Key concepts like computer science, data science, artificial intelligence, and machine learning are defined and differentiated.
The typical machine learning cycle involves fetching data, cleaning it, training a model, evaluating performance, and deploying to production.
Initial research reveals existing models are inadequate, leading to the decision to use computer vision and object detection for the problem.
The initial pre-processing step involves manually labeling juggling objects in images using Azure Custom Vision, a time-consuming and unscalable process.
Using raw, user-generated content without cleaning leads to poor model performance, highlighting the necessity of filtering data before training.
A multi-cloud pipeline using AWS, Azure, and Google Cloud services automates data collection, cleaning, and preparation for model training.
The model is trained and evaluated using both Azure and Google Cloud platforms, revealing some humorous misclassifications along the way.
The trained model is exported as a Docker container, enabling easy and consistent deployment across local environments and multiple cloud providers.
Cloud platforms democratize technology by providing managed services that reduce the required expertise and time to build and deploy complex applications.
Key takeaways include the benefits of serverless architecture and automation, with future plans for a CI/CD pipeline and expanded model capabilities.
23:19Nočnica Fee
1:01:02Oliver Arafat
49:45Kris Howard
48:38Flo Pachinger
43:36Jens Eickmeyer
39:02Marek Suppa
34:21Aarno Aukia
48:07Sebastien Stormacq
.webp?w=240&auto=compress,format)
.gif?w=240&auto=compress,format)
Jobs that call for the skills explored in this talk.
Aws Emea Sarl
webLyzard
Vienna, Austria