Erwan Fagnou

Scaling Direct Feedback Learning with Jacobian Alignment Guarantees

Globally 1-Lipschitz Attention Without Sequence-Length Dependence

Trading Complexity for Expressivity: Theoretical Exploration of Linear and Causal Token Mixing Strategies

Accelerated training through iterative gradient propagation along the residual path

Backpropagation-Free Learning through Gradient Aligned Feedbacks

Activation Decay by Loss Smoothing to Enhance Generalization