PRISM: Diversifying Dataset Distillation by Decoupling Architectural Priors

TL;DR

PRISM introduces a novel dataset distillation framework that leverages diverse teacher models to enhance intra-class diversity and generalization in synthetic data, outperforming existing methods on ImageNet-1K.

Contribution

PRISM decouples architectural priors during dataset synthesis by supervising logits and BN alignment with different teacher models, improving diversity and performance.

Findings

Outperforms single-teacher and multi-teacher methods on ImageNet-1K.

Generated data exhibits significantly richer intra-class diversity.

Scalable cross-class batch formation enables fast parallel synthesis.

Abstract

Dataset distillation (DD) promises compact yet faithful synthetic data, but existing approaches often inherit the inductive bias of a single teacher model. As dataset size increases, this bias drives generation toward overly smooth, homogeneous samples, reducing intra-class diversity and limiting generalization. We present PRISM (PRIors from diverse Source Models), a framework that disentangles architectural priors during synthesis. PRISM decouples the logit-matching and regularization objectives, supervising them with different teacher architectures: a primary model for logits and a stochastic subset for batch-normalization (BN) alignment. On ImageNet-1K, PRISM consistently and reproducibly outperforms single-teacher methods (e.g., SRe2L) and recent multi-teacher variants (e.g., G-VBSM) at low- and mid-IPC regimes. The generated data also show significantly richer intra-class diversity, as reflected by a notable drop in cosine similarity between features. We further analyze teacher selection strategies (pre- vs. intra-distillation) and introduce a scalable cross-class batch formation scheme for fast parallel synthesis. Code will be released after the review period.