Pith Number

pith:64SV5BSO

pith:2025:64SV5BSOZBP3Z432G3DIVXP2FH

not attested not anchored not stored refs resolved

On the Unreasonable Effectiveness of Last-layer Retraining

John C. Hill, Tyler LaBonte, Vidya Muthukumar, Xinchen Zhang

Last-layer retraining succeeds mainly because the held-out set has better group balance than the full training data.

arxiv:2512.01766 v2 · 2025-12-01 · cs.LG

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

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{64SV5BSOZBP3Z432G3DIVXP2FH}

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

We present strong evidence for an alternative hypothesis: that the success of LLR is primarily due to better group balance in the held-out set.

C2weakest assumption

That the empirical tests on neural collapse are conclusive enough to rule it out as a contributing factor, and that group balance differences are the dominant cause rather than other unmeasured effects.

C3one line summary

Last-layer retraining succeeds primarily because the held-out set has better group balance, supported by experiments that do not back the neural collapse hypothesis.

References

4 extracted · 4 resolved · 0 Pith anchors

[1] Boosting the margin: A new explanation for the effectiveness of voting methods 1998 · doi:10.1073/pnas.2103091118.url:http://dx.doi.org/10.1073/

[2] A broad-coverage challenge corpus for sentence understanding through inference 2018

[3] Note that Waterbirds is the only dataset that has a distribution shift and MultiNLI is the only dataset which is class-balanceda priori 1992

[4] These pretrained models are used as the initialization for ERM finetuning under the cross-entropy loss 2022

Formal links

2 machine-checked theorem links

Receipt and verification

First computed	2026-05-17T23:39:16.974390Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

f7255e864ec85fbcf37a36c68addfa29c1200a39eb8de91d656f6b98fc79e1c4

Aliases

arxiv: 2512.01766 · arxiv_version: 2512.01766v2 · doi: 10.48550/arxiv.2512.01766 · pith_short_12: 64SV5BSOZBP3 · pith_short_16: 64SV5BSOZBP3Z432 · pith_short_8: 64SV5BSO

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/64SV5BSOZBP3Z432G3DIVXP2FH \
  | 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: f7255e864ec85fbcf37a36c68addfa29c1200a39eb8de91d656f6b98fc79e1c4

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "ac75b42556910ed9a5a2a132fcb3231d6691f4b187e09d64cb8ee5f3262283f3",
    "cross_cats_sorted": [],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2025-12-01T15:08:43Z",
    "title_canon_sha256": "4c1784338706bb0534f91fa5c9a711ef284fd1913be679bb249fca138a7e1b02"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2512.01766",
    "kind": "arxiv",
    "version": 2
  }
}