Braincraft challenge – 1000 neurons, 100 seconds, 10 runs, 2 choices, no reward
7 days ago
- #computational-neuroscience
- #simulation
- #neural-networks
- The BrainCraft Challenge aims to develop biologically inspired neural networks for simple decision tasks in a simulated environment.
- Participants must design a rate-based neural network controlling an agent to locate and reach an energy source in a 10x10 maze.
- The challenge consists of 5 tasks with increasing complexity, each active for 2 months.
- The agent has sensors for distance, wall collision, energy level, and a constant bias input.
- The neural model uses leaky rate units with specific equations for state and output updates.
- Training is limited to 100 seconds, and submissions must be self-contained with dependencies restricted to NumPy, SciPy, and Matplotlib.
- Performance is evaluated based on the mean distance traveled across 10 test runs.
- The challenge encourages open-source submissions and provides a leaderboard for tracking progress.