Content
The laboratory course "Deep Learning" consists of two challenging tasks. The first task is about automated detection and grading of diabetic retinopathy, which is a diabetes complication affecting the eyes, based on retinal images. Deep visualization methods will be used to visualize the functional behavior. In the second task, recurrent neural networks will be applied to recognize human activities from inertial sensor data.
Goals for the first task "Diabetic Retinopathy Classification":
- Build a generic framework for your deep learning pipeline (efficient data loading, training, evaluation, ...).
- Classification of nonreferable and referable diabetic retinopathy.
- You will use one deep visualization method to get insights into your trained model.
Goals for the second task "Human Activity Recognition":
- You can record your own data.
- Classify different activities for a single sensor position based on the accelorometer and gyroscope.
- Hyperparameter optimization through grid search or Bayesian optimization.
- You have the possibility to deploy a model on your smartphone (Android).
The lab will be performed in teams of two.
Learning Objectives
This laboratory course is an excellent opportunity to gain practical experience with Tensorflow and to extend your academic knowledge in deep learning.
Prerequisites
- Passing the Deep Learning exam
- Submission of the Deep Learning programming assignments (individual submissions required)
Registration
Registration takes place in Campus during the first week of lectures (Monday 07:00 - Tuesday 23:59) in the winter term. The registration time does not matter whether you will receive a spot or not. You will be notified on Wednesday if you received a spot.
Contact person
The Deep Learning Lab is only offered during the winter term.