pith:E5JZWUAY
VISD: Enhancing Video Reasoning via Structured Self-Distillation
Structured self-distillation with a video-aware judge improves VideoLLM reasoning accuracy and training efficiency.
arxiv:2605.06094 v4 · 2026-05-07 · cs.CV · cs.AI
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{E5JZWUAYMP5JRYZTBQPZHO5G2X}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
Experiments on diverse benchmarks show that VISD consistently outperforms strong baselines, improving answer accuracy and spatio-temporal grounding quality. Notably, VISD reaches these gains with nearly 2x faster convergence in optimization steps.
The video-aware judge model produces diagnostically meaningful, unbiased privileged information that can be safely used for token-level supervision without introducing new failure modes or reward hacking.
VISD adds structured privileged feedback from a judge model and a direction-magnitude decoupling trick to let VideoLLMs learn token-level credit assignment while keeping RL stable, yielding higher accuracy and roughly 2x faster convergence on video reasoning benchmarks.
Cited by
Receipt and verification
| First computed | 2026-05-25T02:01:22.117869Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
27539b501863fa98e3330c1f93bba6d5d2f766536f0779258990c2093f428886
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/E5JZWUAYMP5JRYZTBQPZHO5G2X \
| 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: 27539b501863fa98e3330c1f93bba6d5d2f766536f0779258990c2093f428886
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "66047564e89d9049c1c3abc2e89c63810c04935c8b1b3eb8ab2084e8a2cb3c3b",
"cross_cats_sorted": [
"cs.AI"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"primary_cat": "cs.CV",
"submitted_at": "2026-05-07T12:13:15Z",
"title_canon_sha256": "93d04c2f3cc99c20cb363f9751e2ec4788e7659aed38c0cea8dddb485f8c67e8"
},
"schema_version": "1.0",
"source": {
"id": "2605.06094",
"kind": "arxiv",
"version": 4
}
}