Google presented the What-If Tool. The tool can visualize the impact of changes in artificial intelligence within TensorBoard. Users only need to add a model and a dataset for this.
At the moment, these kinds of scenarios often have to be created by writing separate code to analyze a specific model, says Venturebeat.
“Not only is this an inefficient process, but it also makes it more difficult for people who don’t program to participate in the process of creating and improving machine learning models,” said James Wexler, AI software engineer at Google.
The tool shows the impact of changes in algorithms. The information is displayed in real time. It is also possible to generate plots that show how the predictions of a model correspond to a function. It also looks at contradictory explanations and analyses of how honest the algorithm is.
The tool can show a comparison between a data point and the nearest data point where the model predicts a different result. It is also possible to show the effects of different thresholds for classifications.
According to Google, the tool has been used internally to detect functions of a dataset that were previously ignored. The tool is also used to discover different patterns in outputs that contribute to improved models. As of today, the open source tool is available.This news article was automatically translated from Dutch to give Techzine.eu a head start. All news articles after September 1, 2019 are written in native English and NOT translated. All our background stories are written in native English as well. For more information read our launch article.