Launched "GeneFlow" beta version, a platform for managing AI learning history and behavior on the blockchain: Ensures transparency of AI learning history and supports project management

Couger Inc (Headquartered in Shibuya-ku, Tokyo; Atsushi Ishii, CEO; hereinafter "Couger") is now offering a limited beta version of "GeneFlow," an AI model management platform that allows users to manage AI models and verify their operation on the blockchain.

“GeneFlow” reduces the cost of complex AI model management, which is essential for AI projects, and improves the reliability of AI models by proving their learning history. The use of blockchain technology in the management platform increases transparency of the learning history and accuracy of AI models not only internally, but also to third parties, such as external companies.

Couger is the first company to use the XAI, "Explainable AI," project, the “GeneFlow '' alpha version was released in 2018. Since then, we have been building and developing the platform through presentations at Stanford University and experiments with Chubu University.

The sophistication and complexity of AI technologies, including deep learning, continues to increase, and how to make them accountable is a major challenge for society. Couger will use its past activities to focus on AI management and contribute to accelerating the implementation of technology in society.

Data Management and Model Transparency Needed for AI Development

The accuracy of AI models varies greatly depending on the content of training data, algorithms, and their tuning. Even an AI that boasts 99% performance cannot be put to practical use if it responds only to specific situations and lacks the generalization ability in the real world. For this reason, AI development sites are repeatedly learning, verifying, and improving. This process involves an enormous amount of data management work.

In addition, as AI technology becomes part of the social infrastructure, the importance of investigating the causes of problems is increasing. However, AI is characterized by the fact that it is difficult to clearly define normal operation. Therefore, in order to create AI whose reliability can be verified, it is necessary to provide transparency and objectivity in learning and verification records. To this end, we provide an infrastructure environment that ensures transparency as a technical mechanism by using blockchain technology to record how AI models are learned and verified, and with what accuracy.

“GeneFlow” was developed against the backdrop of this need to reduce administrative costs and provide reliability in the validation of AI models.

Functional Features of “GeneFlow”

Management of AI development operations

AI training data, training algorithms, generated AI models, and their execution history can be recorded and managed against a distributed blockchain environment.

Management of arbitrary AI models

Existing AI models that have not been created in “GeneFlow” can be uploaded to record and manage the execution results of AI models on the blockchain. This enables management of the accuracy and execution history of any AI model.

Intended Recipients

  • Companies developing AI in-house
  • Companies developing AI through collaboration with external companies
  • Research institutes and research review organizations
  • AI technology community
Recommended articles