Have a dataset and a meaningful problem that can be addressed using AI/ML methods? The Machine Learning Marathon (MLM26) is a semester-long AI/ML hackathon, where the UW–Madison community comes together to tackle real problems submitted by researchers and industry. Challenges run on Kaggle from September through December, with weekly in-person sprint sessions, expert-led workshops (e.g., intro to AWS/GCP), and cloud compute credits to support participating teams. Challenges span a range of skill levels, with suggested prerequisites listed for each. Submitting a challenge puts your dataset and problem in front of a motivated campus audience and Kaggle’s global practitioner community. The initial interest form is due by June 1.
New to the Marathon? Visit our companion site — ML+X Nexus — to explore past challenges and see what is possible.
Quick Links
Eligibility
Before submitting, confirm your challenge has the following:
- A clearly defined problem suitable for AI/ML methods — a specific prediction, classification, estimation, or detection task. Avoid open-ended “explore the dataset” goals.
- A measurable evaluation metric — traditional (accuracy, RMSE, F1, AUC) or task-specific approaches like LLM-based grading, retrieval quality metrics for RAG, or composite system-level scores. A held-out test set is required.
- Shareable data — uploadable to Kaggle by August 1. De-identified or synthetic versions are acceptable if they preserve the problem structure.
Selection criteria: The final challenge set is curated for diversity across application areas and AI/ML methods. We’ll communicate decisions clearly, and strong submissions that aren’t selected are encouraged to resubmit in future cycles.
Interest form due by June 1. See Timeline below for all deadlines. Have questions? Check the FAQ first — for anything not covered there, email endemann@wisc.edu.
Timeline
Submitting the interest form tentatively secures your spot. We will follow up on a rolling basis to give an initial go/no-go on your project concept. If your project gets the green light, the timeline below kicks in. Missing a deadline puts your spot at risk.
- June 1— Interest form deadline. Submit the interest form to secure your spot. We will review your concept and follow up.
- Early June — One-on-one check-in. Initial Kaggle draft started.
- Mid July — Majority of data cleaned and uploaded to Kaggle; metric defined and tested, even if still being refined.
- August 1 — Full draft complete: problem description, all data uploaded, metric finalized, sample submission ready.
- August 17 — Projects announced and participant registration opens.
- September – December — Marathon runs. Weekly sprint events, Wednesdays 4:30–6:30 pm.
August 1 is a firm deadline. Challenges submitted after this date cannot be guaranteed inclusion in MLM26. If you’re unsure whether your dataset will be ready in time, reach out early — we’d rather help you scope things down than miss the window.
Hosting Your Challenge on Kaggle
All challenges use Kaggle‘s free competition format. Your challenge is visible to Kaggle’s global community of millions of practitioners — not just UW–Madison participants — giving your problem and dataset broad exposure. You can start drafting right after submitting the interest form.
- Create a competition draft. Go to Kaggle → Create Competition and select Machine Learning Competition (not a dataset or notebook). Set an initial title and subtitle, then click “Create Competition” to start the draft.
- Add us as a collaborator for review. Under Settings → Collaborators, add Kaggle username qualiaMachine (Chris Endemann). Add whenever you’re ready for feedback, or by July 15 at the latest.
- Completed draft by August 1. Your draft should be substantially complete: problem description written, data uploaded, metric defined, and a sample submission file ready. Final revisions may be made up until Sept 1st.
Frequently Asked Questions
This is an accordion element with a series of buttons that open and close related content panels.
Do I need to be a machine learning expert to submit a challenge?
No. You need a real problem, a dataset, and a clear sense of what a “good answer” looks like. We can help with framing the ML problem and choosing a metric, but this needs to happen early — ideally before or shortly after you submit the interest form. Email endemann@wisc.edu if you’re unsure about framing.
My dataset isn't ready yet. Should I still fill out the interest form?
Yes — earlier is better. If there’s uncertainty about readiness by August 1, we’d rather know now so we can help scope the problem or suggest alternatives (e.g., a de-identified or synthetic version).
My data is sensitive or proprietary. Can I still participate?
Possibly. The data needs to be shareable on Kaggle with all participants by August 1. A de-identified or synthetic version is acceptable as long as it preserves the structure of the original problem. Data with unresolved IRB or proprietary constraints isn’t eligible.
Will my challenge automatically be included?
No — meeting the requirements is necessary but not sufficient. The final set is curated for diversity across application areas (health, environment, social science, engineering, language, industry) and AI/ML methods (vision, NLP, time series, tabular, anomaly detection, generative). Strong submissions may be deferred if similar topics are already represented. We’ll always communicate decisions clearly, and deferred projects are encouraged for future cycles.
Can I propose a non-standard evaluation metric (e.g., RAG or LLM-based scoring)?
Yes. Kaggle supports custom metrics, and we’ve seen strong interest in system-level evaluations beyond classic accuracy/RMSE.
Are multi-stage or two-phase competitions possible?
Unfortunately, no. Multi-stage competitions require Kaggle’s paid plan, which we don’t use. All Marathon challenges use Kaggle’s free prediction competition format: a single leaderboard scored against a held-out test set.
Do I have to advise or mentor teams during the Marathon?
No, but it’s strongly encouraged. Organizers who drop into sprint sessions (even occasionally) consistently see better outcomes from their teams.
What's the time commitment for a challenge organizer?
The bulk of the effort is upfront — preparing the dataset, defining the metric, and drafting the Kaggle page between June and August 1. During the Marathon itself, commitment is light unless you choose to advise: occasional check-ins, optional sprint visits, and availability for questions.
What should my Kaggle competition page include?
See last year’s challenges for examples. Your page should include: a plain-language overview of the problem, a description of the data (fields, size, format), the evaluation metric and why it was chosen, and any recommended prerequisites or starter resources. A starter notebook is a nice bonus but not required.
Can a company or industry lab submit a challenge?
Yes, but industry partners are required to sponsor ML+X to participate as challenge organizers. Sponsorship starts at $2,500 and includes the ML Marathon challenge as one of several benefits. See the ML+X sponsorship page for details or email endemann@wisc.edu to get started.