Competition ML vs. Production ML: A Systematic Gap Analysis
Fuente:
Zenodo
Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Recurso digital |
| Lingua: | inglese |
| Pubblicazione: |
Zenodo
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866901359929851904 |
|---|---|
| author | Emelianov, Ilya |
| author_facet | Emelianov, Ilya |
| contents | <p><span lang="EN-US">Kaggle has 23 million registered accounts. Only 612 of those users have ever earned Grandmaster status. Meanwhile, a Google L3 engineer — entry level — out-earns the 90th percentile of "Kaggle Master" job postings by a wide margin. These facts describe the same industry but fundamentally different skill markets. This paper presents a structured gap analysis between competition ML and production ML across six dimensions: metric choice, temporal dynamics, reproducibility, model complexity, success criteria, and infrastructure. We examine how regulatory frameworks (GDPR Article 22, EU AI Act 2024/1689, FDA GMLP) render common competition techniques legally non-compliant in finance and healthcare. We document the reproducibility crisis — 42% code inclusion at NeurIPS, no empirical study measuring whether the 2021 checklist actually helped — and trace its roots to practices normalized in competition culture. We report a compensation paradox where competition-optimized profiles earn 40–60% less than production-oriented engineers at equivalent seniority. Drawing on published interviews with Kaggle Grandmasters (Bhatia, Guschin, Tunguz, Babushkin), we propose a taxonomy of transferable and non-transferable skills, and argue that ML education programs need to stop treating competitions as sufficient preparation for industry.</span></p> <p><strong><span lang="EN-US">Keywords:</span></strong><span lang="EN-US"> machine learning, Kaggle, MLOps, production ML, reproducibility, AI regulation, ML education</span></p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_18721926 |
| institution | Zenodo |
| language | eng |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Competition ML vs. Production ML: A Systematic Gap Analysis Emelianov, Ilya Machine learning Reproducibility of Results <p><span lang="EN-US">Kaggle has 23 million registered accounts. Only 612 of those users have ever earned Grandmaster status. Meanwhile, a Google L3 engineer — entry level — out-earns the 90th percentile of "Kaggle Master" job postings by a wide margin. These facts describe the same industry but fundamentally different skill markets. This paper presents a structured gap analysis between competition ML and production ML across six dimensions: metric choice, temporal dynamics, reproducibility, model complexity, success criteria, and infrastructure. We examine how regulatory frameworks (GDPR Article 22, EU AI Act 2024/1689, FDA GMLP) render common competition techniques legally non-compliant in finance and healthcare. We document the reproducibility crisis — 42% code inclusion at NeurIPS, no empirical study measuring whether the 2021 checklist actually helped — and trace its roots to practices normalized in competition culture. We report a compensation paradox where competition-optimized profiles earn 40–60% less than production-oriented engineers at equivalent seniority. Drawing on published interviews with Kaggle Grandmasters (Bhatia, Guschin, Tunguz, Babushkin), we propose a taxonomy of transferable and non-transferable skills, and argue that ML education programs need to stop treating competitions as sufficient preparation for industry.</span></p> <p><strong><span lang="EN-US">Keywords:</span></strong><span lang="EN-US"> machine learning, Kaggle, MLOps, production ML, reproducibility, AI regulation, ML education</span></p> |
| title | Competition ML vs. Production ML: A Systematic Gap Analysis |
| topic | Machine learning Reproducibility of Results |
| url | https://doi.org/10.5281/zenodo.18721926 |