Competition participants use publicly available data from the internet to train models. This means that the host’s data is never made public and is only used for validation of competition models in an air-tight manner.
The large sets of training images collected by participants are aggregated to allow platform.ai to produce models that are generalizable to real-world conditions than in-house datasets that may have inherent biases attributed to design and production choices.
Platform is a development environment that uses iteration to rapidly improve model performance. Used in a competition format, platform exponentially increases pace of model development because of the number of people developing models to compete against each other.
Depending on the time and complexity of a project, the cost of developing AI models can run into thousands of dollars. This includes the cost of skilling and hiring. Platform competitions are a compelling alternative for smaller scale projects meant to test the waters.