# Dataset distillation for memorized data: Soft labels can leak held-out teacher knowledge

arXiv: https://arxiv.org/abs/2506.14457 · HTML: https://mlsanity.com/paper/2506.14457/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 31 references that are in the map.
- First posted on arXiv 17 Jun 2025.

## Builds on (most cited first, 22 of 22)

- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Understanding deep learning requires rethinking generalization](https://mlsanity.com/paper/1611.03530/) (2016)
- [Knowledge Distillation: A Survey](https://mlsanity.com/paper/2006.05525/) (2020)
- [When Does Label Smoothing Help?](https://mlsanity.com/paper/1906.02629/) (2019)
- [Progress measures for grokking via mechanistic interpretability](https://mlsanity.com/paper/2301.05217/) (2023)
- [A Survey on Knowledge Distillation of Large Language Models](https://mlsanity.com/paper/2402.13116/) (2024)
- [Towards Understanding Knowledge Distillation](https://mlsanity.com/paper/2105.13093/) (2021)
- [Dataset Distillation: A Comprehensive Review](https://mlsanity.com/paper/2301.07014/) (2023)
- [Generalizing Dataset Distillation via Deep Generative Prior](https://mlsanity.com/paper/2305.01649/) (2023)
- [Rethinking Soft Labels for Knowledge Distillation: A Bias-Variance Tradeoff Perspective](https://mlsanity.com/paper/2102.00650/) (2021)
- [Scaling Laws for Fact Memorization of Large Language Models](https://mlsanity.com/paper/2406.15720/) (2024)
- [A Label is Worth a Thousand Images in Dataset Distillation](https://mlsanity.com/paper/2406.10485/) (2024)
- [Quantifying the Knowledge in a DNN to Explain Knowledge Distillation for Classification](https://mlsanity.com/paper/2208.08741/) (2022)
- [The Pitfalls of Memorization: When Memorization Hurts Generalization](https://mlsanity.com/paper/2412.07684/) (2024)
- [Effectiveness of Arbitrary Transfer Sets for Data-free Knowledge Distillation](https://mlsanity.com/paper/2011.09113/) (2020)
- [Solvable Model for Inheriting the Regularization through Knowledge Distillation](https://mlsanity.com/paper/2012.00194/) (2020)
- [Towards a theory of model distillation](https://mlsanity.com/paper/2403.09053/) (2024)
- [What is Dataset Distillation Learning?](https://mlsanity.com/paper/2406.04284/) (2024)
- [Quantifying Knowledge Distillation Using Partial Information Decomposition](https://mlsanity.com/paper/2411.07483/) (2024)
- [Measures of Information Reflect Memorization Patterns](https://mlsanity.com/paper/2210.09404/) (2022)
- [A Multi-Perspective Analysis of Memorization in Large Language Models](https://mlsanity.com/paper/2405.11577/) (2024)

## Built on by (most cited first, 10 of 10)

- [Towards Understanding Subliminal Learning: When and How Hidden Biases Transfer](https://mlsanity.com/paper/2509.23886/) (2025)
- [Channel Location Constrains the Auditability of Subliminal Learning](https://mlsanity.com/paper/2606.22019/) (2026)
- [ManifoldGD: Training-Free Hierarchical Manifold Guidance for Diffusion-Based Dataset Distillation](https://mlsanity.com/paper/2602.23295/) (2026)
- [Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees](https://mlsanity.com/paper/2605.18654/) (2026)
- [Distilling Tabular Foundation Models for Structured Health Data](https://mlsanity.com/paper/2605.18702/) (2026)
- [Scaling Model-Generated Distillation Data Can Make Latent Teacher Traits More Recoverable](https://mlsanity.com/paper/2608.26958/) (2026)
- [Subliminal Learning as Trait-Direction Drift: A Mechanism and Targeted Control under SFT Distillation](https://mlsanity.com/paper/2609.01091/) (2026)
- [Decoupling Knowledge and Privacy: Post-Task Self-Distillation Replay for LLM Continual Learning](https://mlsanity.com/paper/2609.29711/) (2026)
- [Persistent Watermarking of Text-to-Image Models](https://mlsanity.com/paper/2609.39024/) (2026)
- [Cross-Trait Transfer in Subliminal Learning](https://mlsanity.com/paper/2610.04260/) (2026)

## Abstract (arXiv)

Dataset distillation aims to compress training data into fewer examples via a teacher, from which a student can learn effectively. While its success is often attributed to structure in the data, modern neural networks also memorize specific facts, but if and how such memorized information is can transferred in distillation settings remains less understood. In this work, we show that students trained on soft labels from teachers can achieve non-trivial accuracy on held-out memorized data they never directly observed. This effect persists on structured data when the teacher has not generalized.To analyze it in isolation, we consider finite random i.i.d. datasets where generalization is a priori impossible and a successful teacher fit implies pure memorization. Still, students can learn non-trivial information about the held-out data, in some cases up to perfect accuracy. In those settings, enough soft labels are available to recover the teacher functionally - the student matches the teacher's predictions on all possible inputs, including the held-out memorized data. We show that these phenomena strongly depend on the temperature with which the logits are smoothed, but persist across varying network capacities, architectures and dataset compositions.