capability retention by an average absolute margin of over 20%, matching the efficacy of data-heavy Experience Replay in a prior-data-free regime.
That dynamically scaling learning signals based on model confidence effectively bounds intrinsic parameter disruption risk and secures both target convergence and prior capability retention without introducing new failure modes or limiting necessary adaptation.
ConSFT prevents catastrophic forgetting in fine-tuning flow-matching VLAs by dynamically scaling gradients based on model confidence, retaining over 20% more pre-trained capability than standard SFT without prior data or reference networks.
Receipt and verification
| First computed | 2026-05-20T01:05:16.180327Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
67d705176d781c1322aaf942668369942bd007edb13abc41477feb2956d4361c
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/M7LQKF3NPAOBGIVK7FBGNA3JSQ \
| jq -c '.canonical_record' \
| python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 67d705176d781c1322aaf942668369942bd007edb13abc41477feb2956d4361c
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "c69278ea19f1bb98ffe8f9e7c40c0ec650dbcd746b7299449bd7e919e49c780e",
"cross_cats_sorted": [],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"primary_cat": "cs.RO",
"submitted_at": "2026-05-09T10:59:03Z",
"title_canon_sha256": "c53e7388114353dc17753acb434f7799b4563abced5c0edd7c6d3f6742166908"
},
"schema_version": "1.0",
"source": {
"id": "2605.08879",
"kind": "arxiv",
"version": 2
}
}