Advanced Machine Learning for Physics (PhD 2023)

Course Information and Syllabus 

Contacts: Stefano Giagu (stefano.giagu [at] uniroma1.it) and Andrea Ciardiello (andrea.ciardiello [at] gmail.com)

Program:

The general objective of the course is to become familiar with advanced deep learning techniques based on differentiable neural network models with different learning paradigms; acquire skills in modeling complex problems, through deep learning techniques, and be able to apply them in different contexts in the fields of physics, basic and applied scientific research.

Topics covered include: general overview of differentiable artificial neural networks and use of the pytorch library for ANN design, training and testing. Basic architectures: MLP, Convolutional neural network, neural network for sequence analysis (RNN, LSTM/GRU). Bayesian-NN. Attention, Self-Attention, Transformers and Visual Transformers, Models for object detection and semantic segmentation and applications. Graph Neural Networks and Geometrical Deep Learning. Generative models based on VAE, GAN, autoregressive models, invertible networks, diffusion models, normalising flow, and generative GNNs. Advanced learning techniques:  transfer learning, domain adaptation, adversarial learning, self-supervised and contrastive learning, model distillation.  Explainable and interpretable AI. Quantum Machine Learning on near-term quantum devices.

Approximately 50% of the lectures are frontal lessons supplemented by slide, aimed at providing advanced knowledge of Deep Learning techniques. The remaining 50% is based on hands-on computational practical experiences that provide some of the application skills necessary to autonomously develop and implement advanced Deep Learning models for solving various problems in physics and scientific research in general.

Indispensable prerequisites: basic concepts in machine learning, python language programming, standard python libraries (numpy, pandas, matplotlib, torch/pytorch )

• a basic python course on YT (many others available on web: https://youtu.be/_uQrJ0TkZlc 

• tutorial on numpy, matplotlib, pandas: https://jakevdp.github.io/PythonDataScienceHandbook/)

• basic concepts of ML:  Introduction + Part I (sec. 5: ML basics) of the book I. Goodfellow et al.: https://www.deeplearningbook.org/

• tutorials on pytorch web site: https://pytorch.org/

• an introductory course on pytorch on YT (many others available on web): https://youtu.be/c36lUUr864M

Depending on the requirements of your specific PhD course each students can decided how may lectures/hands-on to attend tu fulfil the required hours:  20h, 40h, 60h (60h corresponds to the whole course).

Discussion group (google group):

• https://groups.google.com/g/aml_2023

Calendar:

Lectures in Aula 6 - Dep. of Physics (E. Fermi) and at this google meet link

Lectures in Aula 3  - Dep. of Physics (E. Fermi) and at this google meet link

Hands-on sessions in LabSS - Dep. of Physcs (G. Marconi, 1st floor) and at this google meet link

• 28.2 - 17:00-18:30: course syllabus and introduction

• 1.3 - 16:00-18:00: recall on ANN

• 7.3 - 14:00-16:00: recall on ANN

• 8.3 - 16:00-18:00: hands-on on use of pytorch

• 14.3 - 14:30-16:00: learning paradigms

• 15.3 - 16:00-18:00: hands-on on residual learning 

• 21.3  - 14:00-16:00: hands-on on contrastive learning  

• 22.3 - 16:00-18:00: hands-on on contrastive learning

• 28.3 - 14:00-16:00: attention and transformers 

• 29.3 - 16:00-18:00: hands-on on transformers

• 4.4  - 14:00-16:00 hands-on on transformers 

• 5.4 - 16:00-18:00: segmentation and object detection models

• 12.4 - 16:00-18:00: hands-on on semantic segmentation

• 18.4 - 14:00-16:00: Graph Neural Networks

• 19.4 - 16:00-18:00: hands-on on object detection

• 26.4 - 16:00-18:00: hands-on on object detection

• 2.5  - 14:00-16:00: AI explainability and interpretability

• 3.5 - 16:00-18:00: hands-on on GNNs

• 9.5  - 14:00-16:00: hands-on on GNNs

• 10.5 - 16:00-18:00: hands-on on GNNs

• 16.5 - 14:00-16:00: hands-on on xAI

• 17.5 - 16:00-18:00: hands-on on xAI/knowledge-transfer

• 23.5 - 14:00-16:00: anomaly detection

• 24.5 - 16:00-18:00: hands-on on anomaly detection

• 30.5 - 14:00-16:00: generative DL

• 31.5 16-18: generative DL (online only at this google meet link)

• 6.6 - 14:00-16:00: hands on on generative DL (normalizing flow models)

• 7.6 - 16:00-18:00: hands on on generative DL (deep diffusion probabilistic models)

• 15.6 - 16:00-18:00: introduction to Quantum computation and Quantum ML and hands on on Quantum ML

Bibliography/References and detailed topics treated during lectures, slides, notebooks, etc.

Given the highly dynamic nature of the topics covered in the course, there is no single reference text. During the course the sources will be indicated and provided from time to time in the form of scientific and technical articles and book chapters.

Some classic readings on Deep Learning based on differentiable neural networks:

• DL: I. Goodfellow, Y. Bengio, A. Courville: Deep Learning, MIT Press (https://www.deeplearningbook.org/)

• PB: P. Baldi, Deep Learning in Science, Cambridge University Press

• GRL: W. L. Hamilton, Graph Representation Learning Book, MCGill Uni press (https://www.cs.mcgill.ca/~wlh/grl_book/files/GRL_Book.pdf)

• lecture 1 - 28.2.2023 (slides, recording) h17:00-19:00  

  • course information

  • google colab and aws sagemaker studio lab clouds

  • artificial neural networks 101: (DL ch 6 (6.1,6.2,6.3, 6.4))

    • artificial neuron model  and XOR gate example

    • MLPs and implementation of a Linear layer in pytorch

    • activations functions for hidden and output layers)

    • weight initialisation

    • universal approximation theorem for ANN

  • a simple example of a MLP implemented in pytorch

• lecture 2 - 1.3.2023 (slides, recording) h16:00-18:00

  • ANN 101: (DL ch 6 (6.1,6.2,6.3, 6.4, 6.5), BIS ch 5 (5.1, 5.2,5.3, 5.5), DUDA ch 6 (6.1, 6.2, 6.3, 6.24, 6.5, 6.8))

    • training of an ANN

    • loss functions

    • SGD, momentum, learning rate, variable lr and optimizers

    • learning curves, bias-variance tradeoff and double descent in DNN

    • regularisation 

      • dropout

      • early stopping

      • noise injection

      • data augmentation

      • weight regularisation L1, L2, L1+L2

  • a simple example of a MLP implemented in pytorch

• lecture 3 - 7.3.2023 (slides, recording) h14:00-16:00

  • Convolutional-NN (SL 9 (9.1, 9.2, 9.3, 9.4, 9.7, 9.8, 9.9, 9.10, 9.11) 

    • image representation and input properties of a CNN (symmetry, translation invariance, self-similarity, compositionality, locality) and learned convolutional filters (DL ch 9 (9.1,9.2, 9.4))

    • local receptive field

    • convolution e shared weights

    • pooling layers

    • CNN architectures: LeNet, AlexNet, VGG, Inception, ResNet, DenseNet, ...

  • Sequence analysis: task definition and problems (DL ch 10 (intro, 10.2))

    • Elementary RNN cell: structure and operating principle

    • StackedRNN, Bidirectional RNN, Encoder-Decoder RNN (seq2seq)

    • Back-propagation through time

    • Long-term correlation and gradient vanishing and exploding problems: Gated Cell and Long Short Term Memory RNN

    • LSTM: description of operations (note by C.Olah and for details DL ch 10 (10.10))

• Hands-on 1 - 8.3.2024 (notebook, recording) h16:00-18:00

  • use of pytorch to design and train a ConvNet for identfication of particles in a RICH detector

• Lecture 4 - 14.3.2023 (slides, recording) h14:00-16:00

• Hands-on 2 - 15.3.2023 (notebook, recording)  h16:00-18:00

  • denoising CNN based on residual learning for refining simulated fast raman spectra

• Hands-on 3 - 21.3.2023 (notebook, recording) h14:00-16:00

  • self supervised contrastive learning SimCLR model for auroral identification and classification (part 1)

• Hands-on 3bis - 22.3.2023 (notebook, recording) h16:00-18:00

  • self supervised contrastive learning SimCLR model for auroral identification and classification (part 2)

• Lecture 5 - 28.3.2023 (slides, recording) h14:00-16:00

  • attention mechanism

  • the RNNSearch encoder-decoder model (arXiv:1409.0473)

  • attention and the Nadaraya-Watson kernel estimator

  • attention layers vs fully connected layers

  • transformer architecture (arXiv:1706.03762)

  • (masked) multi head (self) attention based on scaled dot product

  • layer normalization

  • positional embedding

  • modern evolutions GPT/GPT2/GPT3/BERT

  • vision transformer (arXiv:2010.11929)

• Hands-on 4 - 29.3.2023 (notebook, recording) h16:00-18:00

  • a transformer encoder architecture for jet tagging

• Hands-on 4bis - 4.4.2023 (notebook, recording) h14:00-16:00

  • a transformer encoder architecture for jet tagging + a ViT architecture trained for the same task

• Lecture  6 - 5.4.2023 (slides, recording) h16:00-18:00

  • neural architectures for object detection and segmentation:

    • semantic segmentation, downsampling-upsampling (arXiv:1411.4038., arXiv:1505.04366)

    • object detection, IoU, anchor boxes, non-max supression

    • region proposals, R-CNN, Fast and Faster R-CNN (arXiv:1311.2524,  arXiv:1506.01497 )

    • Yolo model

    • Instance segmentation, Mask R-CNN (arXiv:1703.06870)

    • Pose

• Hands-on 5 - 12.4.2023 (notebook, recording) h16:00-18:00

  • segmentation of MRI images using the MONAI framework for medical imaging

• Lecture 7 - 18.4.2023 (slides, recording) h14:00-16:00

  • Graph Neural Networks (GRL section 5 and 6, PyTorch geometric web site)

    • introduction

    • graphs and representation

    • permutation equivariance

    • graph convolutions and message passing

    • basic GCN layer

    • self-loop GCN

    • graph attention networks

    • solutions to the over-smoothing problem in GNNs

    • normalization

    • GNN python libraries

• Hands-on 6  - 19.4.2023 (notebook, recording) h16:00-18:00

• object detection with the YOLO V3 model - part 1

• Hands-on 6bis  - 26.4.2023 (notebook, recording) h14:00-16:00

  • object detection with the YOLO V3 model - part 2

• Lecture 8 - 2.5.2023 (slides, recording) h14:00-16:00

  • AI explainability and interpretability

• Hands-on 7 - 3.5.2023 (notebook, recording) h16:00-18:00

  • graph neural network with pytorch geometric

• Hands-on 7bis - 9.5.2023 (notebook, recording) h14:00-16:00

  • GNN for point cloud classification (part 1)

• Hands-on 7tris - 10.5.2023 (notebook, recording) h16:00-18:00

  • GNN for point cloud classification (part 2)

• Hands-on session 8 - 16.5.2023 (notebook, recording) h14:00-16:00

  • xAI methods 

• Hands-on session 8bis - 17.5.2023 (notebook, recording) h16:00-18:00

  • xAI methods (part 2)

• Lecture 9 - 23.5.2023 (slides, recording) h14:00-16:00

  • AutoEncoders (DL ch 14)

    • under-complete auto-encoders

    • linear AE and PCA

    • over-complete AEs

    • denoising AE

    • sparse AE

    • contractive AE

    • AE for self-supervised anomaly detection

    • examples of implementation in pytorch

• Hands-on session 9 - 24.5.2023 (notebook, recording) h16:00-18:00

  • anomaly detection

• Lecture 10 - 30.5.2023 (slides, recording) h14:00-16:00

  • generative DL (DL ch. 20)

    • autoregressive models

    • latent variable models

      • VAE, ELBO theorem

      • Generative Adversarial Networks

• Lecture 11 - 31.5.2023 (slides, recording) h16:00-18:00

  • generative DL part 2 (DL ch. 20)

    • flow model: normalising flow models

    • deep diffusion probabilistic models

• Hands-on session 10 - 6.6.2023 (notebook, recording)  h14:00-16:00

  • normalising flow model implementation 

• Hands-on session 10bis - 7.6.2023 (notebook)  h16:00-18:00

  • deep diffusion probabilistic comedy implementation 

• Lecture 12 and hands-on session 11 - 15.6.2023 (slides,  notebook, recording)  h16:00-18:00

  • basic introduction to Quantum Machine Learning (SP ch. 1, 3, 5)

  • implementation of simple models with the pennylane library 

• Special seminar from Sergio Orlandini on HPC DL - 23.6.2023 (slides)