Publications

Oscillatory Predictive Learning (OPL) combines Artificial Kuramoto Oscillatory Neurons with predictive self-supervised pretraining (X-PhiNet): adversarial robustness emerges from architectural and representation-learning inductive biases, without adversarial training or test-time purification.

Combining Artificial Kuramoto Oscillatory Neurons (AKOrN) with PhiNet, a hippocampus-inspired non-contrastive self-supervised method, to obtain encoders that are more robust to adversarial perturbations — without adversarial training.