Couger developed a system “GeneFlow” that secures the reliability of AI by recording the learning history and execution history in blockchain.
Couger Inc. (Headquarters: Shibuya-ku, Tokyo. CEO: Atsushi Ishii, hereinafter referred to as “Couger”) had developed a system that ensure the reliability of AI which called “GeneFlow”, that combines AI, IoT, AR/VR, and blockchain. As one part of the system, which called “Connectome” is a technology that make smart space and records AI learning history and execution history in blockchain. In addition, currently we are looking for external partner companies to start closed test of this technology.
In technology development of AI and robotics, Couger is working on early development of the GeneFlow project, which ensure the reliability in AI learning history and execution history, pursuing technology development that are mainly corresponding to public blockchain “Ethereum” and consortium type blockchain “Quorum”. In the future, in addition to internal demonstration experiments, we will conduct closed tests with external partners and aiming to find countermeasure for performance and large-scale data that required for actual applications.
The field of AI evolves greatly due to machine learning and its related technology, Deep Learning. The feature of machine learning is that it relied heavily on learning data rather than algorithm. AI can be anything depending on the learning data. In order to realize the automation of hardware which make advanced and complicated judgement such as robot, it is essential to utilize innumerable learning data. As a result, there are concerns about “Black Box Problem” follow by unknown basis ground judgment of AI.
Moreover, in the coming era of IoT, the bandwidth and type of is increasing enormously by connecting to innumerable hardware such as automatic driving car, drones, smart speakers, and etc. Along with that, it is expected that AI will process these data in real time, it will be in the state of “AI everywhere” when installed with any devices.
What is important at that time, the history of “What data have AI learned” and “What judgement result of AI ( Execution Result) are for securing the reliability of the decision and operation.
- Purpose (Problem Solving)
GeneFlow is classified into three functions: Learning Processing, Execution Processing ( Inference Processing), and Optimal Model Selection. In learning process, we store the learning history of what kind of data Al learned in blockchain. In doing so, learned models and training data are stored in a distributed file system. In the execution process, the learned model is called out from the distributed file system and the execution result for the target data is displayed. In the optimal model selection, evaluating multiple learned models generated in various combinations and able to select the optimum model.
All of these history are recorded in blockchain, so there will not be tampered or lost, giving great confidence in the operation and formation of AI. In addition, history is committed, it is possible to share reliable and safe learning data and learned models among multiple companies and research institutes.
Developing own technology “Connectome” that create smart space, combines AI, IoT, AR/VR, and blockchain.