The H2O Driverless AI platform enables the following elements of AutoML:
- Support for NVIDIA GPUs—AI models are exploding in complexity, and automated data transformation and deep learning require massive compute power and scalability. H2O Driverless AI supports the latest NVIDIA GPUs to accelerate feature engineering and training of neural networks. NVIDIA’s Multi-Instance GPU (MIG) feature can be used to partition the GPUs, increase overall GPU utilization, and support several types of use cases and deployments with guaranteed quality of service.
- Integrated catalog of recipes and models—H2O Driverless AI offers a rich catalog of AI models, transformers, and scorers for automatic feature engineering and model building.
- Machine learning and deep learning—H2O Driverless AI includes leading open-source transformers, embeddings, and frameworks for machine learning and deep learning techniques to handle various data science use cases. With H2O Driverless AI, you can automatically build models for Independent and Identically Distributed (IID) data, images, text, and more. For example, H2O Driverless AI includes TensorFlow CNNs for image modeling and NLP libraries from PyTorch, including BERT and other state-of-the-art techniques.
- Machine Learning Interpretability (MLI)—H2O Driverless AI provides robust explainability and fairness analysis for machine learning models and helps explore and demystify modeling results. It includes straightforward disparate impact analysis to test for model bias and provides reason codes for every prediction. Maximum transparency and minimal disparate impact are crucial differentiators if you must justify your models to business stakeholders and regulators.
- Automatic model documentation (AutoDoc)—Data scientists must document the data, algorithms, and processes used to create machine learning models for business users and regulators. H2O Driverless AI automatic model documentation relieves you from the time-consuming task of recording and summarizing your workflow while building machine learning models. The documentation includes details about the data used, the validation schema selected, model and feature tuning, MLI, and the final model created. AutoDoc saves data scientists time and removes tedious work so that they can spend more time practicing data science and drive more value for the business.
- Bring-Your-Own Recipes—Experienced data scientists can easily extend H2O Driverless AI with customizations that run within the H2O Driverless AI platform, including data preparation, models, transformers, and scorers. These customizations, called recipes, are Python code snippets that can be uploaded into H2O Driverless AI at runtime, like plugins. H2O Driverless AI can consume recipes with multiple convenient options: uploading from a local machine, consuming from published code in a source control hub (Bitbucket) and linking to a recipe raw code. You can check the GitHub repository for the available and optimized H2O.ai recipes. During training of a supervised machine learning modeling pipeline, H2O Driverless AI can use these recipes as building blocks with or instead of all integrated code pieces. They are used in the automatic machine learning optimization process, eventually creating the winning model. Data science teams can develop customizations specific to their use-cases, industry, or business.