I am a machine-learning and design research practitioner interested in real-time discovery and planning problems in risk governance and engineering design.
Current Research ProblemsCausal Elicitation: How does even one appropriately elicit scenarios for governing risky situations with multiple stakeholders, when the participants potentially have all kinds of differences in their worldview and shaky grasps on the underlying facts? A related question is how does one represent and display the conflicting accounts emerging from the testimony of multiple stakeholders? This project looks at non-parametric elicitation that looks to capture the causal networks people intuitively hold, and see if they can be used to elicit better predictions and more sensible strategies than merely asking about future outcomes.
Design from the perspective of statistical processes: How do you know when your understanding of potential causal factors is complete enough for engineering or policy? I think that design and ethnography practices teach us to look further than we otherwise might (to bias exploration) and that there is much to learn about how design works from a statistical perspective (similar to recent work in cognitive developmental psychology and the psychology of science). In particular, understanding how to take appropriate priors over non-parametric structures should lead to non-parametric engineering design methods.
Making design and diagrams "portable": I think of design as being very project-driven: you try a bunch of things, you finish one of them to completion, and then you take whatever you learn, but the models that you often build end up in the trash. In machine learning, there's a problem called "transfer learning", where you ask given that you know one kind of statistics, what can you infer about a different relationship. There's possibly a design analog, where after you've illustrated one set of relationships, how do you (or somebody else) recover those connections for a different project?
Making causal information "portable": Even beyond representing causal information in the first place, I think there are some challenges in making it "portable". Everyone knows there are active volcanoes in Iceland, yet nobody can be practically expected to use that information to assist their business travel planning. How to do you pull in what you already "know" when making plans? Also, how does one assemble this common knowledge? Can it be mined from text?
Planning across closed-loop supply chains: Closed-loop planning issues interest me because of the temporal effect on values: no matter the social values a given society might determine is appropriate, using up a resource thermodynamically will affect future societies with different social arrangement. This leads to some frighteningly easy decision-theoretic planning, as the difference between perfect recovery versus total depletion tends toward infinity over time. However, rarely are those extremes accurate characterization of the situation at hand.
Materialist foundations for ontology What would a computational implementation of a materialist ontology (i.e. Manuel DeLanda's work) look like?
Interesting Data SetsHere are some data sets I've long thought I might enjoy. The ideal statistics for each would, where applicable, span worldwide or a otherwise have a strong cross-cultural sampling, be separated by region, and spanning as far back as 2000 years ago; and include totals, or where not available averages, medians, and quartiles (I'll settle for variances)
- occupational statistics
- inventories of durable goods owned in a lifetime, ideally separated into each decade of life
- inventories of consumable goods used, monthly or weekly
- major flows of fuel and mined materials, as well as known reserves; including aggregates per region for renewables and other distributed streams
- waste statistics
- dates of inventions and installations of new kinds of equipment (i.e. the first lathe in North America)
- institution statistics (i.e. number and population of schools, prisons, hospitals, financial institutions, and types of businesses)
- agricultural production (crops planted, volume, and in particular nutritional density)
- linguistic and word usage statistics