Software Tool for Designing Machine Learning Models

(Photo: Sergey Sosnin / University of Wien)

Equipment and other chemical compounds are tested as a rule and at multiple levels, a toxicological and ecological risk that is possible. These issues are often so that we can think of ethical ideas back and forth, but we can not help but think people. Rechnergestützte Machine Learning methods can perform some tests, but the chemical compounds are synthetic and tested. Allerdings are fachleute and skeptical approaches to computer models, which are never more nachvollziehbar and transparent.

“That is a study for toxicologists,” says Sergey Sosnin, senior scientist at the Pharmaceutical Information Research Group at the University of Vienna. “If a model is used, because a connection with 99% security is not toxic, the connection is also slightly toxic, but there is no possibility to clear it, because it was false.”

In commodity trading, there is a binary classification: A machine learning model gives a probability between 0% and 100%, whether a chemical compound is typically toxic or not. This probability should indicate the trust of the model in your own property. The previous broader reflections, which are ideal, should only be given near 0% or 100% with correct predictions. If the model is no longer safe and gives a confidence rating of typically 51%, these previous acquisitions and alternative methods for risk assessment will be re-examined. There is a problem that occurs if the model is false with the high expectations that have arisen in the form of the sosnin.

Chemical connections with a 2D Ebene

The only thing you can do is that you can organize lessons in the form of identification, at the model of the blind Flecken hat and avoid that. If the model is used, the consequences of chemical compounds for a part of the product are treated – a mixture and a fehleranfällige warming.

“Um diese Forschenden zu Unterstützen”, the fort shows, “entwickelten with interactive graphic workzeuge, the chemical compounds on a 2D-Ebene projizieren, such as with the different maps. With the costs that the connection entails, those with higher security -falsch-vorhergesagt-wurden, soft drinks are used as Cluster Roter Punkte identifies können. The map is interactive and ermöglicht of the Benutzern, the chemical Raum is subordinate and besorgniserregende Bereiche zu erkunden.“

The methodology of the Research Team of the Uni has a number of models with binding and the tested Estrogen recipe. By the visual analysis of the chemical particles another model is used for the action of steroids and polychlorinated biphenyls, if it is small, the compounds are not so versatile and can not be used. The software compass integrated in this project is the community of erasure on Github that can be used for free. Sosnin hofft, the Mol-Compass Chemikern and Toxicologists are busy investigating the data of computer models.