Pith Number

pith:IGINEITW

pith:2026:IGINEITWTVLJ477W3AH4PA3C4Y

not attested not anchored not stored refs resolved

Inference-Time Dynamic Modality Selection for Incomplete Multimodal Classification

Chen Qin, Declan P. O'Regan, Siyi Du, Xinzhe Luo

DyMo uses task loss computed at inference time as a proxy to select which recovered modalities to fuse for each sample.

arxiv:2601.22853 v3 · 2026-01-30 · cs.CV

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{IGINEITWTVLJ477W3AH4PA3C4Y}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

DyMo significantly outperforms state-of-the-art incomplete/dynamic MDL methods across various missing-data scenarios by adaptively identifying and fusing reliable recovered modalities using a task-loss proxy for information maximization.

C2weakest assumption

The theoretically established connection between multimodal task-relevant information and the computable task loss at inference time is a valid and sufficient proxy for guiding modality selection without introducing bias from the recovery process.

C3one line summary

DyMo dynamically selects reliable recovered modalities at inference by using task loss as a proxy for task-relevant information, outperforming prior discard-or-impute methods on image datasets.

References

13 extracted · 13 resolved · 2 Pith anchors

[1] Best of both worlds: Multimodal contrastive learning with tabular and imaging data 2026

[2] Adam: A Method for Stochastic Optimization · arXiv:1412.6980

[3] Multi- modal deep learning 2026

[4] Deep learning and the information bottleneck principle 2015

[5] Deep Multimodal Learning with Missing Modality: A Survey · arXiv:2409.07825

Cited by

1 paper in Pith

Beyond Surface Artifacts: Capturing Shared Latent Forgery Knowledge Across Modalities

Receipt and verification

First computed	2026-05-18T02:45:05.650612Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

4190d222769d569e7ff6d80fc78362e60d465632a9b1e2a02679499525fd44ba

Aliases

arxiv: 2601.22853 · arxiv_version: 2601.22853v3 · doi: 10.48550/arxiv.2601.22853 · pith_short_12: IGINEITWTVLJ · pith_short_16: IGINEITWTVLJ477W · pith_short_8: IGINEITW

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/IGINEITWTVLJ477W3AH4PA3C4Y \
  | 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: 4190d222769d569e7ff6d80fc78362e60d465632a9b1e2a02679499525fd44ba

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "c6d980674ff291a7786b841d8d358c9969a5678a4c379b3f084f249655e2ac0b",
    "cross_cats_sorted": [],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.CV",
    "submitted_at": "2026-01-30T11:24:43Z",
    "title_canon_sha256": "ce8c0a1f6a6cabc93275bdce39b18abc36a71a9a393efb409d87b8e1f81d4b47"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2601.22853",
    "kind": "arxiv",
    "version": 3
  }
}