Teach and Test

2018: Services for testing of the systems of artificial intelligence

On February 20, 2018 the Accenture company announced start of new services for testing of the systems of the artificial intelligence (AI) based on own methodology of Teach and Test which is intended for creation, control and measurement of the AI reliable systems in own infrastructure or in a cloud.

The technique of Teach and Test from Accenture provides acceptance with the AI systems of the correct solutions in two stages. At a grade level the choice of data, models and algorithms which are used in computer-based training is performed.

Accenture starts new services for testing of the systems of artificial intelligence

According to  the India Times edition, during training are investigated and different models are statistically estimated to select the most effective model for deployment from production.

On a test stage results of the AI system are compared to key performance indicators and is established whether a system can explain how this or that result was achieved. In decision making process the innovative technologies and cloud tools for permanent observation of a system for ensuring stable work are used. For example, the patented method of normalization uses a unique algorithm for faster testing of object recognition.

The Accenture company also used methodology of Teach and Test for training of the interactive virtual agent for the website of financial company that it could conduct a conversation with potential clients and the nobility when the person needs to retell the matter of a conversation. The agent was trained for 80% quicker, than it was possible earlier, and reached 85 percent accuracy in terms of providing recommendations to clients.

Besides, methodology for efficiency evaluation of service of a brand by the analysis of social networks, news and other sources in real time. Training time for models was cut by half that allowed to accelerate the analysis and to improve results.^[1]

Notes