DIENS

Working on various industrial case-studies in the past decade, we realized that, given the very quick evolution of hardware platforms, and the growing complexity of embedded software, it is, more than ever, necessary to start with a very general point of view. On one hand we need to understand the physical timing constraints and tolerances as determined by control engineers for a given application (sampling frequencies and jitters, end-to_end latencies, etc.), without mentioning software entities like tasks or scheduling. On the other hand we need to characterize precisely the computation and communication performances of a given execution platform, and assess their predictability. Designing the implementation means finding space and time allocations for the application functional parts, in such a way that the physical constraints are met by construction, and in a provable way for certification authorities. We will use examples with the Kalray MPPA, to illustrate the nature and specification of the timing constraints, the implementation schemes, and how the constraints can be met.

“Big mechanisms are large, explanatory models of complicated systems in which interactions have important causal effects. The collection of big data is increasingly automated, but the creation of big mechanisms remains a human endeavor made increasingly difficult by the fragmentation and distribution of knowledge. To the extent that the construction of big mechanisms can be automated, it could change how science is done”. Paul Cohen, DARPA 2014.
If biological big data has underlying “big mechanisms”, there is little hope that any mental construction can ever comprehend these “large explanatory models”. Moreover, because of the combinatorics of protein-protein interactions, no static map, database, or ontology list, can suffice to complement human brains in the design of these models. These simple observations have taken a large swath of molecular biologists away from any (formal or semi-formal) modeling activity. Most experimental biologists give little credit to existing models, which they know are partial, outdated and have little predictive power. Yet, while rejecting models, experimental biologists have to rely on implicit mental constructs in order to decide which experiment to set up. Thus every biologist has a model, either implicitly or explicitly. In this presentation, we argue that the lack of success story of formal models in Systems biology is not due to the impossibility of having accurate and u! seful models, but rather to the fact that one is confusing models of interactions with explanations of biological functions. Our approach is to view models as programs and biological functions as computations. If models are to be treated as programs what language one should use to assemble them is a critical point. In this talk we will present the issue of modeling biological "big mechanisms": i.e, protein-protein interaction networks whose specification is too combinatorial to be explicitly represented. In particular we will discuss the lessons that can be learned from the DARPA "BIGMEC" project, in which we took part, and that involved conjuncts efforts from A.I experts, biologists and language theory computer scientists.