AIcrowdGeneral MLRegistration closed · In progress
AIcrowdARC White-Box Estimation Challenge 2026
Official pageOpen the competition on AIcrowd↗
- Effort
- 40–100 hours
- Difficulty
- Research
- Compute
- CPU only
- Scoring
- Automatic
About the task
Full rules ↗The organizer's text, in Chinese; there is no English translation yet.
Alignment Research Center与AIcrowd举办的白盒估计挑战。给定随机ReLU网络，估计标准正态输入下的神经元激活期望。第二阶段为1024宽、16层MLP，每网络预算2^41 FLOPs；以预算调整后的末层MSE排名，最终奖金采用私有重测。报名与组队已于2026年10月2日23:59 UTC截止，最终提交截止10月17日23:59 UTC；10月24日论文说明和10月31日赛期结束不延长提交。须年满18且达到当地成年年龄，资格与奖金限制以官方条款为准。
- Scoring
- Budget-adjusted final-layer MSE (lower is better)
- Teams
- Up to 5 per team
- Host
- Alignment Research Center
Is it for you?
Prepare before joining
Beginner fit12/100
Learning value91/100
The editors' notes, in Chinese; there is no English translation yet.
Who it suits
适合熟悉概率、数值计算和神经网络的研究者；当前仅已报名选手还能提交，新访客可以学习公开 starter kit。
What makes it hard
需要在严格计量的 FLOP 预算中推导并实现高精度估计，同时满足 Python/flopscope 计算限制和私有测试泛化要求。
Getting started
Skills to bring
- Python
- Probability and statistics
- Linear algebra
- Numerical methods
- Neural networks
- Monte Carlo estimation
We haven't pulled tutorials, data or baselines from the official page yet; the task and the data are there.
First competition? Read No GPU? Pick Competitions That Still Teach You Good Habits