pith. sign in

arxiv: 2310.05773 · v2 · pith:GFG5GZHB · submitted 2023-10-09 · cs.CV

Towards Lossless Dataset Distillation via Difficulty-Aligned Trajectory Matching

pith:GFG5GZHBopen to challenge →

classification cs.CV
keywords datasetsyntheticdistillationmethodspatternslargerlosslesssamples
0
0 comments X
read the original abstract

The ultimate goal of Dataset Distillation is to synthesize a small synthetic dataset such that a model trained on this synthetic set will perform equally well as a model trained on the full, real dataset. Until now, no method of Dataset Distillation has reached this completely lossless goal, in part due to the fact that previous methods only remain effective when the total number of synthetic samples is extremely small. Since only so much information can be contained in such a small number of samples, it seems that to achieve truly loss dataset distillation, we must develop a distillation method that remains effective as the size of the synthetic dataset grows. In this work, we present such an algorithm and elucidate why existing methods fail to generate larger, high-quality synthetic sets. Current state-of-the-art methods rely on trajectory-matching, or optimizing the synthetic data to induce similar long-term training dynamics as the real data. We empirically find that the training stage of the trajectories we choose to match (i.e., early or late) greatly affects the effectiveness of the distilled dataset. Specifically, early trajectories (where the teacher network learns easy patterns) work well for a low-cardinality synthetic set since there are fewer examples wherein to distribute the necessary information. Conversely, late trajectories (where the teacher network learns hard patterns) provide better signals for larger synthetic sets since there are now enough samples to represent the necessary complex patterns. Based on our findings, we propose to align the difficulty of the generated patterns with the size of the synthetic dataset. In doing so, we successfully scale trajectory matching-based methods to larger synthetic datasets, achieving lossless dataset distillation for the very first time. Code and distilled datasets are available at https://gzyaftermath.github.io/DATM.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 9 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Condensing Large-Scale Datasets Directly with Minimal Information Loss

    cs.CV 2026-07 unverdicted novelty 7.0

    CIM directly aligns data distributions to condense large-scale datasets with minimal information loss, achieving new SOTA results on ImageNet-1K distillation at IPC=10.

  2. DMGD: Train-Free Dataset Distillation with Semantic-Distribution Matching in Diffusion Models

    cs.CV 2026-05 unverdicted novelty 7.0

    DMGD achieves better performance than fine-tuned SOTA methods in dataset distillation on ImageNet subsets by using semantic matching through conditional likelihood optimization and OT-based distribution matching in a ...

  3. Direct Discrepancy Replay: Distribution-Discrepancy Condensation and Manifold-Consistent Replay for Continual Face Forgery Detection

    cs.CV 2026-04 unverdicted novelty 7.0

    A replay method for continual face forgery detection condenses real-fake distribution discrepancies into compact maps and synthesizes compatible samples from current real faces to reduce forgetting under tight memory ...

  4. Robust Trajectory Distillation: Hybrid Reweighting Meets Teacher-Inspired Targets

    cs.CV 2026-06 unverdicted novelty 6.0

    Introduces SGR and TIAT for robust dataset distillation that suppresses noise while preserving knowledge under noisy supervision.

  5. Rank-Aware Hyperbolic Alignment for Vision-Language Dataset Distillation

    cs.CV 2026-06 unverdicted novelty 6.0

    RAHA applies rank-aware hyperbolic alignment to vision-language dataset distillation by enforcing geodesic alignment in the shared low-rank range and regularizing the residual subspace for improved transfer.

  6. DIVER:Diving Deeper into Distilled Data via Expressive Semantic Recovery

    cs.CV 2026-05 unverdicted novelty 6.0

    DIVER is a dual-stage distillation method using diffusion models to enhance semantic preservation and cross-architecture generalization in dataset distillation.

  7. Distill Once, Adapt Life-Long: Exploring Dataset Distillation for Continual Test-Time Adaptation

    cs.CV 2026-06 unverdicted novelty 5.0

    DO-ALL uses dataset distillation to create synthetic source anchors that enable stable long-term continual test-time adaptation without storing original source data.

  8. Distill Once, Adapt Life-Long: Exploring Dataset Distillation for Continual Test-Time Adaptation

    cs.CV 2026-06 unverdicted novelty 5.0

    DO-ALL applies dataset distillation to generate synthetic source anchors that stabilize continual test-time adaptation under evolving domains without storing original source data.

  9. DIVER:Diving Deeper into Distilled Data via Expressive Semantic Recovery

    cs.CV 2026-05 unverdicted novelty 5.0

    DIVER applies a pre-trained diffusion model in a dual-stage process of semantic inheritance, guidance, and fusion to improve semantic expression and cross-architecture generalization in dataset distillation.