Google researchers introduces Simula.

Google researchers have introduced Simula, a new method for generating synthetic data that promises to improve the quality and diversity of data for specialized AI applications.

In response to the growing demand for specialized AI models, Google has presented a new approach to synthetic data generation. In an article written by Tima R. Davidson and Hamza Harkos, the launch of Simula is described—a framework that allows for more accurate and diverse data generation.

Modern general-purpose AI models have succeeded due to the availability of large volumes of internet data. However, in specialized and confidential areas where data is scarce or unavailable, new methods of data acquisition are required. The simulation of synthetic data opens new possibilities for developing reliable AI models.

Simula addresses the challenges associated with traditional data generation. Unlike manual prompts and evolutionary algorithms, Simula employs a 'reasoning-based methodology' that allows for the complete creation of datasets, independent of time constraints. This makes the approach more flexible and autonomous.

The core idea of Simula is to break down the data generation process into several controlled axes: global and local diversity, complexity, and quality. Global diversity ensures coverage of a wide range of topics rather than a narrow focus. This is achieved through the construction of hierarchical taxonomies, which allows for better data management.

Local diversity prevents redundancy among recurring concepts, while 'complexification' finds ways to increase the complexity of scenarios, making them more realistic. Quality control of data, based on bilateral assessment, eliminates the need for manual verification, ensuring high accuracy of labels.

Thus, with the help of Simula, researchers hope not only to improve the process of synthetic data generation but also to expand the boundaries of AI application in various fields. This work represents a significant step forward in enhancing the accessibility and quality of data necessary for the safe and effective functioning of AI.