Algorithmic stability via ensembling
Averaging-based ensembles get a general stability bound tied to a covariance operator.
Barber and Samworth give a framework for measuring how much an ensemble resists data perturbations. The main guarantee depends on the norm of a covariance operator describing the ensembling process. They say the framework covers several practical perturbation examples and can produce sharper guarantees than privacy-based arguments. ArXiv · AI/CL/LG's note
Barber and Samworth give a framework for measuring how much an ensemble resists data perturbations. The main guarantee depends on the norm of a covariance operator describing the ensembling process. They say the framework covers several practical perturbation examples and can produce sharper guarantees than privacy-based arguments. ArXiv · AI/CL/LG's note
score 4