Gradient-based inference of abstract task representations for generalization in neural networks

Open in new window