Challenge #1-2 PI Leads: david.kaplan@tufts.edu; neuro014@gmail.com; ssblemker@virginia.edu; mbuehler@mit.edu; figueroc@med.umich.edu;
IMAG POCs: TGLewis@niaid.nih.gov; beth.e.lewandowski@nasa.gov; david.miller3@nih.gov; ferrantem@nih.gov;
8:30-8:50am: Update on the Accomplishments for Challenges #1-2 thusfar,
8:50-9:00am: Discussion of the gaps to be addressed in the future.
Questions to pose to the MSM Consortium related to these challenges:
- Just a comment: There is ongoing controversy in the AI arena regarding the efficacy of undirected Ai approaches in creating generalizable outputs (i.e. avoid brittle solutions); suggest we need to avoid possible trap from over-emphasis on "data-centric" models versus using existing knowledge structures to inform, direct and constrain applications of these methods. Also need to recognize that AI/ML is not a solution to data sparsity, but rather is limted by it. Important to acknowledge the importance of simulated/synthetic data in overcoming limitaitons of data sparsity. - From Gary An (sorry, can't figure out how James is adding Comments...)
Comment
Challenge 2 comes up with important issues relevant elsewhere. How do we describe (ontologies, language standards, interfaces) changes in time and space? How do we capture semiquantitative data, e.g. increase, decrease, bigger, smaller,...? Finally, how do we deal with intrinsic stochasticity in models and individuals? How do we quantify and compare results from spatiotemporal models? These are issues we'll hear about throughout the day. Interesting that they are also concerned about lack of standards for agent/cell models.
Terminology question: Combining continuum models (e.g., CFD, FEM) with other techniques such as Agent Based Modeling has great potential. I have seen some examples in various papers, but am wondering if there is some standard terminology we could use to identify more easily when these techniques are combined. I have heard "hybrid models" used in a related context, but that seems perhaps too generic. Thanks.
How about software? How does panel feel about the current software tools to integrate methods across domains?
Need to include cell behavior level in hierarchy, cells do mechanical and dynamic things, they are not just chemical reactors. Would be helpful to try to define components and interfaces, so that we could understand issues in model sharing and interconnection.