Frequently Asked Questions

Do I need prior ML experience?

Some. You should know Python basics, machine learning fundamentals, and enough GitHub to collaborate. Past participants have ranged from near-beginners to AI researchers. Challenges list their own prerequisites on the challenge page, so you can pick one that fits where you are. If you need to get up to speed, work through the relevant self-paced materials before kickoff – Intro Python, Collaborating with GitHub Desktop, and Intro to Machine Learning.

This isn’t a guided course. Expect to search a lot, break things, and learn from the people next to you. Advisors are there when you’re genuinely stuck; your teammates are your first resource.

What’s the time commitment?

Weekly Wednesday evening sessions from September through December, most running 4:30-7:30PM, plus async work with your team between sessions. A few land on other weekdays; see the schedule.

Participants are expected to attend nearly all scheduled events, as laid out in the rules. Occasional absences for exams or unavoidable conflicts are understandable, but your teammates are counting on you. Arriving after 5PM is fine as long as your team is okay with it.

This is a semester-long commitment, not a weekend hackathon. If your Wednesday evenings are already spoken for, this probably isn’t your year.

What does the ticket cost, and can I get a refund?

$150, priced to break even on costs. A good chunk of that is food: most weekly sessions run past 6:30PM and come with a catered meal, plus light refreshments at the 9/2 mixer.

We also encourage researchers and lab affiliates to ask their advisors/departments if they are willing to cover the costs, or pay with a department funding string. This is professional development, and plenty of participants get it covered.

Refunds are available through 9/3 (following the team mixer event), minus a $25 processing fee from Eventbrite. Email endemann@wisc.edu by 9/3 at 11:59pm to request a refund.

What compute do I get?

Roughly $50 in AWS credits per participant, plus a workshop on how to use it well (SageMaker and Bedrock, 10/7). Some people end up with more – it depends on how heavily the shared pool gets used across the season.

We also host LLMs on campus infrastructure and share them with projects that need them.

That all said, we encourage everyone to start on their laptops. Get direction on the problem first, then scale up. Cloud compute is easier to spend than it is to spend well.

Are there prizes?

This is a community learning event, not a cash competition. The leaderboards are there to give you something concrete to push against, but the atmosphere is far more collaborative than competitive. Teams share solutions across the season, building off others’ progress with appropriate credit.

Recognition, though, yes: the winning team in each challenge gives a 10-minute method spotlight at the closing celebration on 12/9.

How do you decide who gets a seat?

Applications are open to anyone affiliated with UW–Madison with an interest in applied ML/AI — undergrads, grad students, postdocs, faculty, staff, and recent alumni. Past participants have ranged from complete beginners to AI experts.

We expect roughly twice as many applicants as seats, so admission is by application rather than first-come. The following factors are considered when reviewing applications:

• Fit with the challenges you ranked, including the prerequisites they list.
• A mix of disciplines, roles, and experience levels, both within each challenge and across the cohort. Teams work better when people aren’t all coming from the same background.
• What you tell us about how you’d apply ML/AI in your own work.
• Making sure every challenge has enough people to field at least one team. This is why the flexibility checkbox helps, and why a few people get their second or third choice.
• Affiliation with research activities on campus: while all are welcome to apply, our priority is to advance the research mission of the university. Some priority is given towards graduate students and researchers.
• First-timers get a slight edge when everything else is equal.

What happens if I’m waitlisted?

Everyone who applies starts on the waitlist. We start reviewing on 8/24 and admit on a rolling basis from there. There’s no application deadline, but we expect twice as many applicants as seats. Apply by 8/24 to be included in the first review; after that we take applications as they arrive. If you don’t get a seat, you can still work the challenges. See below.

Each acceptance comes with a payment link good for 48 hours. Unclaimed links put the seat back in the pool, so we backfill from the waitlist in rapid rounds. If you haven’t received a seat by 9/7, we couldn’t fit you this year.

Can I work on a challenge without being in the Marathon?

Yes. All MLM26 challenges are public on Kaggle, and the challenge pages list prerequisites, data, resources, and the Kaggle link for each one (as they go live leading up to Showcase on 8/19). You don’t need a seat, and you don’t need to have applied.

You can follow along with the MLM participants’ progress on Kaggle. They post solutions and questions to the Kaggle discussion boards for their challenge, and resources go up throughout the semester tied to specific projects. Contribute your own if you have something worth sharing.

Asynchronous participation is self-organized and unstaffed. No advisors, no compute credits, no workshops, and no access to the Wednesday sessions at Discovery. In-person capacity is capped at 75 and we can’t add to it. If you’re working a challenge this way and want the full program, apply next year. You’ll be coming in having already worked one of these problems, which is a good place to start from.

Do I need a team before I apply?

No. The application asks whether you already plan to work with specific people so we can attempt to keep you together (assuming all applicants meet prerequisites and a challenge has capacity).

Teams come together at the 9/2 mixer, and plenty of people arrive without one.

How is my challenge assigned, and what if I don’t get my first choice?

The application asks you to rank your top three challenges. Your assignment comes with your acceptance email, before you pay. We balance assignments so every challenge has enough people to field at least one team. That means some people land on their second or third choice.

How are teams formed, and how big are they?

Teams are 3-6 people (see the rules) and form at the team formation mixer on 9/2, 4:30-6PM in the Discovery Atrium. The mixer is organized by challenge, so you’ll be meeting the people assigned to the same challenge as you, not the whole room all at once.

If you already applied as a group of 3 or more, you just confirm your roster at the mixer and you’re set.

Most teams leave the mixer settled. Teams should be finalized by the kickoff on 9/9.

Can I still register if I can’t make the mixer?

Yes. Teams form at the 9/2 mixer, so plan on being there if you can. If you have a conflict, post in your challenge’s Slack channel by 8/31 – or within 24 hours of receiving your ticket, if you’re admitted after that – and work with the channel to find a team asynchronously. The channel name and the Slack invite come with your ticket confirmation. Tell people your background in ML and what you’re hoping to get out of the challenge. If you don’t post and don’t show up, we’ll release your seat to the waitlist.

What if my team loses someone mid-season?

Your team keeps going. The 3-6 range is enforced when teams form, not for the rest of the semester. In rare cases, team mergers may be requested. These requests are typically approved so long as both teams are on board.