You are here: Home Kolloquium Probabilistic Machine Learning – Inference and Structure

Probabilistic Machine Learning – Inference and Structure

Dr. Nico Piatkowski, TU Dortmund

Event details
What
    When Jan 10, 2020
    from 11:15 to 12:00
    Where Raum 0.016, Endenicher Allee 19a
    Add event to calendar vCal
    iCal
    Probabilistic methods provide the basis for various data science techniques, including logistic regression, latent Dirichlet allocation, Markov random fields, and Boltzmann machines. They allow us to learn discriminative and generative models of the data, which achieve high classification and sampling performance in a variety of tasks, while providing statistically sound uncertainty estimates. The two major computational challenges involved in learning such models from data are probabilistic inference and structure estimation. I will give an overview of my contributions to that field. We will see (1) that probabilistic models can be scaled down to ultra-low-power devices, (2) how quantum computation can help us with probabilistic inference, and (3) scalable approximate inference via stochastic quadrature. Finally, (4) we connect probabilistic models and deep learning to derive a recent result for the depth of the latent space. An outlook on future challenges concludes the talk.

    Vorab wird in einem kurzen Vortrag einen Einblick in eine Lehrveranstaltung zu Thema 
    Entscheidungsbäume
    gegeben.
     
    Wir machen uns mit Entscheidungsbäumen und ihren grundlegenden Eigenschaften vertraut.
    Basierend auf bekannten Datenstrukturen werden Entscheidungsbäume formal definiert. Anschließend
    wird ein Algorithmus für das Lernen von Entscheidungsbäumen aus Datensätzen vorgestellt.
    Document Actions