RESUMEN
Supervised machine learning classification is the most common example of artificial intelligence (AI) in industry and in academic research. These technologies predict whether a series of measurements belong to one of multiple groups of examples on which the machine was previously trained. Prior to real-world deployment, all implementations need to be carefully evaluated with hold-out validation, where the algorithm is tested on different samples than it was provided for training, in order to ensure the generalizability and reliability of AI models. However, established methods for performing hold-out validation do not assess the consistency of the mistakes that the AI model makes during hold-out validation. Here, we show that in addition to standard methods, an enhanced technique for performing hold-out validation-that also assesses the consistency of the sample-wise mistakes made by the learning algorithm-can assist in the evaluation and design of reliable and predictable AI models. The technique can be applied to the validation of any supervised learning classification application, and we demonstrate the use of the technique on a variety of example biomedical diagnostic applications, which help illustrate the importance of producing reliable AI models. The validation software created is made publicly available, assisting anyone developing AI models for any supervised classification application in the creation of more reliable and predictable technologies.
RESUMEN
A central question in the study of ecology and evolution is: "Why are there so many species?" It has been shown that certain forms of the Lotka-Volterra (L-V) competition equations lead to an unlimited number of species. Furthermore, these authors note how any change in the nature of competition (the competition kernel) leads to a finite or small number of coexisting species. Here we build upon these works by further investigating the L-V model of unlimited niche packing as a reference model and evolutionary game for understanding the environmental factors restricting biodiversity. We also examine the combined eco-evolutionary dynamics leading up to the species diversity and traits of the ESS community in both unlimited and finite niche-packing versions of the model. As an L-V game with symmetric competition, we let the strategies of individuals determine the strength of the competitive interaction (like competes most with like) and also the carrying capacity of the population. We use a mixture of analytic proofs (for one and two species systems) and numerical simulations. For the model of unlimited niche packing, we show that a finite number of species will evolve to specific convergent stable minima of the adaptive landscape (also known as species archetypes). Starting with a single species, faunal buildup can proceed either through species doubling as each diversity-specific set of minima are reached, or through the addition of species one-by-one by randomly assigning a speciation event to one of the species. Either way it is possible for an unlimited number or species to evolve and coexist. We examine two simple and biologically likely ways for breaking the unlimited niche-packing: (1) some minimum level of competition among species, and (2) constrain the fundamental niche of the trait space to a finite interval. When examined under both ecological and evolutionary dynamics, both modifications result in convergent stable ESSs with a finite number of species. When the number of species is held below the number of species in an ESS coalition, we see a diverse array of convergent stable niche archetypes that consist of some species at maxima and some at minima of the adaptive landscape. Our results support those of others and suggest that instead of focusing on why there are so many species we might just as usefully ask, why are there so few species?
