Research map / Probabilistic and scientific modeling / Variational and generative sampling
Latent Autoregressive Source Separation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 LASS instead of BASIS Glow | MNIST source separation · MNIST test | PSNR: 22.7 -> 24.2 | Helped |
| Use LASS instead of BASIS NCSN | CelebA source separation · CelebA test | FID: 7.55 -> 8.96 | Hurt |
| Use LASS instead of blind unsupervised baselines | SLAKH2100 Drums + Bass separation · SLAKH2100 test | SDR (Avg): 1.04 -> 4.86 | Helped |
| Use LASS instead of supervised Conv-Tasnet | SLAKH2100 Drums + Bass separation · SLAKH2100 test | SDR (Avg): 5.47 -> 4.86 | Hurt |
| Use LASS instead of BASIS NCSN | MNIST source separation · MNIST | inference time: 53.34 s $\pm$ 0.51 s -> 4.49 s $\pm$ 0.27 s | Helped |
| Use LASS instead of PnF | SLAKH2100 3-second source separation · SLAKH2100 | inference time: 42.29 min $\pm$ 1.08 s -> 1.33 min $\pm$ 0.87 s | Helped |
| Use stochastic sampling rather than greedy or beam-search MAP methods (ablation) | LASS · MNIST validation | PSNR: 16.96 $\pm{}$ 5.78 -> 24.23 $\pm{}$ 6.23 | Helped |
| Use beam search rather than greedy sampling (ablation) | LASS · SLAKH2100 validation | SDR: 1.23 $\pm{}$ 2.33 -> 5.01 $\pm{}$ 2.39 | Helped |
| Use top-k sampling with k = 32 rather than ancestral sampling (ablation) | LASS · MNIST validation | PSNR: 24.03 $\pm{}$ 6.37 -> 24.23 $\pm{}$ 6.23 | No clear effect |
| Use top-k sampling with k = 32 rather than k = 16 (ablation) | LASS · MNIST validation | PSNR: 23.74 $\pm{}$ 6.55 -> 24.23 $\pm{}$ 6.23 | No clear effect |
| Use top-k sampling with k = 32 rather than k = 64 (ablation) | LASS · MNIST validation | PSNR: 23.85 $\pm{}$ 6.13 -> 24.23 $\pm{}$ 6.23 | No clear effect |
Closest papers by meaning
- Anytime Sampling for Autoregressive Models via Ordered Autoencoding (ICLR 2021 · similarity 0.89)
- Outsourced Diffusion Sampling: Efficient Posterior Inference in Latent Spaces of Generative Models (ICML 2025 · similarity 0.85)
- Stabilize the Latent Space for Image Autoregressive Modeling: A Unified Perspective (NeurIPS 2024 · similarity 0.85)
- Continuous Visual Autoregressive Generation via Score Maximization (ICML 2025 · similarity 0.84)
- Continuous Autoregressive Modeling with Stochastic Monotonic Alignment for Speech Synthesis (ICLR 2025 · similarity 0.84)
- Distilled Decoding 1: One-step Sampling of Image Auto-regressive Models with Flow Matching (ICLR 2025 · similarity 0.84)
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.