Research map / Image generation / Autoregressive and latent generation
Towards Model-Agnostic Dataset Condensation by Heterogeneous Models
Jun-Yeong Moon, Jung Uk Kim, Gyeong-Moon Park
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 |
|---|---|---|---|
| condense with heterogeneous models (ConvNet + ViT-tiny) instead of a single ConvNet | dataset condensation baselines (IDC, DREAM) · CIFAR-10 (IPC 10) | average accuracy across models: 48.17 -> 75.04 (DREAM vs HMDC) | Improved |
| add Gradient Balance Module and Mutual Distillation together (ablation) | HMDC · CIFAR-10 (IPC 10) | ViT average accuracy: 81.25 -> 82.64 | Improved |
| add only Gradient Balance Module (ablation) | HMDC · CIFAR-10 (IPC 10) | average accuracy across models: 74.05 -> 71.68 | Hurt |
| add only Mutual Distillation (ablation) | HMDC · CIFAR-10 (IPC 10) | average accuracy across models: 74.05 -> 73.60 | No clear effect |
| add both GBM and MD (ablation) | HMDC · CIFAR-10 (IPC 10) | average accuracy across models: 74.05 -> 74.85 | No clear effect |
| condense with ViT-tiny only instead of ConvNet only (ablation) | DREAM · CIFAR-10 (IPC 10) | accuracy on ConvNet: 47.62 -> 25.69 | Hurt |
| use HMDC instead of DREAM at IPC 1 | dataset condensation · CIFAR-10 (IPC 1) | average accuracy across models: 18.00 -> 48.87 | Improved |
| use HMDC at IPC 50 | dataset condensation · CIFAR-10 (IPC 50) | average accuracy across models: 79.13 (Random) -> 79.66 | No clear effect |
| condense with two heterogeneous models while needing only 100 iterations | HMDC · CIFAR-10 | iterations: 100 iterations versus 1,200 to 20,000 | Improved |
Closest papers by meaning
- M3D: Dataset Condensation by Minimizing Maximum Mean Discrepancy (AAAI 2024 · similarity 0.88)
- Elucidating the Design Space of Dataset Condensation (NeurIPS 2024 · similarity 0.87)
- Dataset Distillation for Pre-Trained Self-Supervised Vision Models (NeurIPS 2025 · similarity 0.86)
- D^4: Dataset Distillation via Disentangled Diffusion Model (CVPR 2024 · similarity 0.85)
- MGD$^3$ : Mode-Guided Dataset Distillation using Diffusion Models (ICML 2025 · similarity 0.85)
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · 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.
