Signature Kernel
A similarity measure for time-series paths derived from combining path signature theory with kernel methods. It defines a kernel between two sequential paths and exhibits robustness to irregular sampling. The signature kernel can serve as a non-adversarial training objective for generative models, replacing discriminator-based approaches in certain sequence generation tasks.
Timeline (last 90 days)
HyperNSDE uses a non-adversarial signature kernel objective function for optimization, avoiding the training instability issues of GAN-type models.
The signature kernel combines path signature theory with kernel methods, and its training objective is non-adversarial, eliminating problems such as mode collapse and training oscillation.
All Facts (2)
HyperNSDE uses a non-adversarial signature kernel objective function for optimization, avoiding the training instability issues of GAN-type models.
50%UnverifiedThe signature kernel combines path signature theory with kernel methods, and its training objective is non-adversarial, eliminating problems such as mode collapse and training oscillation.
50%