Research map / Image generation / Fast diffusion sampling
Discrete Diffusion for Bundle Construction
Teng Tu, Ai Li, Yunshan Ma, Shuo Xu, Xiaohao Liu, Haokai Ma, Liang Pang, Tat-Seng Chua
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 |
|---|---|---|---|
| add residual vector quantization item codes instead of raw item IDs (ablation) | DDBC · Spotify k=30 | F1: 0.021 -> 0.282 | Improved |
| add the boi token (ablation) | DDBC · Spotify k=30 | F1: 0.176 -> 0.282 | Improved |
| add item-swap data augmentation (ablation) | DDBC · Spotify k=30 | F1: 0.254 -> 0.282 | Improved |
| add token validity filter at inference (ablation) | DDBC · Spotify k=30 | F1: 0.276 -> 0.282 | Improved |
| increase RVQ levels from 1 to 4 (ablation) | DDBC · Spotify k=30 | F1: 0.182 -> 0.282 | Improved |
| increase reverse diffusion steps from 1 to 50 (ablation) | DDBC · Spotify k=30 | F1: 0.2693 -> 0.2850 | Improved |
| increase diffusion steps from 50 to 300 (ablation) | DDBC · Spotify k=30 | F1: 0.2850 -> 0.2816 | No clear effect |
| observe suffix instead of prefix items (ablation) | DDBC · Spotify k=30 | F1: 0.2821 -> 0.2777 | No clear effect |
| use DDBC instead of autoregressive TIGER on a short bundle | DDBC · POG k=4 | F1: 0.213 -> 0.139 | Hurt |
| use DDBC instead of best baseline on long bundles | DDBC · Spotify k=90 | Jaccard: 0.070 -> 0.177 | Improved |
| generate 15 slots instead of trained length 30 without retraining (ablation) | DDBC · Spotify k=30 | F1: 0.2821 -> 0.2066 | Hurt |
| semi-AR decoding with 1 item per pass instead of 30 (ablation) | DDBC · Spotify k=30 | F1: 0.2821 -> 0.1244 | Hurt |
Closest papers by meaning
- Disentangled Contrastive Bundle Recommendation with Conditional Diffusion (AAAI 2025 · similarity 0.83)
- Compositional Discrete Latent Code for High Fidelity, Productive Diffusion Models (NeurIPS 2025 · similarity 0.81)
- Primal-Dual Guided Decoding for Constrained Discrete Diffusion (NeurIPS 2026 · similarity 0.80)
- Diffusion Models Encode the Intrinsic Dimension of Data Manifolds (ICML 2024 · similarity 0.80)
- DisCo-Diff: Enhancing Continuous Diffusion Models with Discrete Latents (ICML 2024 · similarity 0.80)
- Optimization Can Learn Johnson Lindenstrauss Embeddings (NeurIPS 2024 · similarity 0.80)
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.
