Looking for critical review of an NN architecture (possible evaluation bias?) [D]

In three linesAmateur student seeks critical review of a custom neural network architecture (Directional Neural Network) he designed. The architecture outperforms standard MLPs on simple tasks, but the author suspects potential evaluation bias in his comparisons (initialization, optimizer, datasets). Shares a repository with reproducible code.

Summary generated by Claude — human-verified