![]() ![]() Instead, we’ll simply tell it what the forward pass of a model looks like,Īnd the synthesis engine will figure out how to train it. Will be declarative-we won’t need to teach the synthesizer anything There’s already some research interest in verifying learned models īy doing synthesis we get an even richer set of tools. (e.g., that the outputs are always in a reasonable range). We’ll be able to prove properties of learned models Will automatically give us tools to do verification of models. Building the infrastructure for synthesizing machine learning models.Which complete the weights for a single fixed topology,Īnd rely on an external layer (e.g., grid search) to search for good shapes. This is in contrast to most traditional learning algorithms, ![]() The best weights in a fixed model (e.g., a neural network) but also altering the shape of the model We can use it to do superoptimization of the trained model-discovering not just
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |