Oracle Releases MySQL HeatWave ML That Adds Powerful Machine Learning Capabilities to MySQL Applications

Integrating machine learning capabilities to MySQL systems is prohibitively difficult and time-consuming. The process involves extracting data from the database and into another system to construct and deploy machine learning models. As data flows around, this strategy produces silos for applying machine learning to application data and causes latency. This results in data leakage, making the database more open to security attacks. Moreover, existing machine learning (ML) solutions lack the ability to explain why the model developers build delivers specific predictions.

Recently, Oracle released MySQL HeatWave, the only MySQL cloud database service that supports in-database machine learning (ML). It automates the ML lifecycle and saves all trained models in the MySQL database, removing the need to migrate data or models to a machine learning tool or service. This decreases application complexity, saves costs, and increases data and model security. It produces a model with the best algorithm, features, and hyper-parameters for a specific data collection and application. 

The team released ML benchmarks based on various publicly available ML classification and regression datasets. Customers can now retrain their models more frequently and keep up with changes to data because training can now be done very efficiently and quickly with MySQL HeatWave. This also increases the accuracy of the models by keeping them up to date.

When compared to other cloud database systems, HeatWave ML has the following features:

• Automated Model Training: So far, other cloud database providers have integrated with external machine learning capabilities. This task requires considerable human input from developers throughout the ML training process. With HeatWave ML, all stages of developing a model are fully automated and do not require any developer participation. The resulting fine-tuned model t is more precise and requires no manual labor.

• Hyper-Parameter Tuning: The most time-consuming stage of ML model training is hyper-parameter tuning. For this, HeatWave ML uses a new gradient search-based reduction technique. This allows for concurrent execution of the hyper-parameter search without affecting model correctness. 

• HeatWave ML selects the optimum ML method for training using the concept of proxy models, which are basic models that exhibit the attributes of a full complicated model. Algorithm selection can be made quickly and accurately with the help of a simple proxy model. 

• Most cloud service providers use random data sampling to develop machine learning models without considering the peculiarities of the data distribution. On the other hand, HeatWave ML uses intelligent data sampling to sample a tiny percentage of the data during model training to increase performance. This results in sample data which contains all representative data points. 

• Feature Selection: It is useful in determining the characteristics of the training data that influence the resulting predictions. HeatWave ML can efficiently find the important features in a fresh data set using these acquired statistics and meta information from broad training data sets.

• Explanations for Models and Inferences: Prediction explainability refers to strategies for determining why a machine learning model generated a particular prediction. As part of its model training procedure, HeatWave ML incorporates both model and prediction explanations. As a result, all HeatWave ML models may provide model and inference explanations without training data at the time of inference explanation. Oracle has improved existing explanation techniques’ performance, interpretability, and quality. 

Oracle added new features to the MySQL HeatWave service in addition to machine learning capabilities. Customers may scale up or down their HeatWave cluster to any number of nodes in real-time, without downtime or read-only time, and without manually rebalancing the cluster. Further, data compression allows clients to handle twice as much data per node while lowering expenses by roughly half while retaining the same price-performance ratio. Finally, a new pause-and-resume feature allows users to save money by pausing HeatWave. When HeatWave is restarted, the data and statistics required by MySQL Autopilot are immediately reloaded.

Oracle’s MySQL HeatWave cloud service is one of the company’s fastest-growing cloud services. Customers are migrating from other cloud database providers to MySQL HeatWave, which has resulted in considerable performance gains and cost savings. 

Read the MySQL HeatWave ML technical white paper

Reference: https://www.oracle.com/news/announcement/mysql-heatwave-supports-in-database-machine-learning-2022-03-29/

Suggested