Pith Number

pith:VPUYMYPB

pith:2026:VPUYMYPBMGB2PBR76ZODRJ2PMC

not attested not anchored not stored refs resolved

Pessimistic Risk-Aware Policy Learning in Contextual Bandits

Xianyi Wu, Yilong Wan, Yuqiang Li

Optimizing general Lipschitz risk criteria in offline contextual bandits incurs no additional statistical cost beyond expected-reward optimization.

arxiv:2605.15620 v1 · 2026-05-15 · stat.ML · 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{VPUYMYPBMGB2PBR76ZODRJ2PMC}

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

By developing novel empirical concentration inequalities for importance sampling-based distributional estimators, our analysis derives data-dependent suboptimality bounds with an Õ(1/√n) rate, without relying on restrictive uniform overlap assumptions. This rate is minimax optimal and matches that of risk-neutral offline policy optimization, indicating that optimizing general Lipschitz risk criteria incurs no additional statistical cost relative to the expected-reward.

C2weakest assumption

The risk functionals under consideration are Lipschitz continuous (invoked to unify mean-variance, entropic risk, CVaR, etc.), and the novel empirical concentration inequalities for the importance-sampling distributional estimators hold under the paper's data-dependent conditions rather than uniform overlap.

C3one line summary

A distributional framework for optimizing Lipschitz risk functionals in offline contextual bandits yields data-dependent suboptimality bounds of Õ(1/√n) that match risk-neutral rates and are minimax optimal.

References

48 extracted · 48 resolved · 0 Pith anchors

[1] Mathematical Finance , volume = 1999 · doi:10.1111/1467-9965.00068

[2] Policy learning with observational data 2021

[3] Regret bounds for risk-sensitive reinforcement learning 2022

[4] Bellemare, Will Dabney, and R \'e mi Munos 2017

[5] From predictive to prescriptive analytics 2019 · doi:10.1287/mnsc.2018.3253

Formal links

1 machine-checked theorem link

Receipt and verification

First computed	2026-05-20T00:01:08.569800Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

abe98661e16183a7863ff65c38a74f609acc3b4f5ba010438fe2c47f372b18ca

Aliases

arxiv: 2605.15620 · arxiv_version: 2605.15620v1 · doi: 10.48550/arxiv.2605.15620 · pith_short_12: VPUYMYPBMGB2 · pith_short_16: VPUYMYPBMGB2PBR7 · pith_short_8: VPUYMYPB

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/VPUYMYPBMGB2PBR76ZODRJ2PMC \
  | 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: abe98661e16183a7863ff65c38a74f609acc3b4f5ba010438fe2c47f372b18ca

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "50373642908d5d2856918dfd4d66f73ac2a09880e82adcaf66f556428726e16e",
    "cross_cats_sorted": [
      "cs.LG"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "stat.ML",
    "submitted_at": "2026-05-15T05:02:39Z",
    "title_canon_sha256": "1cebe2e302eaee8d624a7f67f11397e57cbc39183447f338bc17e23d1dab15de"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.15620",
    "kind": "arxiv",
    "version": 1
  }
}