Noticias en español
Desarrollo de algoritmos de machine learning para generación de mapas LCLU de Latinoamérica y el Caribe para el proyecto CopernicusLAC Chile
A partir del cruce de múltiples productos internacionales de LCLU se generan datasets diferidos para el entrenamiento de algoritmos de machine learning especializados hacia los diferentes climas presentes en Latinoamérica y el Caribe. El producto resultante se contrasta con aquellos ya disponibles y se establecen los pasos a seguir para la obtención de resultados más precisos y la inclusión de una mayor cantidad de coberturas de suelo en las siguientes iteraciones del proyecto Land Cover del Centro Regional CopernicusLAC Chile.
Read MoreAttention is Turing Complete.
Abstract: Alternatives to recurrent neural networks, in particular, architectures based on self-attention, are gaining momentum for processing input sequences. In spite of their relevance, the computational properties of such networks have not yet been fully explored. We study the computational power of the Transformer, one of the most paradigmatic architectures exemplifying self-attention. We show that the Transformer with hard-attention is Turing complete exclusively based on their capacity to compute and access internal dense representations of the data. Our study also reveals some...
Read MoreProphet Inequalities Require Only a Constant Number of Samples.
Abstract: In a prophet inequality problem, n independent random variables are presented to a gambler one by one. The gambler decides when to stop the sequence and obtains the most recent value as a reward. We evaluate a stopping rule by the worst-case ratio between its expected reward and the expectation of the maximum variable. Because of its connections with resource allocation and posted-price mechanisms, this problem has been intensively studied, and several variants have been introduced. While most of the literature assumes that distributions are known to the gambler, recent work has...
Read MoreOn Wu’s Inequality and the Poisson-Föllmer Process.
Resumen: In the discrete setting the Poisson distribution is a ubiquitous object, as the Gaussian distribution is in the Euclidean setting. In spite of that, it does not satisfy Gross’ log-Sobolev inequality. Nevertheless, Bobkov and Ledoux were able to prove that it satisfies a “modified” version of it, which was subsequently reinforced by Wu. In the first part of this talk we will exhibit a new stochastic proof of Wu’s modified log-Sobolev, via an entropy-minimizing process constructed by Klartag and Lehec, which we call the Poisson-Föllmer process. We will also see how this stochastic...
Read MoreEstudio Comparativo De La Evaluación Antropogénica Y Geogénica De La Concentración De Metales Pesados Y Lábiles En Sedimentos Marinos Superficiales Y Suelos De La Zona Norte De Chile.
Resumen: La investigación, permite la identificación, priorización, jerarquización y evaluación preliminar de la contaminación en zonas muy degradadas ambientalmente y posibilita la determinación de los niveles de los cuales se establecen los límites de inocuidad y peligrosidad en ambientes costeros. El propósito del estudio es elaborar un diagnóstico regional de los suelos y sedimentos marinos del borde costero del norte de Chile con potencial presencia de elementos trazas y metales pesados. El estudio, analiza la potencial presencia de 10 metales pesados en el borde costero...
Read MoreCoarse Graph Theory I – The Questions
Resumen: Coarse graph theory is concerned with the large-scale geometric properties of infinite graphs, especially the properties preserved by quasi-isometries. Although such inquiries have been raised since the 80’s, it was usually in the context of Cayley graphs and geometric group theory. Recently the topic migrated to graph theory with intriguing outcomes. Eventually, the field of coarse graph theory was officially inaugurated and systematically treated in a paper by Georgakopoulos and Papasoglu. In this talk, I will explain the basics of coarse graph theory, exploreits...
Read More