Amazon launches new tools for preparing training data in SageMaker
Amazon is introducing AI data preparation capabilities into SageMaker AI – synthetic data generation, dataset quality assessment, and incorporation of expert feedback via the Python SDK or the AWS AI-ML agent skill.
Amazon Web Services has expanded Amazon SageMaker AI with new AI data preparation capabilities designed for preparing training and evaluation data during model development. The tools make it possible to generate synthetic data, assess and improve the quality of existing datasets, and incorporate feedback from human experts into the process. Access is possible via the Python SDK API or via the AWS AI-ML agent skill, which, based on a natural-language description of the task, itself selects and runs the appropriate function.
According to the company, the individual capabilities can be used on their own or chained into a custom workflow – from automatic data labeling based on a small sample set (where uncertain cases go to human review) to a complete path from raw documents to a trained model. The tools integrate into the SageMaker Studio JupyterLab, Kiro, and local IDE environments. According to the company, source data, generated datasets, and labels all remain in the customer's Amazon S3 buckets in the chosen AWS region, and execution runs under an IAM role that the customer manages themselves.
As a usage example, the company cites Splunk (a subsidiary of Cisco), which, according to a statement from Yash Sheth, Director of Engineering for Splunk Agent Observability, uses synthetic data generation to create statistically representative datasets for GenAI evaluation models, doing so faster and with less need for manual data processing.
The new capabilities are generally available in the US East (N. Virginia) and US West (Oregon) regions.
Why it matters
According to the source, preparing quality training and evaluation data is often the slowest part of ML projects – due to a lack of labeled data, privacy rules, and quality issues (duplicates, noise in labeling). Automating synthetic data generation and dataset validation directly in SageMaker can reduce the time data scientists spend on manual data processing and make it easier to train smaller specialized models on quality data instead of relying on general-purpose large models.
Two audiences, two different impacts
What this means
For individuals
Data scientists and ML engineers working with SageMaker gain, directly in the notebook or IDE, a tool for generating and validating training data without having to assemble their own pipeline from multiple separate scripts and services.
For a business
According to the available example (Splunk), companies developing specialized models can speed up the preparation of evaluation and training datasets and reduce manual labeling costs, while data remains in their own S3 storage under the control of their access roles.
DevelopmentCheck the original
Event sources
clearly official source · 1 publisher, 0 independent. We count feeds from the same owner only once.