Diffusion Models - PyTorch Implementation
Learn to implement and train diffusion models in PyTorch, covering unconditional and conditional versions, classifier-free guidance, and exponential moving average. Gain hands-on experience with state-of-the-art generative AI techniques.
Google Colaboratory and Python: Deep Learning Basics for Computer Vision Tasks
Master Google Colab for image processing while learning Python fundamentals, CNN architecture, and hands-on model training using CIFAR-10 dataset for computer vision tasks.
Premonitions of Public Data for Private ML
Explore how public data pre-training enhances private machine learning, examining its impact on accuracy, zero-shot learning, and the future of privacy-preserving AI techniques.
Reflected Diffusion Models: Principled Data Constraints in Generative Modeling
Explore reflected diffusion models for improved image generation. Learn about reversing stochastic differential equations, addressing numerical errors, and incorporating data constraints in principled ways.
One Pixel Adversarial Attacks via Sketched Programs
Explore a novel approach for one-pixel adversarial attacks using program synthesis, achieving state-of-the-art success rates with fewer queries and transferable programs for neural network classifiers.
Comparing Elixir and Python for Simple Neural Networks
Compare Elixir's Nx library with Python's machine learning ecosystem for training neural networks. Explore development experiences and performance differences using MNIST and CIFAR-10 datasets.
Differentially Private Diffusion Models Generate Useful Synthetic Images
Explore privacy-preserving synthetic image generation using diffusion models. Learn techniques for fine-tuning pre-trained models to achieve state-of-the-art results on CIFAR-10 and Camelyon17 datasets.
Elucidating the Design Space of Diffusion-Based Generative Models
Explore advancements in diffusion-based generative models, including improved sampling, training, and score network preconditioning. Learn about state-of-the-art results and modular design changes for enhanced efficiency and quality.
Splitting the Difference on Adversarial Training - A Method for Robust Neural Networks
Discover a novel approach to adversarial training that enhances neural network robustness while maintaining optimal accuracy by splitting data classes into clean and adversarial categories.