This paper introduces consistency models, a new family of generative models that map any point on a Probability Flow ODE trajectory to its origin. They support fast one-step generation by design, while allowing multi-step sampling for improved quality and zero-shot editing tasks like inpainting and colorization.
This paper introduces Discrete Denoising Diffusion Probabilistic Models (D3PMs), which generalize diffusion to discrete state-spaces using structured Markov transition matrices. D3PMs include uniform, absorbing-state, and discretized Gaussian corruption processes, drawing a connection between diffusion and masked language models.
This paper introduces Latent Diffusion Models (LDMs), which apply denoising diffusion in the latent space of pretrained autoencoders. By separating perceptual compression from generative learning and adding cross-attention conditioning, LDMs achieve FID 1.50 on Places inpainting and FID 3.60 on ImageNet class-conditional synthesis, with competitive text-to-image generation, at a fraction of the compute cost of pixel-space diffusion.
Proposes ‘InterFlow’, a method to learn continuous normalizing flows between arbitrary densities using stochastic interpolants. It avoids ODE backpropagation by minimizing a quadratic objective on the velocity field, enabling scalable ODE-based generation. On CIFAR-10, NLL matches ScoreSDE (2.99 bits per dim) with simulation-free training, though FID (10.27) trails dedicated image models (ScoreSDE: 2.92); the primary strength is tractable likelihood with efficient training cost.
Introduces Flow Matching, a scalable method for training CNFs by regressing vector fields of conditional probability paths. It generalizes diffusion and enables Optimal Transport paths for straighter, more efficient sampling.
Introduces ‘Rectified Flow,’ a method to transport distributions via ODEs with straight paths. Uses a ‘reflow’ procedure to iteratively straighten trajectories, enabling high-quality 1-step generation with optional lightweight distillation.
This paper provides a rigorous probabilistic foundation for Denoising Autoencoders by proving they are mathematically equivalent to Score Matching on a kernel-smoothed data distribution. It derives a specific energy function for DAEs and justifies the use of tied weights.
This paper unifies previous score-based methods (SMLD and DDPM) under a continuous-time SDE framework. It introduces Predictor-Corrector samplers for improved generation and Probability Flow ODEs for near-exact likelihood computation, setting new records on CIFAR-10.
Kingma and Welling’s 2013 paper introducing Variational Autoencoders and the reparameterization trick, enabling end-to-end gradient-based training of generative models with continuous latent variables by moving the stochasticity outside the computational graph so that gradients can flow through a deterministic path.
Burda et al.’s ICLR 2016 paper introducing Importance Weighted Autoencoders, which use importance sampling to derive a strictly tighter log-likelihood lower bound than standard VAEs, addressing posterior collapse and improving generative quality. The model architecture remains the same.
Discover how Importance Weighted Autoencoders (IWAEs) use the same architecture as VAEs with a fundamentally more powerful objective to leverage multiple samples effectively.
A NeurIPS 2021 method paper introducing Noise Contrastive Priors to address the VAE ‘prior hole’ problem, where standard Gaussian priors assign high density to regions of latent space that don’t correspond to realistic data, using energy-based models trained with contrastive learning to match the aggregate posterior.