Date: November 12 at 11 am
Speaker: Sara Perez Carabaza – CeADAR
Title: Application of convolutional neural networks in remote sensing classification
Abstract:
This presentation will be looking at the applications of Convolutional Neural Networks (CNN) on image classification in the remote sensing field. CNN are powerful deep learning techniques to automatically learn patterns from labelled images. It has been successfully applied across multiple visual tasks: image classification, object recognition, and image segmentation; so applying CNN techniques to remote sensing classification problems is a promising research field. In this talk we will address some of the challenges it involves: the lack of labelled imagery, and the need to integrate multispectral and temporal information from remote sensing images. We will focus on two remote sensing problems: i) crop classification from Sentinel-2 satellite imagery and ii) classification of Irish habitats using aerial imagery.
Zoom link: https://ucd-ie.zoom.us/j/87501651320?pwd=cVFrKzhQQlJQZHVrZVVuemk3WEJYQT09
Online Recording: https://youtu.be/X6BZcEZxoaA
Date: November 19 at 11 am
Speaker: Lucas Pereira – Universidade de Lisboa & PRSMA
Title: Energy Monitoring, Storage Control and AI in the Wild: Experiences from a real-world demonstrator
Abstract:
The European electricity sector is the main stage for several smart-grid demonstration projects. In this respect, the Smart IsLand Energy systems – SMILE project demonstrates different smart-grid solutions in three large-scale demonstrators in the Orkneys, Samsø, and Madeira islands.
In Madeira Island, one of the main goals is to demonstrate the potential for increasing the share of renewable energy sources (RES) in the electric grid by leveraging the potential of stationary energy storage. More concretely, small-scale storage to increase self-consumption rates in domestic and commercial settings, and large-scale storage for voltage control and load-leveling at the substation level.
While deploying laboratory prototypes in real-world demonstrators is technically challenging, reports on real-world experiences are seldom reported in the literature. Therefore, in this talk, we will discuss the challenges and experiences of deploying and maintaining such large-scale real-world demonstrators. Special attention will be given to the hardware-software research infrastructures that support the demonstrator life cycle (from data acquisition to actuation/control and end-user interactions/reactions) and the deployment and evaluation of AI models that support different storage control algorithms.
Zoom link: https://ucd-ie.zoom.us/j/82813276507?pwd=b3F0Q3VTN0RtcW1Eb1FYTk8wQU5WQT09
Online Recording: https://youtu.be/hRuZ5byl9is
Date: November 26 at 11 am
Speaker: Ihsan Ullah – CeADAR
Title: Overview of Federated Learning framework, tools, and examples
Abstract:
AI or more specifically deep learning need data to train on. However, majority of data is limited mainly because of privacy and security reasons. A solution is proposed by Google is a federated learning approach.
In this talk, questions like What is federated learning?, Why is federated learning adopted?, Architecture (or its types) of Federated learning? Application of federated learning, what are the challenges it is facing, libraries and devices that can be utilized as well as some examples of how researchers adopted/implemented it, will be discussed.
Zoom link: https://ucd-ie.zoom.us/j/89422706781?pwd=NVRlTU5iUnNOeDUzY0kzcDNGRnNpUT09
Online Recording: https://youtu.be/DQf7AeYjb1A
