We propose a large-scale community benchmark and competition to catalyze research on biologically inspired credit assignment algorithms. This project originates from the ARNI Biological Learning Working Group and addresses the reliance of modern AI on back propagation and human supervision. Our goal is to develop a testbed for algorithms that utilize hyper-modal data—temporally aligned streams of vision, audio, spikes, and sensors—to learn rich representations without supervision, mirroring the learn- ing dynamics of biological systems. This project will deliver a standardized open-source benchmark, a leaderboard competition, and a suite of “no-backprop” baselines, fostering collaborative efforts between machine learning, neuroscience, and cognitive science commu-
nities.
