Mark Coletti

Validating Safecast data by comparisons to a U. S. Department of Energy Fukushima Prefecture aerial survey

Safecast is a volunteered geographic information (VGI) project where the lay public uses hand-held sensors to collect radiation measurements that are then made freely available under the Creative Commons CC0 license. However, Safecast data fidelity is uncertain given the sensor kits are hand assembled with various levels of technical proficiency, and the sensors may not be properly deployed. Our objective was to validate Safecast data by comparing Safecast data with authoritative data collected by the U. S. Department of Energy (DOE) and the U. S. National Nuclear Security Administration (NNSA) gathered in the Fukushima Prefecture shortly after the Daiichi nuclear power plant catastrophe. We found that the two data sets were highly correlated, though the DOE/NNSA observations were generally higher than the Safecast measurements. We concluded that this high correlation alone makes Safecast a viable data source for detecting and monitoring radiation.

Towards Validating a Model of Households and Societies in East Africa

One of the major challenges of social simulations is the validation of the models. When modeling societies, where experimentation is not practical or ethical, validation of models is inherently difficult. However, one of the significant strengths of the agent-based modeling (ABM) approach is that it begins with the implementation of a theory of behavior for relatively low-level agents and then produces high-level behaviors emerging from the low-level theory’s implementation. Our ABM model of societies is based on modeling the decision making of rural households in a 1,600 km (1,000 mile) square around Lake Victoria in East Africa. We report on the first validation of our model of households making their living on a daily basis by comparing resulting activities against societal data collected by anthropologists.

The relationship between evolvability and bloat

Bloat is a common problem with Evolutionary Algorithms (EAs) that use variable length representation. By creating unnecessarily large individuals it results in longer EA runtimes and solutions that are difficult to interpret. The causes of bloat are still uncertain, but one theory suggests that it occurs when the phenotype (e.g. behaviors) of the parents are not successfully inherited by their offspring. Noting the similarity to evolvability theory, which measures heritability of fitness, we hypothesize that reproductive operators with high evolvability will be less likely to cause bloat. We set out to design a new crossover operator for Pittsburgh approach classifier systems that has high phenotypic heritability. We saw an opportunity using the nearest neighbor representation to perform crossover cuts in phenotype space rather than on the genomes. We demonstrate that our operator tends to be less susceptible to bloat and has higher evolvability than a standard Pittsburgh approach crossover operator. Our hope is that this will lead to a general approach to reducing bloat for any representation.

The effects of training set size and keeping rules on the emergent selection pressure of learnable evolution model

Evolutionary algorithms with computationally expensive fitness evaluations typically have smaller evaluation budgets and population sizes. However, smaller populations and fewer evaluations mean that the problem space may not be effectively explored. An evolutionary algorithm may be combined with a machine learner to compensate for these smaller populations and evaluations to increase the likelihood of finding viable solutions. Learnable Evolution Model (LEM) is such an evolutionary algorithm (EA) and machine learner (ML) hybrid that infers rules from best- and least-fit individuals and then exploits these rules when creating offspring. This paper shows that LEM introduces a unique form of emergent selection pressure that is separate from any selection pressure induced by parent or survivor selection. Additionally this work shows that this selection pressure can be attenuated by how the best and least fit subsets are chosen, and by how long learned rules are kept. Practitioners need to be aware of this novel form of selection pressure and these means of adjusting it to ensure their LEM implementations are adequately tuned. That is, too much selection pressure may mean premature convergence to inferior solutions while insufficient selection pressure may mean no sufficient solutions are found.

A preliminary study of learnable evolution methodology implemented with C4.5

The learnable evolution model (LEM) introduces a machine learning-based birth operator into an evolutionary computing algorithm. New individuals are generated from hypotheses learned by the operator from the most-fit and least-fit parent sub-populations. The LEM allows for arbitrary machine learning mechanisms, though, so far, only an AQ (Algorithm Quasi-optimal) based machine learner has been used in LEM implementations. This paper describes preliminary results using a different machine learner in a LEM implementation - C4.5.