ANTARES Deep learning workshop

Samenvatting

Workshop on introduction on foundational and advanced Deep learning concepts to PhDs, Postdocs and Senior researchers at BioSense Institute, Serbia

Day 1 (10:00-18:00)
10:00-12:00
Topic 1A: Welcome / Workshop Overview / General Deep Learning Overview
Topic 1B: Introduction to PyTorch
- PyTorch tensors and variables, using the GPU, gradient computation.
- Practical: Fitting a linear model
Lunch break
13:00-14:30
Topic 2: Basic Neural Networks and Datasets in PyTorch.
- Defining a basic tensor dataset. Defining a sequential neural network model. Representation learning with XOR dataset as an example.
- Practical: The XOR dataset, the moons dataset
Coffee break
14:45-16:00
Topic 3: Image Classification
- Using PyTorch’s library of existing datasets. Building a feedforward neural network for image classification. Saving / Loading a model.
- Practical: Digit classification on the MNIST dataset. (Grayscale images)
Coffee break
16:15-18:00
Topic 4: Basics of Fully Convolutional Neural Networks (FCNs) for Image Segmentation
- Convolutions in PyTorch. Data Enrichment.
- Practical: Grape Stalk Segmentation (RGB Images)

Day 2 (9:00-18:00)
09:00-10:30
Topic 1: FCNs for more complex image datasets
- Max Pooling / Unpooling in PyTorch. Data Enrichment.
- Practical: Dog Segmentation in RGB images
10:30-12:00
Topic 2: Hyperspectral data and dealing with large datasets
- Introduction to Hyperspectral Imaging
- Practical: Preparing the NIR apple dataset.
13:00-14:30
Topic 3: FCNs for Hyperspectral Image Segmentation
- Introduction to Hyperspectral imaging
- Practical: Damage detection in NIR hyperspectral images of apples.
14:45-16:00
Topic 4: Unsupervised learning on hyperspectral data / Autoencoders / Reusing neural network components in other models
- Practical: Autoencoder on NIR apple dataset.
16:00-18:00
(Possible/Time permitting) Topic 5: Using multi-GPU machines
- Discussion (No practical might be possible)

Day 3 (t.b.d, max 2 hours):
- General discussion