BROADCASTS.com

Anyone who has done any work with artificial intelligence can tell you that collecting enough data to train the AI can be as difficult as building the neural networks themselves. There are some decent libraries for standard data sets but say you needed to train an AI to identify a particular model of a product category. Finding the right images to train the system could be near impossible. To help solve this problem is Mindtech Global.

The company specializes in creating synthetic training data to help take an artificial intelligence system from concept to knowledgable. There are all sorts of scenarios where real-world data is not possible. Take, for example, drone data. An airport isn't going to allow an AI creator to fly a drone in their airspace to train what a good or bad drone might look like. By using synthetic data from Mindtech Global, you can still train your system without the need to inconvenience the airport.

In addition to difficult or impossible scenarios, there is also a bias in training data. When Project Natal, which became Xbox Kinect, was first demonstrated in 2009, it was done entirely with light-skinned people on stage. This is because the detection algorithms were not able to detect people with darker skin tones. Artificial intelligence systems face similar bias issues because of their training data. Humans tend to teach based on our experiences, which drives an AI to miss things. With the synthetic datasets, we eliminate humanity's tendency towards unconscious bias.

Currently, a lot of the bias is detected and solved by humans. For example, if a data scientist notices that systems have issues detecting people wearing backpacks, they will ensure that the dataset will include more images of people wearing backpacks. That means that future training sessions will have improved identification of people with backpacks.

To learn more about the Mindtech Global technology or engage their services, check out their website.

Interview by Scott Ertz of F5 Live: Refreshing Technology and Christopher Jordan of The Talking Sound.