If you have ever been involved with machine learning-based systems, you already know enough about trained data. Data must be delivered in the correct format and be accurate before you pour it into an AI model to train the model. Suppose you make a fraud detection engine that uses a popular machine learning system in a public cloud. You first create a dataset to train the model: in this case, a million transactional records with the fraudulent transactions labeled as such. The model thus learns which transactions have a high chance of being fraudulent and which are not.
(There are of course different types of training data, labeled or not.) Once trained, the model can continue to learn by gaining experience. If you had the time, the model could train itself to monitor transactions that would otherwise be perceived as fraudulent by people or other systems. AI as a mentor What is striking about this approach to training AI is that you must have a very good set of training data. Sometimes you can pick it up from open or proprietary data brokers. But was it if we could train one trained model the other?
That is not a new idea, since the emergence of AI we have been looking at plans to impart something to one system by either sharing data or even sharing experiences and automated sharing. If one AI could be a mentor to the other, the model becomes even more valuable and effective. That is easier said than done. Machine learning algorithms do not usually talk to each other even if they use the same software. They are built from scratch to be stand-alone students and only talk to non-AI or people. But inter-AI training is already on the radar of most suppliers. Recently I have seen two important trends that could herald this era: The use of SaaS-based AI engines that communicate with other AI engines in a public cloud or on-premises. You can see these as SaaS clouds specializing in teaching other AI engines a specific set of skills, everything from recognizing fraudulent transactions to medical diagnoses and machine maintenance, and more. AI engines are able to combine with your previously learned models, giving you a kind of super AI that not only uses experiences from its own training data, but can also apply global lessons. This is a trend to keep in mind if you want to create more value with the help of AI, such as machine learning and deep learning. In addition, many companies encounter the issue that they do not have enough training data to make machine learning functional. This would be a good solution for both challenges.