01/12/2022
Αύριο Παρασκευή 2 Δεκεμβρίου στις 17:00 στην αίθουσα 145Π58 του κτηρίου Επιστημών, θα πραγματοποιηθεί ομιλία στα πλαίσια του μεταπτυχιακού μαθήματος MLDS απο τον Ralf Klinkenberg, Founder & Head of Research στην RapidMiner.Η RapidMiner είναι μία παγκόσμια αναγνωρισμένη εταιρία εστιασμένη στο data science without code. Περισσότερα στο abstract της ομιλίας!
Title: From Interactive Data Analysis and Visual Process Design to Automated Machine Learning, Real-Time Data Streaming, and Machine Learning Model Deployment from an Easy-to-Use Graphical User Interface
Abstract: Increasing amounts of available data and advanced data analysis and machine learning enable new insights, forecasts, automation, and other value creating solutions for many use cases across many industries. The value of such solution increases significantly, if a wider variety of data sources can be integrated and if real-time predictions based on real-time data support decision processes and automations. Simplifying the design and deployment of data analysis and machine learning processes enables broader groups of users to leverage the power of machine learning for their use cases. The EU-funded R&D project INFORE (Interactive Extreme-Scale Analytics and Forecasting) addressed the challenges posed by huge datasets and data streams and paved the way for real-time, interactive extreme-scale analytics and forecasting. Today, at an increasing rate, industrial and scientific institutions need to deal with massive data flows, streaming-in from maritime surveillance applications, financial forecasting applications or cancer cells growth simulations as well as a multitude of other sources. The ability to forecast, as early as possible, a good approximation to the outcome of a time-consuming and resource demanding computational task allows to quickly identify undesired outcomes and save valuable amounts of time, effort and computational resources. Since not everyone is a data scientist and since not everyone knows how to configure and program the tools needed for real-time data streaming, making the design and deployment of data analysis processes simpler is crucial for a wider adoption of these technologies and to enable users in many industries to leverage the value creation potential for their use cases. Within the INFORE project, the focus of RapidMiner was to provide a unified software platform seamlessly integrating all data sources, data streaming technologies (Kafka, Flink, Spark Streaming, etc.), machine learning algorithms and libraries (Python, R, Google TensorFlow, DL4J, H2O, Keras, Weka, etc.), model validation schemes (cross-validation, sliding window validation, etc.), deployment options, visualizations, model monitoring and operations with easy-to-use interfaces. Users can visually design data analysis workflows in an easy-to-use Graphical User Interface (GUI) without having to code and they can train and deploy machine learned models locally on their computer or server or on distributed data streams, Hadoop or Spark clusters, in the cloud, or on the edge – all from a single unified graphical user interface. The goal is to ease and accelerate the process from data and idea to productive analysis processes and value creation with machine learning for as many users as possible, including not only data scientists but also domain experts from various industries like b**r brewers, electrical engineers, manufacturers, etc. as well as business analysts and managers. Within INFORE, RapidMiner and its project partners have developed an easy-to-use framework for handling and integrating large data streams from various sources using various standard technologies like Kafka, Flink, and Spark Streaming, for data preprocessing, for machine learning and parameter optimization (mostly offline), and for real-time model deployment on real-time data streams (online). The INFORE project demonstrated the applicability of this framework and its time series and forecasting capabilities and its Complex Event Detection and Prediction (CEP) capabilities on use cases in various domains including maritime surveillance and issue detection, financial time series forecasting, and cancer cell growth simulations and predictions. This presentation first provides an overview of the RapidMiner data science platform and then focuses on the developed real-time data streaming and machine learning framework and its user interface in RapidMiner.
