Rules

This document outlines the official rules for three contests of the International Olympiad in Artificial

Intelligence (IOAI) 2025:

• Individual Contest
• Team Challenge
• GAITE Contest

The rules may be updated to address omissions, inconsistencies, or new information prior to the event, but they will not change substantially in format.

Team Leaders are responsible for ensuring that all team members fully understand and comply with these rules.

1. General Requirements

1.1. Code of Conduct

OATH. We, the participants of this International AI Olympiad, solemnly swear to compete responsibly and honestly, embracing the spirit of fair play and collaborating ethically with AI. For the glory of humankind and the honor of our teams, we swear!

Respect. All individuals involved are expected to show respect and consideration; to create a welcoming environment that celebrates diversity and fosters mutual understanding.

Integrity. Participants must engage in fair play, avoid deception, and maintain the spirit of honest contest.

Confidentiality. Sensitive information, especially regarding problem sets and solutions, must be safeguarded to ensure the integrity of the contest.

Professionalism. Interactions should be characterized by formality and decorum, reflecting the serious nature of the academic pursuit and respect for fellow competitors and officials.

Safety and Welfare. The IOAI is committed to providing a secure environment. Any behavior that endangers the emotional or physical well–being of participants is strictly prohibited.

Compliance with Rules. Understanding and adhering to the official rules are obligatory. This includes guidelines for the contest structure, submission deadlines, and conduct during events.

Reporting Misconduct. Transparency is crucial. Any unethical behavior or rule violations observed should be promptly reported to ensure fairness and accountability.

Enforcement. The IOAI will enforce this code rigorously. Infractions can lead to sanctions, including but not limited to disqualification and barring from future participation.

2. Individual Contest

During the Individual Contest, all contestants from each country use different computers, must not communicate during the contest, and their results will be computed individually. It is important to know that medal allocations are only based on results of the Individual Contest. The tasks of the Individual Contest are designed in accordance with the IOAI Syllabus.

2.1. Contest Schedule

• At–home contest: There are 3 problems provided to each team about one month before IOAI 2025. These problems are for educational purposes, and the results of the at–home contest do not affect the final results of IOAI 2025.

• Practice session: A 2-hour practice session will be held before Contest 1 to help contestants familiarize themselves with the contest hall and the contest system. This has no impact on the scores and medals.

• Individual Contest 1: There are 3 tasks in this contest which are an extension and continuation of the at–home contest tasks. Contestants will have six hours to solve these problems.

• Individual Contest 2: There are 2 or 3 tasks in this contest which are novel and different from the at–home tasks. Contestants have six hours to solve these problems.

2.2. Contest Environment

• All contestants will have access to identical local machines during the contest, though minor technical differences may exist.

• Contest tasks may include, but are not limited to: writing code, fitting models on training data, running inference with trained models on test data.

• All contestants will have access to identical GPU resources during the contest, with a minimum RAM of 24 GB.

• The Bohrium (https://bohrium.dp.tech/) platform will be used to access the task statements and datasets, submit solutions, and see the scores.

• A web–based Jupyter notebook environment will be made available within Bohrium – this is how GPUs will be accessible.

• Additional common editors will be installed locally: VSCode, Vim, PyCharm.

• The contestants will have access to GPT-4o (or a newer version, announced before the event), as integrated within Bohrium. The use of any other LLM–based chat, copilot, or AI–code assistant tools is prohibited.

• Access to the internet will be restricted to the following list of websites (Contestants must not log in to these websites, post questions, or use them to communicate with anyone outside the contest hall):

◦ scikit–learn.org

◦ pytorch.org

◦ stackoverflow.com◦ huggingface.co

◦ numpy.org

◦ github.com

◦ python.org

◦ pypi.org

◦ A search engine (result access restricted to the list above)

◦ A translation website

• On a per–task basis, contestants may be asked to submit their code, trained models, and/or model predictions. There will be no additional time for model training after the end of the contest.

• By default, the use of pre–trained models and additional data is not allowed, unless otherwise specified in the task statement.

• Please use the following form to submit requests for:

◦ Additional editors to be installed on the contestants’ machines

◦ Additional websites to be added to the white list

◦ An online platform for translation from English to the team’ s language(s) translation Form submissions will be accepted until June 30, 2025.

• The final list of available websites, editors, tools, search engine and translation system will be published before the event.

2.3. Translation

• To provide translated tasks to the contestants, a translation session will start two hours before each contest. During the translation session, TLs will receive the tasks in English and machine–translated versions in different languages. The TLs must check and correct machine translations if there is any

issue.

• If several teams have contestants who speak in the same language, the TLs of those countries may collaborate during the translation session.

• All translations should match the original English version. Providing extra information or hints in translations is prohibited

• During the translation session, the TLs should treat the tasks as highly confidential. They cannot communicate with anybody outside the hall of the translation session about the tasks or send them to anybody else or post them online.

• During each contest, all students will have access to the original English version of the tasks as well as all translations.

• All translations will be made public after the contest on IOAI official website. If any violation of above rules (e.g. providing any hint or extra information) is reported after the contest, the team who violated the rules might face penalties.

2.4. Scoring

• Each task may consist of multiple subtasks, with each subtask contributing a portion of the total score.

• Maximum Score: The maximum score for each task is 100 points.

• Final Score Calculation: The final score for a task is the sum of the scores of its subtasks, rounded to two decimal places.

• Baseline Solution: A baseline solution, typically provided in the task statement, represents a simple or trivial solution that helps establish the minimum acceptable performance for the task.

• Normalized Score Calculation: For each task, the score of a submission is normalized using the

following formula:

Norm_Score = (Submission_Score – Min_Score) / (Max_Score – Min_Score) * 100

• Submission_Score = The raw score obtained for the submission (e.g., accuracy, AUC, F1-score, or any other metric specified in the task statement).

• Min_Score = The score of the baseline solution.

• Max_Score = The maximum score, determined by the greater of the following two values:

◦ 0.9 × SC_Solution, where SC_Solution is the score of an (almost) optimal solution written by the Scientific Committee, or

◦ Max_Submission, the highest unnormalized score achieved by any contestant’ s submission during the contest.

• Example: Assume a task is evaluated based on accuracy, and the raw accuracy score of a contestant‘s submission is 85%. The baseline solution has an accuracy score of 60%, and the optimal solution from the Scientific Committee has an accuracy score of 95%.

◦ Min_Score = 60%

◦ SC_Solution = 95%

◦ Max_Submission = 90% (let’s assume another contestant achieved 90%)

◦ Max_Score = max(0.9 × 95%, 90%) = 90%

◦ Norm_Score = (85% – 60%) / (90% – 60%) × 100 = (25%) / (30%) × 100 = 83.33

◦ Thus, the normalized score for this submission would be 83.33.

2.5. Feedback (Scoreboard)

• Contestants will receive provisional feedback for each submission, showing the score per subtask and indicating the result for each test case.

• During the contest, each contestant will have access only to their own scores during the contest. They cannot see the scores or rankings of the other contestants. But they can see the Min_Score (score of the baseline solution) and the Max_Submission (the highest unnormalized score achieved by any other contestant at each time of the contest).

• There is a live–scoreboard publicly available during the contest for the people outside the contest hall. Contestants are not allowed to see it during the contest.

• Right after the contest, contestants can see the scoreboard. The results provided in the scoreboard are provisional and subject to change. The final scoreboard will be published after the appeals review session by the International Scientific Committee (ISC).

2.6. Quarantine Rules

• From the start of the translation session until the end of each contest, all contestants are under quarantine.

• During this period, contestants cannot use cell phones, laptops, or other communication devices. Volunteers must remain with contestants to ensure compliance.

• All communication between contestants and others (e.g., team leaders) regarding contest tasks is strictly prohibited. Violations may result in disqualification.

• Contestants must give all communication devices (cell phones, laptops, etc.) to their team volunteers by 10 PM on the evening before each contest. These items will be returned after the contest concludes for all contestants.

2.7. Supplies

• Provided Supplies: Blank paper, writing tools, Clarification Request Forms, snacks, and water will be provided in the contest hall.

• Approved Items: Writing utensils, small mascots, non–electronic earplugs, ID badges, snacks, and water are approved. Contestants may ask the Scientific Committee to bring their own keyboard or mouse to the contest hall. This request should be made during the practice session. Bringing an external monitor is not allowed.

• Prohibited Items: Contestants may not bring items such as electronic devices (computers, phones, earphones, calculators, any sort of communication or Bluetooth–enabled items), books, manuals, data storage media, or any items that could store or transmit data.

• Medical items (e.g. tablets, glucometers) are allowed but must be approved by the Scientific Committee before being brought into the contest hall. These and other special need requests can be made during the practice session. If any medical device requires a Bluetooth connection to software on a cell phone, the cell phone should be kept by a volunteer in the contest hall who is previously educated on what to do in case of any abnormal situation. If there are any other situations, the Scientific Committee should be informed before the practice session.

2.8. Starting the Contest

• Contestants must be seated at least 5 minutes before the start of the contest.

• Contestants should not touch the workstation or any tools until instructed by the organizers.

2.9. Clarification Requests

• Clarifications regarding task details, rules, or grading may be submitted to the Scientific Committee via the contest system or written forms.

• The Scientific Committee will respond to clarification requests with clear Yes/No answers or further guidance when necessary.

• The Scientific Committee may deny responding to a question that is ambiguous or unclear.

• Any non–trivial and substantial answers will be broadcasted to all contestants.

2.10. Assistance Requests

• If a contestant faces any technical issues (e.g., computer, network problems), they should raise a colored card to request assistance.

• Assistance staff will address such issues but will not answer questions regarding contest tasks.

2.11. Ending the Contest

• Contestants will receive three warnings at 15, 5, and 1 minute before the end of the contest.

• At the end of the contest, all contestants must immediately stop working and await instructions to leave their desks.

• In case of a substantial technical problem, contestants may be given extra time, on a case–by–case basis.

2.12. Cheating and Violations

• Contestants are prohibited from engaging in any form of cheating, including tampering with the contest system, communicating with team leaders or deputy leaders during the quarantine period, direct or indirect communication with other contestants or people outside the contest hall during the contest, bringing prohibited items into the contest hall, or attempting unauthorized access to test data.

• The screen of each contestant’s computer may be recorded or remotely proctored.

• Assistance staff will log any suspected cheating or rule violations observed during the contest, including discussions they had with or warnings given to contestants. This information will be reviewed by the Scientific Committee.

• Violating any of these rules may result in disqualification or other penalties.

2.13. Appeal Process

• Results Embargo period: The Scientific Committee may share each contestant’s preliminary results with their team leader before the official results are announced. Team leaders are not allowed to share this information with contestants.

• Team leaders may file an appeal regarding any perceived grading issues by submitting an Appeal Form to the Scientific Committee.

• The final score for any task may change due to an accepted appeal. Appeals may result in a higher or lower score, but repeated appeals for the same issue will not be accepted.

3. Team Challenge

• Team Challenge is an official part of IOAI in which teams, rather than individual contestants, participate in a creative and AI–oriented challenge.

• The style of the Team Challenge changes every year. During the IOAI 2024 Team Challenge (formerly known as “Practical Round”), teams were asked to use existing AI tools and platforms for an exciting task: designing the album cover and music video for the next hit song of a Bulgarian pop star. The style of the Team Challenge for IOAI 2025 will be announced later.

• During the Team Challenge, all contestants of each team will be seated together and can communicate and cooperate.

• The number of computers that a team can use during the Team Challenge and the websites and tools allowed to be used are task–specific and will be announced before the Team Challenge.

• The tasks of the Team Challenge are provided in English, and there is no translation session for the Team Challenge. The contestants can use the same web–based translation system of the Individual Contest during the Team Challenge.

• During the Team Challenge, the contestants of each team cannot communicate with people outside the contest hall.

• Approved and Prohibited items in the Team Challenge are similar to those of the Individual Contest.

• The scoring for the Team Challenge is task–specific and will be announced in the task statements.

• The teams who achieve the topmost results of the Team Challenge will receive Team Challenge awards during the closing ceremony of IOAI.

• The duration of the Team Challenge depends on the style of running it and might take a few hours to a full day.

4. GAITE Contest

• Global AI Talent Empowerment (GAITE) Contest is a special simplified version of the Individual Contest. To participate, teams must pre–register in GAITE (https://ioai–official.org/gaite).

• The GAITE Contest is very similar to the Individual Contest (described in section 2 above) and consists of two days, 6 hours each, which run concurrently with Individual Contests 1 and 2.

• The tasks of GAITE Contest Day 1 are similar to the tasks of Individual Contest 1. The main difference is that GAITE contestants might ask for hints. The details of scoring, given hints, will be provided before the contest.

• The tasks of GAITE Contest Day 2 are simpler than the tasks of Individual Contest 2. Furthermore, the GAITE contestants might ask for hints.

• There is a separate scoreboard for the GAITE Contest.

• The contestants who achieve the top results of the GAITE Contest will receive awards (but not medals) during the closing ceremony.If you have any questions about the contest rules, please contact the IOAI International Scientific Committee (ISC) by sending an email to isc@ioai–official.org.