Google has developed the home property of Entwicklermesse Google I/O an einheitliche Cloud-Plattform für Machine-Learning-Anwendungen (ML). Vertex AI combines Services vom Training der ML-Modelle tot zum produktiven Einsatz. The platform is based on data scientists and ML engineers and has abdecken Anwendungen’s gesamten Lebenszyklus about Machine Learning Operations (MLOps).
Der ML-Lebenszyklus aus Sicht von Vertex AI
(Image: Google)
Vertex AI combines various internet connections with a central user interface and its own API. Die platform bietet Werkzeuge zum Erstellen and Trainieren, Ausrollen and Verwalten von ML-Modellen.
Centralization and automation
Data scientists can model with manual training for AutoML (Automated Machine Learning). Letzteres is a relativistic young disziplin, who makes the settings of ML models automatable. Statt Modelle iterativ beim Training manual anzupassen, erfolgt die Optimierung des Models auf Basis der eingegebenen Trainingdaten automatic. Google has found other Cloud Provider AutoML over time in its portfolio.
Das iX-Developer-Sonderheft “Machine Learning” covered on 148 Seiten include ML-Frameworks, Textanalyse, Reinforcement Learning and the Produktivstellung von Modellen. I support the practical application, and most of the articles consist of Jupyter notebooks as the basis for own ML projects.
One of the articles published by Cloud-Angebote about Machine Learning from Google, Microsoft, Amazon and Co. are distributed is another article described with the Verwalten and Bereitstellen of ML models.
The platform is developed by Google internally developed Toolkit for standard ML analytics for computer vision, text analysis, data analysis and analysis. Die Modelle lie in a central repository. Vertex Visor is a job seeker that optimizes hyperparameter. The platform is used to use Vertex data labels for manual labeling of data by Menschen and Vertex Explainable AI, as a look at the Arbeit der Modelle gibt and at Nachvollziehen der Ergebnisse helfen soll. Schließlich exists a direct connection for Jupyter notebooks.
Pipelines and features
Vertex Pipelines help automate MLOps processes. Whether it’s different components, the jewels need to be updated in the workflow or the workflow, and multiple links and operations can be done to get a container image. A typical ML-Ops-Pipeline has a component of Aufnehmen and one of the Vorverarbeiten der Daten. Follow the following components for training and evaluating models. Schließlich übernimmt a component of the implementation of the models in the productive Betrieb. It is a fact that we have used several components for the monitoring and Verwaltung of the models.
Vertex AI uses components and services to set up and apply ML models.
(Image: Google)
Neben dem Repository für die Modelle bietet die Platform einen Bereich zum zentralisierten Organisieren and Speichern von ML-Features. Der dafür verantwortliche Vertex Feature Store hat alldings zum Start der Platform nor Preview-Status. Weitere Details zu Vertex AI sich dem Google Cloud Blog logged in.
(rm)