Research map / Probabilistic and scientific modeling / Variational and generative sampling
Tensor Monte Carlo: Particle Methods for the GPU era
Laurence Aitchison
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use TMC instead of IWAE with the same particle count | MNIST multi-layer VAE · MNIST | variational lower bound: 20 -> 20^5 | Improved |
| Use TMC instead of IWAE | Gaussian latent-variable toy model · Gaussian toy experiment, N=128 | log-marginal likelihood bound | Improved |
| Use TMC instead of SMC | Gaussian latent-variable toy model · Gaussian toy experiment | log-marginal likelihood bound | Improved |
| Use TMC instead of SMC | Gaussian latent-variable toy model · Gaussian toy experiment | runtime | Improved |
| Use a non-factorised rather than factorised proposal (ablation) | TMC · Correlated Gaussian latent chain, N=100 | log-marginal likelihood bound | Improved |
| Train with TMC rather than IWAE | MNIST multi-layer VAE · MNIST | variational lower bound | Improved |
| Evaluate an IWAE-trained model with TMC rather than IWAE (ablation) | MNIST multi-layer VAE trained with IWAE · MNIST | variational lower bound | Improved |
| Add STL to TMC (ablation) | MNIST multi-layer VAE · MNIST | variational lower bound | Hurt |
| Add DReGs to TMC (ablation) | MNIST multi-layer VAE · MNIST | variational lower bound | Hurt |
| Add DReGs to TMC (ablation) | MNIST multi-layer VAE · MNIST | runtime | Hurt |
| Increase recognition-model capacity from factorised or small non-factorised to large non-factorised (ablation) | TMC and IWAE · MNIST | TMC-bound advantage over IWAE | Mixed |
Closest papers by meaning
- Monte Carlo Variational Auto-Encoders (ICML 2021 · similarity 0.87)
- Learning Energy-Based Model with Variational Auto-Encoder as Amortized Sampler (AAAI 2021 · similarity 0.86)
- Expected Tensor Decomposition with Stochastic Gradient Descent (AAAI 2016 · similarity 0.86)
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 2024 · similarity 0.85)
- Online Variational Sequential Monte Carlo (ICML 2024 · similarity 0.85)
- Tensorion: A Tensor-Aware Generalization of the Muon Optimizer (NeurIPS 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
