Hands-On One-shot Learning with Python

Image generation

The raw binary character image, , is generated with a stochastic rendering function, which maps the stroke trajectory with grayscale ink. This is achieved by assigning independent Bernoulli probabilities to each pixel.

BPL is a highly intuitive model that can model concepts with simple programs under the Bayesian framework. The parameters of the probability distributions are learned from the training data. On single-shot computer vision tasks of classification and generation, the model's performance is at par with human-level performance, with data requirements much less than some recent deep learning models. The probabilistic ...