Dataset Distillation Comparison Hub

Reference Surfaces

Top Papers

• From Fewer Samples to Fewer Bits: Reframing Dataset Distillation as Joint Optimization of Precision and Compactness(8.0)

  QuADD offers a quantization-aware dataset distillation framework optimizing data compactness and precision for efficient machine learning.

• Effective Dataset Distillation for Spatio-Temporal Forecasting with Bi-dimensional Compression(7.0)

  STemDist is a novel dataset distillation method that enhances spatio-temporal forecasting by compressing both spatial and temporal dimensions.

• EVLF: Early Vision-Language Fusion for Generative Dataset Distillation(7.0)

  Improve dataset distillation by fusing visual and textual embeddings early in the generative process, creating more visually coherent synthetic data for downstream classification tasks.

• HIERAMP: Coarse-to-Fine Autoregressive Amplification for Generative Dataset Distillation(7.0)

  HIERAMP leverages a vision autoregressive model to amplify hierarchical semantics in dataset distillation, improving validation performance by focusing on discriminative parts and structures.

• Difficulty-guided Sampling: Bridging the Target Gap between Dataset Distillation and Downstream Tasks(6.0)

  A method that enhances dataset distillation with difficulty-guided sampling to optimize deep learning model training.

• Towards Principled Dataset Distillation: A Spectral Distribution Perspective(5.0)

  Create a compact and efficient dataset distillation tool for imbalanced data that improves model training stability and performance.