Dataset Distillation

Dataset Distillation aims to compress a large dataset into a small synthetic one while maintaining predictive performance. We show that as different demographic groups exhibit distinct predictive patterns, the distillation process struggles to simultaneously preserve informative signals for all subgroups, regardless of whether group sizes are mildly or severely imbalanced. Consequently, models trained on distilled data can experience substantial performance drops for certain subgroups, leading to fairness gaps. Crucially, these gaps do not disappear by merely correcting group imbalance, since they stem from fundamental mismatches in subgroup predictive patterns rather than from sample-size disparities alone. We therefore formally analyze the interaction between these two sources of bias and cast the solution as identifying a group-imbalance-agnostic barycenter of the predictive information that induces similar representations across all subgroups. By distilling toward this shared aggregate representation, we show that group fairness concerns can be reduced. Our approach is compatible with existing distillation methods, and empirical results show that it substantially reduces bias introduced by dataset distillation. Code is available at https://github.com/mhmoslemi/COBRA.

Dataset condensation constructs compact synthetic datasets that retain the training utility of large real-world datasets, enabling efficient model development and potentially supporting downstream research in governed domains such as healthcare. Trajectory matching (TM) is a widely used condensation approach that supervises synthetic data using changes in model parameters observed during training on real data, yet the structure of this supervision signal remains poorly understood. In this paper, we provide a geometric characterisation of trajectory matching, showing that a fixed synthetic dataset can only reproduce a limited span of such training-induced parameter changes. When the resulting supervision signal is spectrally broad, this creates a conditional representability bottleneck. Motivated by this mismatch, we propose Bezier Trajectory Matching (BTM), which replaces SGD trajectories with quadratic Bezier trajectory surrogates between initial and final model states. These surrogates are optimised to reduce average loss along the path while replacing broad SGD-derived supervision with a more structured, lower-rank signal that is better aligned with the optimisation constraints of a fixed synthetic dataset, and they substantially reduce trajectory storage. Experiments on five clinical datasets demonstrate that BTM consistently matches or improves upon standard trajectory matching, with the largest gains in low-prevalence and low-synthetic-budget settings. These results indicate that effective trajectory matching depends on structuring the supervision signal rather than reproducing stochastic optimisation paths.

Despite the perceived success of large-scale dataset distillation (DD) methods, recent evidence finds that simple random image baselines perform on-par with state-of-theart DD methods like SRe2L due to the use of soft labels during downstream model training. This is in contrast with the findings in coreset literature, where high-quality coresets consistently outperform random subsets in the hardlabel (HL) setting. To understand this discrepancy, we perform a detailed scalability analysis to examine the role of data quality under different label regimes, ranging from abundant soft labels (termed as SL+KD regime) to fixed soft labels (SL) and hard labels (HL). Our analysis reveals that high-quality coresets fail to convincingly outperform the random baseline in both SL and SL+KD regimes. In the SL+KD setting, performance further approaches nearoptimal levels relative to the full dataset, regardless of subset size or quality, for a given compute budget. This performance saturation calls into question the widespread practice of using soft labels for model evaluation, where unlike the HL setting, subset quality has negligible influence. A subsequent systematic evaluation of five large-scale and four small-scale DD methods in the HL setting reveals that only RDED reliably outperforms random baselines on ImageNet-1K, but can still lag behind strong coreset methods due to its over-reliance on easy sample patches. Based on this, we introduce CAD-Prune, a compute-aware pruning metric that efficiently identifies samples of optimal difficulty for a given compute budget, and use it to develop CA2D, a compute-aligned DD method, outperforming current DD methods on ImageNet-1K at various IPC settings. Together, our findings uncover many insights into current DD research and establish useful tools to advance dataefficient learning for both coresets and DD.

Large-scale dataset distillation requires storing auxiliary soft labels that can be 30-40x larger on ImageNet-1K and 200x larger on ImageNet-21K than the condensed images, undermining the goal of dataset compression. We identify two fundamental issues necessitating such extensive labels: (1) insufficient image diversity, where high within-class similarity in synthetic images requires extensive augmentation, and (2) insufficient supervision diversity, where limited variety in supervisory signals during training leads to performance degradation at high compression rates. To address these challenges, we propose Label Pruning and Quantization for Large-scale Distillation (LPQLD). We enhance image diversity via class-wise batching and batch-normalization supervision during synthesis. For supervision diversity, we introduce Label Pruning with Dynamic Knowledge Reuse to improve label-per-augmentation diversity, and Label Quantization with Calibrated Student-Teacher Alignment to improve augmentation-per-image diversity. Our approach reduces soft label storage by 78x on ImageNet-1K and 500x on ImageNet-21K while improving accuracy by up to 7.2% and 2.8%, respectively. Extensive experiments validate the superiority of LPQLD across different network architectures and dataset distillation methods. Code is available at https://github.com/he-y/soft-label-pruning-quantization-for-dataset-distillation.

Dataset distillation (DD) compresses a large training set into a small synthetic set, reducing storage and training cost, and has shown strong results on general benchmarks. Decoupled DD further improves efficiency by splitting the pipeline into pretraining, sample distillation, and soft-label generation. However, existing decoupled methods largely rely on coarse class-label supervision and optimize samples within each class in a nearly identical manner. On fine-grained datasets, this often yields distilled samples that (i) retain large intra-class variation with subtle inter-class differences and (ii) become overly similar within the same class, limiting localized discriminative cues and hurting recognition. To solve the above-mentioned problems, we propose FD$^{2}$, a dedicated framework for Fine-grained Dataset Distillation. FD$^{2}$ localizes discriminative regions and constructs fine-grained representations for distillation. During pretraining, counterfactual attention learning aggregates discriminative representations to update class prototypes. During distillation, a fine-grained characteristic constraint aligns each sample with its class prototype while repelling others, and a similarity constraint diversifies attention across same-class samples. Experiments on multiple fine-grained and general datasets show that FD$^{2}$ integrates seamlessly with decoupled DD and improves performance in most settings, indicating strong transferability. Code is available at https://github.com/Guang000/FD2.

Dataset distillation, a training-aware data compression technique, has recently attracted increasing attention as an effective tool for mitigating costs of optimization and data storage. However, progress remains largely empirical. Mechanisms underlying the extraction of task-relevant information from the training process and the efficient encoding of such information into synthetic data points remain elusive. In this paper, we theoretically analyze practical algorithms of dataset distillation applied to the gradient-based training of two-layer neural networks with width $L$. By focusing on a non-linear task structure called multi-index model, we prove that the low-dimensional structure of the problem is efficiently encoded into the resulting distilled data. This dataset reproduces a model with high generalization ability for a required memory complexity of \tildeΘ$$(r^2d+L), where $d$ and $r$ are the input and intrinsic dimensions of the task. To the best of our knowledge, this is one of the first theoretical works that include a specific task structure, leverage its intrinsic dimensionality to quantify the compression rate and study dataset distillation implemented solely via gradient-based algorithms.

Dataset distillation seeks to synthesize a compact surrogate dataset that enables performance comparable to training on the original dataset for downstream tasks. For the scenario where pre-trained self-supervised models serve as priors, traditional Linear Gradient Matching optimizes synthetic images by encouraging them to mimic the gradient updates induced by real images on the linear probe. However, this batch-level formulation requires loading thousands of real images and applying multiple differentiable augmentations to synthetic images at each distillation step, leading to substantial computational and memory overheads. In this paper, we revisit the linear gradient and theoretically derive that it is essentially a local relative distribution directed from target class centers toward non-target class centers, which we term flow. This property causes suboptimality and instability, often necessitating expensive multiple augmentations to compensate. To address this, we introduce Statistical Flow Matching, an optimal, stable, and efficient supervised learning framework that optimizes synthetic images by aligning global statistical flows in the original data. Our approach loads raw statistics only once and performs a single augmentation pass on the synthetic data, achieving performance comparable to or better than the state-of-the-art method with 10x less GPU memory usage and 4x faster distillation time. Moreover, increasing the number of augmentations for our method yields further performance gains while incurring lower additional cost. Our code is publicly available at https://github.com/einsteinxia/SFM.

Recent years have witnessed the remarkable success of deep learning in remote sensing image interpretation, driven by the availability of large-scale benchmark datasets. However, this reliance on massive training data also brings substantial storage and computational costs. To address this challenge, this study introduces the concept of dataset distillation into the field of remote sensing image interpretation for the first time. Specifically, we propose discriminative prototype-guided diffusion (DPD), a diffusion-based generative distillation framework that condenses a large-scale remote sensing dataset into a compact and representative distilled dataset. To improve the semantic fidelity and diversity of the synthesized samples, we extract representative prototypes for each category in the latent space. We then construct hyperspherical semantic anchors around the prototypes to guide the reverse denoising trajectory. Furthermore, to enhance the discriminative quality of the generated samples, multiple candidates are generated for each prototype and ranked by a latent classifier using a logit-margin criterion, with the most discriminative candidates selected to form the final distilled dataset. Experiments on three high-resolution remote sensing scene classification benchmarks show that the proposed method can distill realistic, diverse, and discriminative samples for downstream model training. Code and pre-trained models are available online (https://github.com/YonghaoXu/DPD).

Given a dataset of finitely many elements $\mathcal{T} = \{\mathbf{x}_i\}_{i = 1}^N$, the goal of dataset condensation (DC) is to construct a synthetic dataset $\mathcal{S} = \{\tilde{\mathbf{x}}_j\}_{j = 1}^M$ which is significantly smaller ($M \ll N$) such that a model trained from scratch on $\mathcal{S}$ achieves comparable or even superior generalization performance to a model trained on $\mathcal{T}$. Recent advances in DC reveal a close connection to the problem of approximating the data distribution represented by $\mathcal{T}$ with a reduced set of points. In this work, we present a unified framework that encompasses existing DC methods and extend the task-specific notion of DC to a more general and formal definition using notions of discrepancy, which quantify the distance between probability distribution in different regimes. Our framework broadens the objective of DC beyond generalization, accommodating additional objectives such as robustness, privacy, and other desirable properties.

Training large neural networks on large-scale datasets requires substantial computational resources, particularly for dense prediction tasks such as object detection. Although dataset distillation (DD) has been proposed to alleviate these demands by synthesizing compact datasets from larger ones, most existing work focuses solely on image classification, leaving the more complex detection setting largely unexplored. In this paper, we introduce OD3, a novel optimization-free data distillation framework specifically designed for object detection. Our approach involves two stages: first, a candidate selection process in which object instances are iteratively placed in synthesized images based on their suitable locations, and second, a candidate screening process using a pre-trained observer model to remove low-confidence objects. We perform our data synthesis framework on MS COCO and PASCAL VOC, two popular detection datasets, with compression ratios ranging from 0.25% to 5%. Compared to the prior solely existing dataset distillation method on detection and conventional core set selection methods, OD3 delivers superior accuracy, establishes new state-of-the-art results, surpassing prior best method by more than 14% on COCO mAP50 at a compression ratio of 1.0%. Code is available at: https://github.com/VILA-Lab/OD3.

Dataset distillation provides an effective approach to reduce memory and computational costs by optimizing a compact dataset that achieves performance comparable to the full original. However, for large-scale datasets and complex deep networks (e.g., ImageNet-1K with ResNet-101), the vast optimization space hinders distillation effectiveness, limiting practical applications. Recent methods leverage pre-trained diffusion models to directly generate informative images, thereby bypassing pixel-level optimization and achieving promising results. Nonetheless, these approaches often suffer from distribution shifts between the pre-trained diffusion prior and target datasets, as well as the need for multiple distillation steps under varying settings. To overcome these challenges, we propose a novel framework that is orthogonal to existing diffusion-based distillation techniques by utilizing the diffusion prior for patch selection rather than generation. Our method predicts noise from the diffusion model conditioned on input images and optional text prompts (with or without label information), and computes the associated loss for each image-patch pair. Based on the loss differences, we identify distinctive regions within the original images. Furthermore, we apply intra-class clustering and ranking on the selected patches to enforce diversity constraints. This streamlined pipeline enables a one-step distillation process. Extensive experiments demonstrate that our approach consistently outperforms state-of-the-art methods across various metrics and settings.