Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Conference on Health, Inference, and Learning (CHIL) · brycewang-stanford | DecimalAI

brycewang-stanford/conference-on-health-inference-and-learning

SkillScore73Maintained

Use when targeting Conference on Health, Inference, and Learning (CHIL) or deciding whether a computer-science manuscript fits this venue. Encodes conference fit, framing, evidence bar, submission-cycle checks, rebuttal posture, and desk-reject risks for AI for health.

researchby brycewang-stanford

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Research pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/brycewang-stanford-conference-on-health-inference-and-learning/SKILL.md

decimalai skills pull brycewang-stanford-conference-on-health-inference-and-learning --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/brycewang-stanford-conference-on-health-inference-and-learning/SKILL.md https://app.decimal.ai/s/brycewang-stanford-conference-on-health-inference-and-learning/SKILL.md

Raw file: /s/brycewang-stanford-conference-on-health-inference-and-learning/SKILL.md ↗ · machine-readable summary at /s/brycewang-stanford-conference-on-health-inference-and-learning.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 19d ago

Yes — +59% lift · at higher cost+55% tokens0% turns73% of whole cases pass with skill (16/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-01	✗→✓	▲ Improved	35%	0%
case-02	✗→✓	▲ Improved	46%	0%
case-03	✗→✓	▲ Improved	29%	0%
case-10	✗→✓	▲ Improved	46%	0%
case-11	✗→✓	▲ Improved	21%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/10/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

brycewang-stanford / awesome-journal-skills

Computer-Science-Conference-Skills/skills/conference-on-health-inference-and-learning/SKILL.md

View on GitHub ↗

License MITImported 8/10/2026Version v1

Skill contents

SKILL.md

Conference on Health, Inference, and Learning (CHIL)

Conference positioning

Conference on Health, Inference, and Learning (CHIL) is a top computer-science conference venue for machine learning for health, clinical inference, health systems, fairness, robustness, and deployment constraints. It rewards a health ML paper with clinical grounding, validation discipline, and clear claims about use context. Treat this skill as a fit / venue-selection / re-framing tool for conference submission strategy, not as a substitute for the current year's CFP, author kit, ethics policy, or submission portal.

Because CS conferences change deadlines, templates, page limits, review workflow, artifact rules, AI-use policy, and rebuttal formats every cycle, always verify the live official instructions before making a submission-ready recommendation. Start from the official source anchor recorded for this venue in ../../resources/conference-roster.md and ../../resources/official-source-map.md.

When to trigger

The author names CHIL / Conference on Health, Inference, and Learning as the target venue.
A manuscript in machine learning for health needs a conference-fit read before being formatted or submitted.
The paper must be re-framed from journal style or arXiv style into a selective CS conference narrative.
The author needs an evidence-gap, anonymity, artifact, rebuttal, or re-routing diagnosis for this venue.

Scope & topic fit

Core fit: machine learning for health, clinical inference, health systems, fairness, robustness, and deployment constraints.
Best submissions make a precise contribution type visible: algorithm, theorem, system, dataset, benchmark, empirical finding, design artifact, tool, or socio-technical analysis.
The paper should explain why the result matters to CHIL's reviewers, not just why it is interesting to the authors' lab or product context.
Position related work against the most recent conference-cycle papers in this venue and its closest siblings; stale comparisons are a common early-review weakness.
If the contribution is interdisciplinary, state which part is CS research and which part is domain evidence.

Venue-specific calibration

Reviewer lens: Treat CHIL as a AI for health venue whose reviewers expect the scope and evidence to match its own community. Do not submit a generic CS paper until the introduction names the exact subcommunity, contribution type, and proof or empirical standard.
Contribution hook to foreground: the venue-specific contribution bar.
Scope vocabulary to use naturally in the abstract and introduction: machine learning for health, clinical inference, health systems, fairness, robustness, and deployment constraints.
Distinctive fingerprint for reviewer calibration: machine, learning, health, clinical, inference, fairness, robustness, deployment, constraints, venue-specific, contribution, chilconference.
Official anchor domain: www.chilconference.org. Quote annual rules only after opening that source and the current-year CFP/author kit.

Close-neighbor routing guardrail

Use this profile only when the manuscript's central contribution is genuinely in AI for health

and the author can say why CHIL reviewers are the primary audience, not merely a convenient deadline.

Closest roster neighbors to compare before final routing: conference-on-lifelong-learning-

agents (CoLLAs), international-conference-on-automated-machine-learning (AutoML Conference), machine-learning-for-health (ML4H), acm-sigkdd-conference-on-knowledge- discovery-and-data-mining (KDD). Break ties by contribution type, evidence shape, reviewer community, and the current official CFP from www.chilconference.org.

CHIL-specific routing detail

Prefer CHIL when the novelty depends on health context: clinical prediction, care delivery, health-system deployment, fairness across patient populations, robustness under clinical shift, or validated inference from health data.
Use KDD/ICML/NeurIPS only when the general machine-learning or data-mining contribution is strong without the clinical setting; CHIL reviewers should see why the health setting changes the method, validation, or failure analysis.
Do not route optimization, scheduling, or operations-research papers here just because they use medical examples. If the central contribution is solver design, resource allocation, or constraint modeling, compare with CPAIOR, ICAPS, INFORMS-facing venues, or health-operations outlets first.

Method & evidence bar

Compare against current strong baselines and explain exactly what changes in the algorithm, objective, data, or inference procedure.
Report ablations that isolate the claimed mechanism; do not rely on aggregate benchmark wins alone.
Document data, compute, hyperparameters, model selection, and failure cases so the result can be reviewed as science rather than demo output.
For CHIL, the evidence must support the venue-specific signature: a health ML paper with clinical grounding, validation discipline, and clear claims about use context.
Include limitations, negative results, compute/resource reporting, data provenance, and ethics details when they affect the claim.

Structure & house style

Frame the contribution as a reusable idea: method, theory, benchmark, dataset, system, or socio-technical finding.
Separate main claims from exploratory results; reviewers at top AI venues punish overclaiming and hidden cherry-picking.
Use the current official template exactly; do not guess page limits, font sizes, supplement rules, anonymity exceptions, or camera-ready requirements from old cycles.
The introduction should answer: problem, why now, what is new, why this venue, and what evidence proves the claim.
Put the strongest result in the main paper, not only in the appendix or supplement; reviewers should not have to reconstruct the contribution.

Official-cycle checklist

Open the live official venue page: https://www.chilconference.org/
Re-check the current cycle's CFP, author kit, submission system, abstract/paper deadlines, page limits, supplementary-material rules, anonymity policy, dual-submission policy, ethics policy, AI-use policy, artifact/code/data expectations, rebuttal/author-response format, and camera-ready requirements.
Confirm the review workflow and portal: OpenReview / CMT / HotCRP / PCS / START or society portal, as specified for the current cycle.
Check whether accepted papers require in-person presentation, separate registration, artifact badges, proceedings copyright, or post-acceptance release forms.
If the live official instructions conflict with this skill, the official instructions win.

Pre-submission self-check

] One sentence states why this manuscript belongs at CHIL, using the venue's scope rather than generic "top conference" language.
] The claim is calibrated to the evidence: no broader than the datasets, proofs, systems, user studies, deployments, or threat model support.
] Related work includes the nearest current-cycle AI for health papers and explains the technical delta.
] The paper satisfies the current official template, anonymity, ethics, artifact, and rebuttal requirements.
] The main paper is self-contained enough for reviewers to evaluate novelty and correctness without hunting through external links.

Common desk-reject triggers

Leaderboard-only novelty with weak explanation of why the method works.
Unclear data contamination, missing baselines, or evaluation that cannot be reproduced.
Claims about safety, fairness, health, or society without matching evidence and limitations.
Formatting, anonymity, dual-submission, external-link, or supplement violations under the current-year policy.
A contribution framed for a neighboring field while giving CHIL reviewers too little technical or empirical substance.

Re-routing decision

If the paper misses CHIL's bar, compare against neural-information-processing-systems / international-conference-on-machine-learning / international-conference-on-learning-representations / aaai-conference-on-artificial-intelligence. Re-route based on contribution type, not prestige: theory to a theory venue, systems to a systems venue, application-heavy work to a domain venue, and early ideas to workshops or shorter tracks when the official CFP supports them.

Output format

text
[Fit] High / Medium / Low (one-line reason)
[Target] Conference on Health, Inference, and Learning (CHIL)
[Contribution type] algorithm / theory / system / dataset / benchmark / empirical / design / security / other
[Main evidence gap] <single most important missing proof, experiment, study, artifact, or policy check>
[Official items to re-check] CFP / author kit / deadline / format / anonymity / ethics / AI-use / artifact / rebuttal / camera-ready
[Top rejection risk] <venue-specific risk>
[Re-route suggestion] <better-matched conference or journal if not a fit>

Illustrative Want to see what brycewang-stanford/conference-on-health-inference-and-learning would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

9/3/2026, 04:04 UTC · 22 cases

Lift

+59%

73% vs 14%

Agent turns

1.0 vs 1.0

Tokens

+55%

3,679.95 vs 2,373.95

fail → pass · 14pass → pass · 2fail → fail · 5pass → fail · 1

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-07 We are designing a prospective clinical trial to validate our CHIL model in a local hospital ward. Please perform a statistical power analysis to calculate the required patient sample size to achieve 80% power at alpha = 0.05 given an expected effect size of d = 0.35.	fail→fail	22,158	22,506	+2%	1	1	0%	3,117	5,132	+65%	0	0	—
case-08 We developed a mixed-integer linear programming algorithm for operating room schedule optimization that minimizes surgical delay penalties, tested on 5,000 surgical records from a university hospital. We want to submit this to CHIL. Evaluate our venue fit and output the assessment in the standard bracketed structure.	fail→fail	17,156	7,320	-57%	1	1	0%	1,875	3,069	+64%	0	0	—
case-01 I am preparing a manuscript on a novel temporal graph neural network designed for predicting hospital readmission risks using Electronic Health Records across multi-site clinical deployments. I'm considering submitting to CHIL (Conference on Health, Inference, and Learning). Please evaluate my paper's suitability and provide your feedback formatted as: 1) Fit level with a concise rationale, 2) Target venue confirmation, 3) Contribution type categorization, 4) The main evidence gap in my current evaluation, 5) Official guidelines and policy items I need to review, 6) Primary rejection risk for this venue, and 7) A recommended re-route alternative if the paper would be better placed elsewhere.	fail→pass	15,473	15,080	-3%	1	1	0%	2,595	3,503	+35%	0	0	—
case-02 We drafted a paper presenting a subpopulation fairness framework for automated triage algorithms operating under distribution shifts in emergency department data. We want to check if the Conference on Health, Inference, and Learning is the right target before finalizing our formatting. Could you assess our draft and report your diagnosis using these structured fields: Fit rating and reason, target venue, contribution category, critical evidence gap to address, official checklist items to re-check, top venue-specific rejection risk, and alternative submission re-route suggestions?	fail→pass	20,692	15,153	-27%	1	1	0%	2,452	3,587	+46%	0	0	—
case-03 I have an arXiv preprint describing a curated multimodal benchmark dataset for ICU physiological time-series forecasting, complete with baseline models and missingness stress tests. I am aiming to submit this to CHIL. Please evaluate my paper against the conference requirements and give me a submission assessment containing: venue fit assessment (High/Medium/Low with reason), target conference, paper contribution classification, primary missing evidence or validation gap, official submission/policy items to verify, biggest reviewer rejection risk, and suggested venue re-routing options.	fail→pass	20,786	12,756	-39%	1	1	0%	2,569	3,324	+29%	0	0	—
case-04 I am writing a paper for the Conference on Health, Inference, and Learning on clinical risk prediction. Here is my current draft abstract: 'We propose a model for predicting ICU mortality using electronic health records. Our method achieves 0.85 AUC on MIMIC-III.' Please rewrite and expand this abstract draft into a camera-ready 250-word paragraph.	fail→fail	16,520	13,155	-20%	1	1	0%	1,907	4,006	+110%	0	0	—
case-05 We are preparing a benchmark submission for Conference on Health, Inference, and Learning using MIMIC-IV clinical EHR data. Please write a complete Python PyTorch script containing the model class, data loader, loss function, and training loop for a baseline LSTM sepsis mortality predictor.	fail→fail	28,122	30,432	+8%	1	1	0%	5,674	7,092	+25%	0	0	—
case-06 For our clinical inference study targeting the Conference on Health, Inference, and Learning, we need to apply for institutional IRB approval. Please draft the complete IRB research protocol document, including patient consent waiver arguments and HIPAA compliance risk mitigations.	fail→fail	29,804	23,434	-21%	1	1	0%	3,887	5,639	+45%	0	0	—
case-09 We designed a generic non-negative matrix factorization algorithm that speeds up convergence on large sparse matrices by 4x. To demonstrate it, we ran it on hospital billing codes alongside synthetic numerical tables. Should we submit this to Conference on Health, Inference, and Learning? Provide the assessment using the standard bracketed submission output block.	pass→pass	11,625	11,358	-2%	1	1	0%	1,637	3,069	+87%	0	0	—
case-10 We built an automated machine learning framework that searches over hyperparameter configurations specifically for tabular clinical EHR pipelines, evaluating pipeline candidate efficiency. We are targeting Conference on Health, Inference, and Learning. Provide a venue fit assessment using the bracketed structured layout.	fail→pass	21,663	9,921	-54%	1	1	0%	2,391	3,481	+46%	0	0	—
case-11 We created a lifelong continual learning algorithm that updates radiology classifier weights continuously over a 10-year period while preventing catastrophic forgetting of historic patient cohorts. Assess this for CHIL submission using the bracketed output structure.	fail→pass	19,457	12,842	-34%	1	1	0%	2,733	3,303	+21%	0	0	—
case-12 We developed a clinical inference method designed specifically for predicting acute kidney injury from continuous ICU telemetry under cross-hospital domain shifts, evaluated with extensive clinical subgroup validation across three health systems. Assess this manuscript for submission using the required bracketed fields.	fail→pass	22,570	12,525	-45%	1	1	0%	3,339	3,008	-10%	0	0	—
case-13 We conducted a rigorous empirical audit evaluating algorithmic subpopulation fairness and disparate impact in emergency department triage systems across diverse patient demographics across multi-site urban hospitals. Evaluate our manuscript for CHIL using the bracketed summary block.	fail→pass	6,210	7,246	+17%	1	1	0%	960	3,166	+230%	0	0	—
case-14 Our manuscript reports a 3% improvement in AUC for mortality prediction on MIMIC-III using a complex deep learning architecture, but we have not conducted ablation studies to isolate individual components or analyzed failure cases across patient cohorts. Evaluate this for CHIL using the bracketed output format.	pass→fail	13,642	6,237	-54%	1	1	0%	2,065	3,027	+47%	0	0	—
case-15 We are submitting a clinical NLP paper for sepsis detection to CHIL, but our related work section compares our method only against baseline algorithms published before 2018. Evaluate our manuscript's submission readiness using the bracketed format.	fail→pass	13,339	12,273	-8%	1	1	0%	2,094	3,212	+53%	0	0	—
case-16 Our paper claims that our deep learning model guarantees zero-risk clinical decision making and should immediately replace emergency room triage nurses, based solely on offline retrospective ROC-AUC scores on a single hospital dataset. Provide a venue assessment block in the bracketed structure.	pass→pass	16,033	8,264	-48%	1	1	0%	1,759	3,250	+85%	0	0	—
case-17 When preparing our CHIL submission strategy analysis, which official website domain should be verified for the latest author guidelines, ethics rules, and submission system details? Format your response in the standard bracketed structure.	fail→pass	1,966	5,598	+185%	1	1	0%	265	3,011	+1036%	0	0	—
case-18 We built a multi-center open-access benchmark dataset containing aligned EHR time-series and chest X-rays for multi-organ failure prediction, complete with missingness stress tests and baseline evaluations. Evaluate this paper for CHIL using the bracketed block.	fail→pass	25,166	11,532	-54%	1	1	0%	2,154	2,953	+37%	0	0	—
case-19 We designed an ultra-low-power embedded neural hardware architecture for continuous real-time ECG arrhythmia monitoring on wearable patient devices, evaluated under microsecond latency and energy constraints. Evaluate this for CHIL using the bracketed output structure.	fail→pass	14,225	8,535	-40%	1	1	0%	2,238	3,288	+47%	0	0	—
case-20 We conducted a multi-month qualitative and quantitative study analyzing how intensive care unit clinicians interact with and override automated sepsis alert recommendations in real-world clinical workflows. Assess this submission for CHIL in the bracketed output layout.	fail→pass	18,839	13,203	-30%	1	1	0%	2,032	3,121	+54%	0	0	—
case-21 Our team wants to know what official cycle items and portal rules we must double-check before submitting our clinical ML paper to CHIL. Present your response using the required bracketed fields.	fail→pass	19,625	7,430	-62%	1	1	0%	2,171	3,080	+42%	0	0	—
case-22 We wrote an interdisciplinary manuscript combining a novel Bayesian variational inference technique with a retrospective clinical observational study of diabetes outcomes. How should we structure the introduction for CHIL reviewers? Provide your response in the bracketed output structure.	fail→pass	37,341	18,928	-49%	1	1	0%	2,313	4,638	+101%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +59 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 13.6%
Pass rate with the skill: 72.7%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 1
Skill version measured: v1
Run completed: 9/3/2026, 04:04 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-01

Prompt

I am preparing a manuscript on a novel temporal graph neural network designed for predicting hospital readmission risks using Electronic Health Records across multi-site clinical deployments. I'm considering submitting to CHIL (Conference on Health, Inference, and Learning). Please evaluate my paper's suitability and provide your feedback formatted as: 1) Fit level with a concise rationale, 2) Target venue confirmation, 3) Contribution type categorization, 4) The main evidence gap in my current evaluation, 5) Official guidelines and policy items I need to review, 6) Primary rejection risk for this venue, and 7) A recommended re-route alternative if the paper would be better placed elsewhere.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run73

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+86% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 18 days ago

Other measured skills in the registry, with their headline benchmark lift.