Search Results for "model-builder"

The supervised-reptile repository contains code associated with the paper “On First-Order Meta-Learning Algorithms”, which introduces Reptile, a meta-learning algorithm for learning model parameter initializations that adapt quickly to new tasks. The implementation here is aimed at supervised few-shot learning settings (e.g. Omniglot, Mini-ImageNet), not reinforcement learning, and includes scripts to run training and evaluation for few-shot classification. The fundamental idea is: sample a task, train on that task (inner loop), and then move the initialization parameters toward the adapted parameters (outer loop). ...

The purpose of the project is summarising effort from a number of analytic libraries, adding interactive web-based user interface and making a free open source solution for risk analytics and stress testing. Feb 8, 2012 Paul Glasserman's Importance Sampling and Tail Approximations as well as plain Monte Carlo have been implemented for for the widely used normal copula model of portfolio credit risk. The package includes source code, examples, spreadsheet with results and references to the papers.