Data-driven and model-driven procedures for environmental studies and beyond
Tipologia Documento
Tipologia
- documento ricercabile
Contenuto
- Seminario tenuto alla Scuola di Dottorato in Statistica e Matematica finanziaria dell'Università di Milano-Bicocca il 2018-0528 su invito del Coordinatore, Prof. Fulvia Mecatti.Part One: Unsupervised classification of scattering patterns from airborne material particles.
Talk summary.An experimental method known as TAOS (two angle-resolved optical scattering) consists of collecting intensity patterns of LASER light scattered by single airborne material particles. A TAOS-based instrument operating in the Baltimore, MD area a few years ago sampled outdoor air and produced thousands of scattering patterns. Said patterns span a solid angular sector which includes the half sphere in the backward scattering directions.Discrimination result: TAOS patterns from bacterial (Bacillus subtilis) spores have been discriminated (http://hdl.handle.net/10281/47995) against patterns from (presumably) innocuous materials; the implemented algorithm consists of two interacting modules: extraction of morphological descriptors and linear classification. Open problem: to form classes within the set of TAOS patterns from outdoor sampling by a procedure which has to be unsupervised, because no patterns from known (=reference) materials are available.Part Two: Time sequences and orbits of dynamical systems.The qualitative theory of dynamical systems aims at characterising the solution to e.g., a system of ordinary differential equations without attempting their integration. Orbits are curves describing how the system evolves in time with respect to a reference frame where the system state components are the coordinates. The properties of orbits can be determined or estimated by means of qualitative theory. In a model-driven context one starts with time sequences of experimental data of any measurable quantity, then orbits from said sequences are formed and qualitative properties of orbits are looked for. Relations between quantities can be thus inferred by means of available models. Applications range from demography, to epidemiology and financial markets and aim at discovering the dynamics underlying observed processes.
Come si vede, la presentazione è divisa in due parti. Gli aspetti progettuali della prima parte compaiono alle pp. 31 e 32 e si riassumono con l'espressione order from TAOS. Gli aspetti progettuali della seconda parte sono soltanto accennati nelle pp. da 37 a 40, che elencano alcune applicazioni. In particolare, l'applicazione che figura a p. 40 ebbe séguito nel Sottoprogetto: Biologia molecolare.