Learning without Global Backpropagation via Synergistic Information Distillation

Chenhao Ye, Ming Tang

Backpropagation (BP), while foundational to deep learning, imposes two critical scalability bottlenecks: update locking, where network modules remain idle until the entire backward…