If you’re figuring out how to prepare for the IMLC Pre-Final Round, try a timed set of machine learning problems you haven’t seen before. A familiar question lets you practice a method you already know. An unfamiliar one makes you decide what the question is asking, whether your first idea is working, and when another problem deserves your attention. The clock gives you a record of those decisions, not just a count of how many answers you finished.
Pick four problems from topics you’ve studied, but make sure you haven’t read the questions or their solutions. Ask a teacher or classmate to choose them if you tend to peek while browsing. The training and practice problems page is one place to look. Choose problems with different demands: one might ask you to calculate a prediction from a small table, while another asks you to explain why a model behaves differently on two data sets. Aim for unfamiliar questions, not four topics you’ve never studied.
Try a 40-minute session: use 3 minutes to scan, 30 minutes to solve, and 7 minutes to check. Those numbers are for this drill, not the official exam duration or format. Set a countdown timer and use a separate sheet to record elapsed minutes. Keep notes, solutions, and tutorials closed until the timer ends. If you run out of time mid-calculation, leave it unfinished. That’s useful evidence when you review.
During the scan, find one problem you can start with a specific action, such as filling in the first row of a table. Mark any problem you don’t yet know how to start. Don’t spend the scan trying to rank every question by difficulty. A clear starting point is enough to choose what to do first.
With 30 minutes of solving time and four problems, try a first check-in about 6 minutes into each first attempt. That leaves room in the solving block to return to something you skipped. The check-in isn’t a hard stop. Ask yourself: What did I just learn, and what will I do next? If you’ve computed two of three quantities and know how to get the last one, staying another minute or two may pay off. If you’ve reread the same sentence three times and still can’t name the quantity you need, switch problems.
Sometimes you can name a next step, but it’s too broad to help. “Try a different formula” doesn’t tell you much. “Check which examples belong in the denominator” gives you a short test with a clear end. Give yourself time for one such test if it looks promising. If it doesn’t resolve the issue, move on rather than starting a second open-ended search.
Before switching, leave a short note that makes it easy to return. For example: Found both groups’ error counts. Still need to check whether the question asks for the overall rate or the average of the two group rates. If you come back, you can restart at that decision instead of redoing the arithmetic. When the checking block begins, look first for things you can verify quickly, such as whether you answered every part or copied a number correctly.
Write down the elapsed minute when you start a problem, switch away, or return. Add a few words about why. Don’t try to record every calculation. A log that takes longer to write than it does to read is likely to distract you from the drill.
| Elapsed minutes | Work | Note |
|---|---|---|
| 0 to 3 | Scan the set | Start with B because the first step is clear. |
| 3 to 9 | Problem B | Finished the table and wrote the prediction. |
| 9 to 19 | Problem A | At minute 15, still unsure which examples to count. Kept rereading. |
| 19 to 25 | Problem C | Reached an answer but wasn’t sure about one assumption. |
| 25 to 30 | Problem D | Set up the calculation. Still needed to finish it. |
| 30 to 33 | Return to A | Listed the examples before counting and found the missing step. |
| 33 to 40 | Check answers | Checked B and C, then went back to D. |
That sample log shows why total minutes alone don’t tell the whole story. The first six minutes on A gave you a chance to understand the problem. The next four went to rereading without a new approach. The return visit, though shorter, produced a specific step. When you review your own log, mark the first moment you could have made a better decision, not just the moment you finally switched.
Before opening the solutions, give an unfinished problem a brief untimed retry. If you can finish it with a fresh read, compare that attempt with what you did under the clock. Maybe you missed a qualifier such as “on the test set,” or maybe listing the examples made the next step obvious. If you still can’t start and the solution uses an idea you don’t understand, put that idea on your study list. Moving on sooner helps you use time elsewhere, but it doesn’t replace learning the missing idea.
Check your completed answers too. A quick answer that uses the wrong denominator is a different pacing problem from a correct answer that took ten minutes. For the first, build in a quick check of what you’re dividing by. For the second, see whether those ten minutes were mostly productive. A long problem isn’t automatically a poor use of time.
The competition rounds page is the place to check how the Pre-Final Round fits into IMLC. For your next practice session, keep the same time budget and use a fresh set of similar difficulty. Change one habit based on your log. If rereading was the main drain, try this rule: after the second read, write the question in your own words or mark it for a return visit. Afterward, check whether you gave another problem an earlier start without leaving work you were close to finishing.
Should I move on as soon as my time check comes up?
No. Stay briefly if you can name a specific step that will bring you closer to an answer. If you’re repeating the same approach without learning anything new, leave a note and switch.
Can I reuse the same problems for my next timed session?
You can reuse them to practice a method, but pick fresh questions to test pacing. Remembering how a problem ends changes the decisions you make under the clock.
About the International Machine Learning Competition: an online machine learning competition for high school students from around the world. Qualification Round 2026: submissions are open until Sunday, 13 December 2026. How the rounds work