💡 What is this?
The MindRL Hub Modeling Challenge is a cognitive modeling competition designed to redefine how we evaluate AI-era models. We move beyond simple predictive accuracy to a tripartite framework: Prediction + Explanation + Robustness (Attack).
Our Core Philosophy
- Prediction ≠ Explanation: High accuracy doesn’t guarantee a correct mechanism.
- Testability: Explanations must be falsifiable and scientifically rigorous.
- Robustness: A truly “good” model should remain valid under adversarial challenge.
🛠 Modeling Approach
We encourage a wide range of methodologies, including but not limited to symbolic cognitive models (e.g. reinforcement learning models, bayesian models), neural networks, large language models and other hybrid models.
The goal is not the method itself, but whether it provides superior prediction and explanation supported by testable claims.
🏁 Two Tracks
We utilize a dual-track system where modeling and auditing are interactive processes:
- Generative Track: Build a predictive model and provide a verifiable “Interpretation Card”.
- Adversarial Track: Systematically audit model claims, looking for simpler alternative explanations or failure cases.
Methodological Diversity: We plan to admit a total of 12 teams. To foster a rich scientific comparison, the organizers will ensure the 12 selected teams represent a balanced variety of modeling approaches.
🔄 Submission & Evaluation
- Submission: Teams maintain their own repositories on GitHub and host larger model on HuggingFace.
- API Standards: Models must expose a standard interface for initialization, resetting, and updating internal states (e.g., Q-value or Bayesian belief updates) after observing actions and rewards.
- Automated Scoring: The committee maintains an automated pipeline for one-step-ahead prediction evaluation using metrics like negative log-likelihood or cross-entropy.
- Auditing: Adversarial teams focus on the Interpretation Card and provide rebuttals to model claims.
📅 Timeline (2026)
- May 17: Official Announcement & Launch
- May 29: Enrollment Opens
- June 15: Enrollment Deadline
- June 22: Notification of Acceptance (Discord Communication Open, Paper Sharing, Warmup)
- June 29: Phase 1 Starts (Task & Dataset Release, Generative Track)
- July 27: Phase 2 Starts (Adversarial Track)
- September 4: Presentation Day & Awards
🏆 Why Join?
We are not looking for a simple leaderboard. Our goal is to contribute to a standard for cognitive modeling in the AI era. You will gain:
- Deep experience in sequential decision-making.
- Exposure to systematic model auditing and failure-mode analysis.
- Publication Opportunities: Participants are encouraged to publish scientific papers based on the work and results developed during the competition.
👥 We Are Looking For
This is a pilot competition for researchers. We are looking for teams that:
- Are Technically Proficient: Official language is Python. Participants should be comfortable using GitHub, although GitHub proficiency itself will not be a major factor in ranking.
- Have Experience: At least one member should have prior experience in computational modeling.
🎓 Guidance
Advisors
Mentors
⚙️ Organizers
🔗 Links
- Info Session: Sign up here
- Community: Join our Discord
- Note: The community is open to all; specific channels are restricted to participants.
- Wiki: Detailed Documentation
- Email: rldmjc2023@gmail.com
✏️ Sign Up Now
Applications are open until June 15th, 2026
Submit Application →