2024 IAIFI Summer School

1. Deep generative models: A latent variable model perspective

(Author: Gilles Louppe)

Instructor: Gilles Louppe Slides: https://glouppe.github.io/iaifi-summer-school-2024/lecture/

Useful materials

Lecture Notes:

• Gilles Louppe, Deep Learning lecture notes, 2023-2024 (lectures 10, 11 and 12).
• Siddarth Mishra-Sharma, Generative modeling, with connection to and applications in physics, 2023.
• Karsten Kreis, Ruiqi Gao, Arash Vahdat, Latent Diffusion Models: Is the Generative AI Revolution Happening in Latent Space?, 2023.

Books:

• Jakub M. Tomczak, Deep generative modeling, 2022.
• Simon J.D. Prince, Understanding deep learning, 2023 (chapters 16, 17, 18).

Code:

• Siddarth Mishra-Sharma, Minified generative models: A repository with minimal/pedagogical implementations of some generative models.

Tutorial

(Author: Gaia Grosso)

Instructor: Gaia Grosso Notebook 1: Diffusion models for galaxy images generation Notebook 2: Variational auto-encoders for anomaly detection at the LHC Notebook 3: How good is your generative model?

2. Geometric Machine Learning

(Author: Melanie Weber)

Useful materials

Tutorial

(Author: Thomas Harvey & Sokratis Trifinopoulos)

• Tutorial Materials

  • The exercise notebooks can be found here: https://github.com/iaifi/summer-school-2024/tree/main/Geometric_NNs

• Reference papers and book chapters for tutorials:

  • Equivariant machine learning structured like classical physics: https://arxiv.org/pdf/2106.06610

3. Scaling and renormalization in high-dimensional regression

(Author: Cengiz Pehlevan & Alex Atanasov)

Useful materials

Tutorial

(Author: Alex Atanasov)

• Tutorial Materials

  • Plase see the folder: https://github.com/iaifi/summer-school-2024/tree/main/RMT_Scaling_Laws

Reference papers and book chapters for tutorials

Key paper:

• Scaling and Renormalization in High Dimensional Regression https://arxiv.org/abs/2405.00592

Related work:

• Dynamical Model of Neural Scaling Laws: https://arxiv.org/abs/2402.01092
• Spectral Bias and Task-Model Alignmeny Explain Generalization: https://arxiv.org/abs/2006.13198

Recommended Books:

• Potters and Bouchaud "A first course in Random Matrix Theory". Strongly recommended!

4. Uncertainty Quantification

(Author: Carol Cueta-Lazaro)

Useful materials

Tutorial

(Author: Jessie Micallef)

• Tutorial Materials

  • Intro slides: https://github.com/jessimic/sbi-tutorial-iaifi/blob/main/assets/SBI_Tutorial_Intro.pdf
  • Tutorials: https://github.com/jessimic/sbi-tutorial-iaifi/tree/a8730f55642b7280c3c3bc1c33c8483af78f8305
    • Main branch has fill in code blanks
    • Answers branch has some solutions, including all solutions for Tutorial 1

• Reference papers and book chapters for tutorials

  • The frontier of simulation-based inference (Cranmer, Brehmer, Louppe): Review paper
  • simulation-based-inference.org: List of papers and resources
  • awesome-neural-sbi: List of papers and resources

5. Hackathon

At the end of the Hackathon on Friday, August 9, we will have a block for presentations of work done on these topics. Forming groups is strongly encouraged!

Prompts

• Train a variational auto-encoder on calorimeter data and use it for anomaly detection and/or for super-resolution of astronomical images.
• Train a diffusion model on galaxy images and generate samples conditionally on noisy observations (e.g., corrupted by noise, missing pixels, etc).
• Train a normalizing flow for simulation-based inference on gravitational lenses (e.g., substructure properties in strong lensing systems).
• Apply simulation-based inference (SBI) to a new dataset (examples below). Can we use different ML methods for SBI outside of dense NNs and CNNs? Transformer? Other methods?
  • Astronomy
  • Breast cancer histology
  • Drift tube chamber data
• Quantum reservoir computing: How does the QRC perform if we only collect 〈Zi〉 expectation values and drop 〈ZiZj〉 correlations? Why do you think this happens? How about adding connected 〈ZiZjZk〉 where {i, j, k} belong to a cluster of sites connected by the nearest neighbor bonds? Hint: Try to modify the readouts vector to exclude/include the correlations that are calculated. Based on this notebook: https://github.com/QuEraComputing/QRC-tutorials/blob/main/QRC%20Demo%20MNIST.ipynb
• Work on your own project!

Datasets

• Astro: https://camels-multifield-dataset.readthedocs.io/en/latest/
• IceCube dataset: https://www.kaggle.com/competitions/icecube-neutrinos-in-deep-ice
• Breast cancer histology: https://www.kaggle.com/datasets/paultimothymooney/breast-histopathology-images
• Drift tube chamber data: https://zenodo.org/records/7128223
  • Associated paper about data quality monitoring for this dataset: https://iopscience.iop.org/article/10.1088/2632-2153/acebb7/pdf
• Calorimeter data (Calo challenge): https://github.com/CaloChallenge/homepage
• Gravitational lens dataset: https://lweb.cfa.harvard.edu/castles/noimages.html
• Galaxies (COSMOS real galaxies dataset): https://zenodo.org/records/3242143

Prize Categories

• Best project (effort, presentation, use of summer school topics): 1st, 2nd, and 3rd place
• Best visualization
• Best team effort

NSF ACCESS Instructions

https://github.com/alexandergagliano/summer-school-2024/tree/main/computing