DiffMEM

Existing tools (NONMEM, SAS, Winbugs) are not well suited for complex models and high-performance. The algorithms used are not always very suitable for use with large numbers of parameters and complicated non-linear models. And more importantly the hardware landscape has changed tremendously since those tools were developed: single core chip performance has begun to stagnate with a resulting move to multicore and more parallelism.

DiffMEM is tool for Non-Linear Mixed Effect model optimization in a frequentist and Bayesian context built specifically for complex models (mostly ordinary/partial differential equations) and performance. Through heavy the use of parallism, extensive low-level optimizations and algorithmic innovation, it is able to rapidly fit complex models. This in turn allows for shorter duty-cycles while building a model, and enables adaptive/optimal trial design and model validation since those typically require a lot of simulation+estimation steps.