Loading the MKC paper replay…
MKC2 extends protected MKC collaboration with application proxies. Develop complementary physical tests, train a private model and calibrate joint predictions against product tests, then use those predictions to select materials together.
This recorded synthetic comparison counts private measurements and complete-product tests separately. At the 90% best-available-product target in the matched four-company study, mean external tests after training (capped at 64 for unfinished programs) fall from 25.7 or 32.1 to 1, depending on material interactions. MKC2 uses 1,168 private curve measurements per program, about 200 more than ordinary MKC. Both approaches also have 48 earlier product tests. The earlier two-company study is shown separately. Laboratory method development is not costed, and these results are not measured industrial time savings.
Enable JavaScript for the develop, train and use walkthrough and interactive charts, or visit the MateriaX essays.