Saved in:
Bibliographic Details
Main Authors: Ran, Fengli, Pu, Xiao, Liu, Bo, Bi, Xiuli, Xiao, Bin
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.02469
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911297878097920
author Ran, Fengli
Pu, Xiao
Liu, Bo
Bi, Xiuli
Xiao, Bin
author_facet Ran, Fengli
Pu, Xiao
Liu, Bo
Bi, Xiuli
Xiao, Bin
contents Dataset distillation compresses large datasets into compact synthetic ones to reduce storage and computational costs. Among various approaches, distribution matching (DM)-based methods have attracted attention for their high efficiency. However, they often overlook the evolution of feature representations during training, which limits the expressiveness of synthetic data and weakens downstream performance. To address this issue, we propose Trajectory Guided Dataset Distillation (TGDD), which reformulates distribution matching as a dynamic alignment process along the model's training trajectory. At each training stage, TGDD captures evolving semantics by aligning the feature distribution between the synthetic and original dataset. Meanwhile, it introduces a distribution constraint regularization to reduce class overlap. This design helps synthetic data preserve both semantic diversity and representativeness, improving performance in downstream tasks. Without additional optimization overhead, TGDD achieves a favorable balance between performance and efficiency. Experiments on ten datasets demonstrate that TGDD achieves state-of-the-art performance, notably a 5.0% accuracy gain on high-resolution benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2512_02469
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle TGDD: Trajectory Guided Dataset Distillation with Balanced Distribution
Ran, Fengli
Pu, Xiao
Liu, Bo
Bi, Xiuli
Xiao, Bin
Computer Vision and Pattern Recognition
Dataset distillation compresses large datasets into compact synthetic ones to reduce storage and computational costs. Among various approaches, distribution matching (DM)-based methods have attracted attention for their high efficiency. However, they often overlook the evolution of feature representations during training, which limits the expressiveness of synthetic data and weakens downstream performance. To address this issue, we propose Trajectory Guided Dataset Distillation (TGDD), which reformulates distribution matching as a dynamic alignment process along the model's training trajectory. At each training stage, TGDD captures evolving semantics by aligning the feature distribution between the synthetic and original dataset. Meanwhile, it introduces a distribution constraint regularization to reduce class overlap. This design helps synthetic data preserve both semantic diversity and representativeness, improving performance in downstream tasks. Without additional optimization overhead, TGDD achieves a favorable balance between performance and efficiency. Experiments on ten datasets demonstrate that TGDD achieves state-of-the-art performance, notably a 5.0% accuracy gain on high-resolution benchmarks.
title TGDD: Trajectory Guided Dataset Distillation with Balanced Distribution
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.02469