Our new training tool is OUT. Starting today, the human friends of Iris.AI can train the AI directly from research paper abstracts and help her understand synonyms.
In May 2016 we launched the first version of our AI training tool to help Iris.AI learn via Ted Talks. A large labelled data set was needed to build an effective training loop. Since then we’ve asked our users to join us in our effort to take the existing stock of research into effective use by participating in this learning experiment.
Fast-forward 10 months and the crowd-training has become one of the backbones of our AI development. Iris.AI trainers around the world have done a tremendous job by training thousands of texts, equivalent to more than half a million trained concepts. These inputs have allowed us not just to improve the accuracy of our algorithm by several percentages but also to verify and assess the quality of it.
Today we’re taking the AI training to new heights by releasing the next version of our training tool. The new curriculum allows Iris.AI to learn directly from research papers and synonyms. This means that the source data of the supervised data set expands to millions of texts letting Iris.ai optimise her neural nets with scientific concepts across research fields.
Our next goal is to gather and inject a trained dataset of 5000 paper abstracts to the algorithm. With those inputs we aim to improve the connections in the neural nets of Iris.AI by approximately 10 %.
Interested in joining the effort and taking Iris.AI to the next grade? Sign up to become a trainer and we’ll help you get started.
Kudos to all our existing AI trainers and welcome new ones! We look forward to sharing the next leaps of this journey with you.