Pith. sign in

REVIEW 21 cited by

Tackling the Generative Learning Trilemma with Denoising Diffusion GANs

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2112.07804 v2 pith:NOITT5AN submitted 2021-12-15 cs.LG stat.ML

Tackling the Generative Learning Trilemma with Denoising Diffusion GANs

classification cs.LG stat.ML
keywords denoisingdiffusionmodelsgansgenerativemodelsamplesampling
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

A wide variety of deep generative models has been developed in the past decade. Yet, these models often struggle with simultaneously addressing three key requirements including: high sample quality, mode coverage, and fast sampling. We call the challenge imposed by these requirements the generative learning trilemma, as the existing models often trade some of them for others. Particularly, denoising diffusion models have shown impressive sample quality and diversity, but their expensive sampling does not yet allow them to be applied in many real-world applications. In this paper, we argue that slow sampling in these models is fundamentally attributed to the Gaussian assumption in the denoising step which is justified only for small step sizes. To enable denoising with large steps, and hence, to reduce the total number of denoising steps, we propose to model the denoising distribution using a complex multimodal distribution. We introduce denoising diffusion generative adversarial networks (denoising diffusion GANs) that model each denoising step using a multimodal conditional GAN. Through extensive evaluations, we show that denoising diffusion GANs obtain sample quality and diversity competitive with original diffusion models while being 2000$\times$ faster on the CIFAR-10 dataset. Compared to traditional GANs, our model exhibits better mode coverage and sample diversity. To the best of our knowledge, denoising diffusion GAN is the first model that reduces sampling cost in diffusion models to an extent that allows them to be applied to real-world applications inexpensively. Project page and code can be found at https://nvlabs.github.io/denoising-diffusion-gan

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 21 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    cs.LG 2022-09 unverdicted novelty 8.0

    Rectified flow learns straight-path neural ODEs for distribution transport, yielding efficient generative models and domain transfers that work well even with a single simulation step.

  2. Safe Few-Step Generation via Velocity Editing

    cs.CV 2026-06 unverdicted novelty 7.0

    VESFlow edits the learned velocity field of flow matching models via a safe-conditional posterior to produce safe images in 4 sampling steps, with an optional risk filter and VESFlow+ variant that also repels from uns...

  3. The Reward Was in Your Data All Along: Correcting Flow Matching with Discriminator-Guided RL

    cs.LG 2026-06 unverdicted novelty 7.0

    DRL trains a discriminator on data versus base-model samples in pretrained representation space and uses its logit as reward in KL-regularized RL, cutting guidance-free FID from 9.38 to 2.62 on SiT and similar gains o...

  4. A Systematic Benchmark of Intraoperative Ultrasound-to-MR Synthesis for Brain Tumour Surgery

    cs.CV 2026-05 conditional novelty 7.0

    On the public ReMIND dataset, a systematic benchmark of six synthesis models across 48 experiments finds LPIPS correlates with downstream segmentation utility while SSIM does not, with SynDiff-2.5D performing best.

  5. IMPLICITSTAINER: Resolution Agnostic Data-Efficient Virtual Staining Using Neural Implicit Functions

    eess.IV 2025-05 unverdicted novelty 7.0

    Neural implicit functions enable resolution-agnostic, deterministic virtual staining from H&E to IHC images with SOTA results and better low-data performance than patch-based GAN or diffusion methods.

  6. Diffusion Policies as an Expressive Policy Class for Offline Reinforcement Learning

    cs.LG 2022-08 unverdicted novelty 7.0

    Diffusion-QL uses conditional diffusion models as expressive policies in offline RL by coupling behavior cloning with Q-value maximization, achieving SOTA on most D4RL tasks.

  7. ROMS-IMLE: A Minimalist Approach to Competitive Single-Step Generative Modelling

    cs.LG 2026-07 conditional novelty 6.0

    A single-step IMLE generator with per-stage supervision and a robust loss reports FID 2.56 on ImageNet-256 by filtering ~5% of samples at test time.

  8. Efficient Diffusion Distillation via Embedding Loss

    cs.CV 2026-04 unverdicted novelty 6.0

    Embedding Loss aligns feature distributions via MMD in random network embeddings to boost one-step diffusion distillation, reaching SOTA FID of 1.475 on CIFAR-10 unconditional generation.

  9. DAG-STL: A Hierarchical Framework for Zero-Shot Trajectory Planning under Signal Temporal Logic Specifications

    cs.RO 2026-04 unverdicted novelty 6.0

    DAG-STL decomposes long-horizon STL planning into decomposition, timed waypoint allocation, and diffusion-based trajectory generation to enable zero-shot planning under unknown dynamics.

  10. MiAD: Mirage Atom Diffusion for De Novo Crystal Generation

    cs.LG 2025-11 conditional novelty 6.0

    Mirage infusion lets crystal diffusion models vary atom counts during generation and raises the S.U.N. rate on MP-20 to 8.2%.

  11. Friend or Foe

    q-bio.QM 2025-08 conditional novelty 6.0

    Friend or Foe is a 64-dataset compendium of 26M+ simulated bacterial interaction environments, with benchmarks showing deep tabular models classify interaction type with mean MCC of about 0.64.

  12. Quantum latent distributions in deep generative models

    cs.LG 2025-08 conditional novelty 6.0

    Quantum latent distributions from boson samplers are shown in theory to expand the output distribution class of invertible Lipschitz generators, and in GAN benchmarks on QM9 to beat Gaussian, Bernoulli, and distinguis...

  13. DOLLAR: Few-Step Video Generation via Distillation and Latent Reward Optimization

    cs.CV 2024-12 unverdicted novelty 6.0

    DOLLAR combines variational score and consistency distillation for few-step video generation plus latent reward optimization, reporting 82.57 VBench score and up to 278x speedup over the teacher diffusion model for 12...

  14. Rectified Flow: A Marginal Preserving Approach to Optimal Transport

    stat.ML 2022-09 unverdicted novelty 6.0

    A single-objective rectified flow variant uses neural ODEs trained by regression to monotonically decrease a fixed convex transport cost while preserving marginal distributions.

  15. Layer-Parallel Inference Reduces Encrypted Nonlinear Depth in Transformers

    cs.LG 2026-07 conditional novelty 5.5

    SNLP reduces encrypted Transformer nonlinear depth from L sequential stages to (L−N)+K, cutting symbolic bootstraps ~2.65× with lower error amplification than sequential inference.

  16. Explorative Modeling: Unlocking a Third Pretraining Axis and End-to-End Generation

    cs.LG 2026-07 conditional novelty 5.0

    Training on the best of K generated candidates improves image, video, and language generative models, with the reported gains growing with scale and enabling single-pass end-to-end generation.

  17. Layer-Parallel Inference Reduces Encrypted Nonlinear Depth in Transformers

    cs.LG 2026-07 conditional novelty 5.0

    SNLP reduces symbolic FHE bootstraps from 53 to 20 on a 0.5B model with +1.2% PPL degradation and lower polynomial-error amplification than sequential inference.

  18. FAIL: Flow Matching Adversarial Imitation Learning for Image Generation

    cs.CV 2026-02 conditional novelty 5.0

    Post-training of flow matching can be framed as adversarial imitation learning, and the proposed FAIL methods improve FLUX's generation quality using 13K expert images without preference pairs.

  19. The Score-Difference Flow for Implicit Generative Modeling

    cs.LG 2023-04 unverdicted novelty 5.0

    Score-difference flow reduces KL divergence between distributions and is formally equivalent to denoising diffusion models and a hidden subproblem in optimal GAN training under stated conditions.

  20. Efficient Difficulty-Aware Dynamic Routing for Diffusion-Based Real-World Image Super-Resolution

    cs.CV 2026-07 reject novelty 4.0

    DDR-SR routes each real-world low-resolution image to one of two diffusion experts based on a high-frequency-loss difficulty score, using a low-compression VAE for hard images and a high-compression VAE for easy image...

  21. Reinforcement Learning: From Algorithms To Foundation Models

    cs.AI 2026-07 conditional novelty 3.0

    A dissertation uniting the author's published results: non-exploitable Nash-DQN policies and the FightLadder benchmark for games, plus diffusion/consistency-model world models for RL — a compilation rather than new results.