Posted On: Dec 19, 2022
Today we announce the general availability of Renate, an open-source Python library for automatic model re-training. The library implements continual learning algorithms to train deep neural networks incrementally when new data becomes available.
Applications of machine learning require updating models as new batches of data become available. Repeatedly re-training deep neural network models from scratch is costly and fine-tuning them with the new data only will lead to a phenomenon called “catastrophic forgetting”. This means that the model will have good performance on the most recent data, but the performance will degrade on the older data. Renate provides algorithms that alleviate the problem of catastrophic forgetting and helps to automatize the re-training process.
With Renate, users run small scale continual learning experiments on their local machine or run large continual learning jobs using Amazon SageMaker. Renate also supports state-of-the-art hyperparameters tuning out-of-the-box, thanks to the integrations with SyneTune.
To learn about the library, checkout our Blog. To get started with Renate, you can consult the following resources: