stories of storages

I believe in storages as a research paradigm. That is, viewing the world in terms of storages provides us with a rich perspective on nature, engineering, design, humanity — maybe on everything. In this note I try to sketch out how I like to think about this perspective.

The basis is the storage (duh!). And, each storage is associated (or equipped) with a set of mechanisms that dictate what the storage can do. Studying the behavior of storages over time does then lead to storage dynamics. The complete state-space of a storage tells us what is possible. Tracing the observed state-space tells us the parts of the complete state-space that were empirically realized. These are the dynamics of a storage’s in- and outputs. The functioning of the storages, as determined by the mechanisms/working-rules associated with a storage.

Storages are an excellent tool to mitigate all kinds of risks. For example, if we know that the yield of some things — say, agricultural fields — varies a lot between years, we can build storages to help us over the years that provide too much or too little of these things (of course the storability itself will also depend on the thing and the kinds of storages we are able to build).

More

TBD

Old

Stock & Flows & Risks as a window into everything. Surely one of my professional deformations. I think everything can be viewed from the angle of stock, flows and risk (in the sense of probability + cost/benefit). Examples: (1) Time flows, but only the fact that we can take stock of the changes makes us realize so. The risk perspective does then allow us to weight on certain paths that time takes. (2) Hydrologists always think in stock and flows. They call them storage and processes, which are perhaps the nicer words — but more difficult to remember. And, we integrate them into practice (water management) by using a risk based approach (think of floods, droughts, hydro-electricity). (3) Machine Learning models consist of an information storage (compression, parametrization, and indexation) plus flows that operate on top of it (saving, retrieval, interpolation). We train/build them by doing empirical risk minimization.

For some reasons some people/disciplines focus only on the benefit or only on the cost part of the risk. I think that must be some kind of bias, since usually both come in pairs, and often we have some kind of tradeoff.