A host for the tutorial material for the machine learning school 2021. This school took place in the week commencing 15 Feburary 2021. Lecture recordings, where taken, are available at https://youtube.com/playlist?list=PLb8u5qKr67D3oTDELd6DXX-OWKYiZjQbr.
This School is a collaboration of working group 4 as part of the League of Advanced European Neutrons Sources (LENS) but was also supported by the STFC's Scientific Machine Learning Group (SciML) group and the Jülich Supercomputing Centre (JSC).
o To use this material, if you are unfamiliar with git we reccommend donloading the entire repository (the green code button and download .zip file)
o To run the tutorials on colab (https://colab.research.google.com/) you will need a google acount, then select github when prompted for a notebook and insert this repository (https://github.com/jfkcooper/LENS_ML_School_2021) which should then find all of the notebooks
o Alot of the notebooks have "lecturer editions" with answers, or answers hidden at the bottom of the page if you get stuck
o A slack workspace has also been created for this school (https://join.slack.com/t/lensmlschool2021/shared_invite/zt-m5hi20cj-NoriZQbku~BuDQgge~BG8A) pleae join the conversation
Lecture 1: Introduction to deep learning and neural networks (Jos Cooper)
o Terminology
o The perceptron
o Fundamentals of deep learning: neural networks, nodes, weights, biases, activation functions, backpropogation and some of the maths behind it
o Introduction to Tensorflow, Pytorch, and Keras
Lecture 2: Dense neural networks and regression (Jos Cooper)
o Supervised learning
o Epochs, metrics, batch processing
o Training, validation, testing, prediction
Lecture 3: Convolutional neural networks and classification (Emmanouela Rantsiou)
o Filters, convolution, layers
o Connections, activations, down sampling
o Training, classification, metrics
o Pre-processing
o Augmentation, regularization
o Hyper-parameter tuning
o Transfer learning
Lecture 4: Traditional ML methods (Andrew McCluskey)
o Decision trees
o Gradient boosting
o Principle component analysis (PCA)
o Bayesian model selection
Lecture 5: Image segmentation (Anders Kaestner)
o Object detection
o Tomography
o SegNet and/or ResNet
o Semi-supervised learning
Lecture 6: Recurrent neural networks (Gagik Vardanyan)
o Time series
o Simple RNNs
o LSTMs
o GRUs
Lecture 7:Generative Adversarial Networks, GANs (Kuangdai Leng)
o Introduction to generative models: VAEs and GANs
o GANs: basics and practice
Lecture 8: Natural language processing and speech recognition (Gagik Vardanyan & Guanghan Song)
o Semantic space, word-to-vec
o NNTK, spacey
o Machine translation, seq-to-seq methods
Lecture 9: Uncertainty and attention (Mario Teixeira Parente)
o Bayesian methods
o Gaussian attention / spatial transformers
Lecture 10: Unsupervised learning - clustering (Marina Ganeva)
o Introduction
o Clustering
o Manifold learning
o Reinforcement learning