Personnal model implementations catalogue

Store implemented models, test scripts and ideas.

Project structure :

├── computer_vision
│   ├── object_detection
│   ├── inception
│   └── yolox
├── diffusion
│   ├── dae
│   ├── ddpm
│   ├── ppca
│   └── vae
├── fixed_income
├── mastery
│   └── Harmonic_Analysis_on_Paths_Spaces.pdf
├── numerical_exercises
│   ├── dealing_cards
│   ├── discrete_execution_optimisation
│   ├── positive_path_count
│   └── proba_negative_asset
├── pdevnet
└── utils
    └── polars_hdf5

Diffusion

Implement Denoising Diffusion Probabilitic Model (DDPM)

• jax and pytorch implementation with variants
• get familiar with lightning, tensorboard
• Ho, Jonathan, Ajay Jain, and Pieter Abbeel. "Denoising diffusion probabilistic models." Advances in neural information processing systems 33 (2020): 6840-6851.

Variational Auto-Encoder

• Sparse and convolutional variants
• Bernoulli or Gaussian distributions
• get familiar with mlflow, pytorch-lion

Probabilistic PCA

• Link with sample generation
• Linear algebra numerical implementation : SVD, bidiagonalisation
• get familiar with pytest
• compare with the EM training

Test with other celebrated models

• Rombach, Robin, et al. "High-resolution image synthesis with latent diffusion models." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022.
  • HuggingFace reference : https://huggingface.co/stabilityai/stable-diffusion-2-1
• Guo, Yuwei, et al. "Animatediff: Animate your personalized text-to-image diffusion models without specific tuning." arXiv preprint arXiv:2307.04725 (2023).
  • HuggingFace reference : https://huggingface.co/guoyww/animatediff-motion-adapter-v1-5-2

Computer Vision

Object detection streaming pipeline

Leverage the opencv Python API to build an object detection pipeline with horizon line detection.

Inception block POC

Implement minimal inception network applied to CIFAR 10 classification.

YOLOX inference

Test YOLOX algorithm for bounding box detection.

Path Development

• Leverage path development inside models
• Application to time-series classification

Mastery

• Litterature review on path developments

• Add example scripts

Fixed Income

• Validate intuition on the difference between PV01 and DV01
• WTI and SPX correlation
• FX pricer with "algebra" design pattern

Numerical implementation

Implement in Rust numerical verification of math questions.

• Counting discrete positive paths starting and ending at zero,
• Probability of a random walk to reach zero under time constraint,
• Optimal strategy to minimize a capped uniform law sequentially under constraints,
• List of other formal brainteasers including probabilities and combinatorics questions of interest.

---

References

• https://documentation.ubuntu.com/wsl/en/latest/howto/gpu-cuda/