Deep learning 2 (5 cr)

Evaluation scale

0 - 5

Objective

The student can apply the methods used in deep learning in the Keras/TensorFlow environment. In addition, the student knows how to use GPU computing and CSC supercomputers in the training of neural networks and can use already trained neural networks in a web browser.

Execution methods

The content and instructions of the course are covered in the introductory lecture (approx. 2h), which can also be watched as a recording later.
The course does not have separate lectures, but the course lecture materials are available as videos.
Separate question-and-answer sessions are organized during the course, where you can ask questions about unclear issues and get help with assignments.

The course discussion takes place on the course's discord channel.

Accomplishment methods

The course does not have a separate exam, but the performance of the course is based on returning the exercises of the course. Passing the course requires returning all course assignments.

Content

- Use of Keras/TensorFlow environment
- Distributed training on multiple GPUs
- The use of CSC's supercomputers in the training of neural networks
- Basics of large language models (LLM).
- Running neural networks in different environments (deployment)
- Implementation of a trained neural network in a web browser
- MLops basics
- Artificial intelligence and ethics

Assessment criteria, satisfactory (1)

A grade of 1 requires the return of all assignments and 50% of the course points. In addition, the returned course exercises must show that the student knows how to use the Keras/Tensorflow environment, load a model into it, and teach and use it.

Assessment criteria, excellent (5)

For a grade of 5, approx. 90% of the points in the course exercises are required. In practice, this means that all returned code works and the reflection sections of the exercises are commendably done.

Qualifications

Deep learning 1