With the rise of deep learning, large datasets and complex models have become common, requiring significant computing power. To address this, data distillation has emerged as a technique to quickly train models with lower memory and time requirements. However, data distillation on text-based datasets hasn't been explored much because of the challenges rising due to its discrete nature. Additionally, existing dataset distillation methods often struggle to generalize to new architectures. In the paper, we propose several data distillation techniques for multilingual text classification datasets using language-model-based learning methods. We conduct experiments to analyze their performance in terms of classification strength, and cross-architecture generalization. Furthermore, we investigate the language-specific fairness of the data summaries generated by these methods. Our approach builds upon existing techniques, enhancing cross-architecture generalization in the text data distillation domain.
Neural networks trained on distilled data often produce over-confident output and require correction by calibration methods. Existing calibration methods such as temperature scaling and mixup work well for networks trained on original large-scale data. However, we find that these methods fail to calibrate networks trained on data distilled from large source datasets. In this paper, we show that distilled data lead to networks that are not calibratable due to (i) a more concentrated distribution of the maximum logits and (ii) the loss of information that is semantically meaningful but unrelated to classification tasks. To address this problem, we propose Masked Temperature Scaling (MTS) and Masked Distillation Training (MDT) which mitigate the limitations of distilled data and achieve better calibration results while maintaining the efficiency of dataset distillation.
Efficiency and trustworthiness are two eternal pursuits when applying deep learning in real-world applications. With regard to efficiency, dataset distillation (DD) endeavors to reduce training costs by distilling the large dataset into a tiny synthetic dataset. However, existing methods merely concentrate on in-distribution (InD) classification in a closed-world setting, disregarding out-of-distribution (OOD) samples. On the other hand, OOD detection aims to enhance models' trustworthiness, which is always inefficiently achieved in full-data settings. For the first time, we simultaneously consider both issues and propose a novel paradigm called Trustworthy Dataset Distillation (TrustDD). By distilling both InD samples and outliers, the condensed datasets are capable of training models competent in both InD classification and OOD detection. To alleviate the requirement of real outlier data, we further propose to corrupt InD samples to generate pseudo-outliers, namely Pseudo-Outlier Exposure (POE). Comprehensive experiments on various settings demonstrate the effectiveness of TrustDD, and POE surpasses the state-of-the-art method Outlier Exposure (OE). Compared with the preceding DD, TrustDD is more trustworthy and applicable to open-world scenarios. Our code is available at https://github.com/mashijie1028/TrustDD
Deep learning has grown tremendously over recent years, yielding state-of-the-art results in various fields. However, training such models requires huge amounts of data, increasing the computational time and cost. To address this, dataset distillation was proposed to compress a large training dataset into a smaller synthetic one that retains its performance -- this is usually done by (1) uniformly initializing a synthetic set and (2) iteratively updating/learning this set according to a predefined loss by uniformly sampling instances from the full data. In this paper, we improve both phases of dataset distillation: (1) we present a provable, sampling-based approach for initializing the distilled set by identifying important and removing redundant points in the data, and (2) we further merge the idea of data subset selection with dataset distillation, by training the distilled set on ``important'' sampled points during the training procedure instead of randomly sampling the next batch. To do so, we define the notion of importance based on the relative contribution of instances with respect to two different loss functions, i.e., one for the initialization phase (a kernel fitting function for kernel ridge regression and $K$-means based loss function for any other distillation method), and the relative cross-entropy loss (or any other predefined loss) function for the training phase. Finally, we provide experimental results showing how our method can latch on to existing dataset distillation techniques and improve their performance.
Data-efficient learning has garnered significant attention, especially given the current trend of large multi-modal models. Recently, dataset distillation has become an effective approach by synthesizing data samples that are essential for network training. However, it remains to be explored which samples are essential for the dataset distillation process itself. In this work, we study the data efficiency and selection for the dataset distillation task. By re-formulating the dynamics of distillation, we provide insight into the inherent redundancy in the real dataset, both theoretically and empirically. We propose to use the empirical loss value as a static data pruning criterion. To further compensate for the variation of the data value in training, we find the most contributing samples based on their causal effects on the distillation. The proposed selection strategy can efficiently exploit the training dataset, outperform the previous SOTA distillation algorithms, and consistently enhance the distillation algorithms, even on much larger-scale and more heterogeneous datasets, e.g., full ImageNet-1K and Kinetics-400. We believe this paradigm will open up new avenues in the dynamics of distillation and pave the way for efficient dataset distillation. Our code is available on https://github.com/silicx/GoldFromOres-BiLP.
The expenses involved in training state-of-the-art deep hashing retrieval models have witnessed an increase due to the adoption of more sophisticated models and large-scale datasets. Dataset Distillation (DD) or Dataset Condensation(DC) focuses on generating smaller synthetic dataset that retains the original information. Nevertheless, existing DD methods face challenges in maintaining a trade-off between accuracy and efficiency. And the state-of-the-art dataset distillation methods can not expand to all deep hashing retrieval methods. In this paper, we propose an efficient condensation framework that addresses these limitations by matching the feature-embedding between synthetic set and real set. Furthermore, we enhance the diversity of features by incorporating the strategies of early-stage augmented models and multi-formation. Extensive experiments provide compelling evidence of the remarkable superiority of our approach, both in terms of performance and efficiency, compared to state-of-the-art baseline methods.
Dataset Distillation is the task of synthesizing small datasets from large ones while still retaining comparable predictive accuracy to the original uncompressed dataset. Despite significant empirical progress in recent years, there is little understanding of the theoretical limitations/guarantees of dataset distillation, specifically, what excess risk is achieved by distillation compared to the original dataset, and how large are distilled datasets? In this work, we take a theoretical view on kernel ridge regression (KRR) based methods of dataset distillation such as Kernel Inducing Points. By transforming ridge regression in random Fourier features (RFF) space, we provide the first proof of the existence of small (size) distilled datasets and their corresponding excess risk for shift-invariant kernels. We prove that a small set of instances exists in the original input space such that its solution in the RFF space coincides with the solution of the original data. We further show that a KRR solution can be generated using this distilled set of instances which gives an approximation towards the KRR solution optimized on the full input data. The size of this set is linear in the dimension of the RFF space of the input set or alternatively near linear in the number of effective degrees of freedom, which is a function of the kernel, number of datapoints, and the regularization parameter $\lambda$. The error bound of this distilled set is also a function of $\lambda$. We verify our bounds analytically and empirically.
The aim of dataset distillation is to encode the rich features of an original dataset into a tiny dataset. It is a promising approach to accelerate neural network training and related studies. Different approaches have been proposed to improve the informativeness and generalization performance of distilled images. However, no work has comprehensively analyzed this technique from a security perspective and there is a lack of systematic understanding of potential risks. In this work, we conduct extensive experiments to evaluate current state-of-the-art dataset distillation methods. We successfully use membership inference attacks to show that privacy risks still remain. Our work also demonstrates that dataset distillation can cause varying degrees of impact on model robustness and amplify model unfairness across classes when making predictions. This work offers a large-scale benchmarking framework for dataset distillation evaluation.
Dataset distillation is attracting more attention in machine learning as training sets continue to grow and the cost of training state-of-the-art models becomes increasingly high. By synthesizing datasets with high information density, dataset distillation offers a range of potential applications, including support for continual learning, neural architecture search, and privacy protection. Despite recent advances, we lack a holistic understanding of the approaches and applications. Our survey aims to bridge this gap by first proposing a taxonomy of dataset distillation, characterizing existing approaches, and then systematically reviewing the data modalities, and related applications. In addition, we summarize the challenges and discuss future directions for this field of research.
Dataset Distillation aims to distill an entire dataset's knowledge into a few synthetic images. The idea is to synthesize a small number of synthetic data points that, when given to a learning algorithm as training data, result in a model approximating one trained on the original data. Despite recent progress in the field, existing dataset distillation methods fail to generalize to new architectures and scale to high-resolution datasets. To overcome the above issues, we propose to use the learned prior from pre-trained deep generative models to synthesize the distilled data. To achieve this, we present a new optimization algorithm that distills a large number of images into a few intermediate feature vectors in the generative model's latent space. Our method augments existing techniques, significantly improving cross-architecture generalization in all settings.
Dataset distillation reduces the network training cost by synthesizing small and informative datasets from large-scale ones. Despite the success of the recent dataset distillation algorithms, three drawbacks still limit their wider application: i). the synthetic images perform poorly on large architectures; ii). they need to be re-optimized when the distillation ratio changes; iii). the limited diversity restricts the performance when the distillation ratio is large. In this paper, we propose a novel distillation scheme to \textbf{D}istill information of large train sets \textbf{i}nto generative \textbf{M}odels, named DiM. Specifically, DiM learns to use a generative model to store the information of the target dataset. During the distillation phase, we minimize the differences in logits predicted by a models pool between real and generated images. At the deployment stage, the generative model synthesizes various training samples from random noises on the fly. Due to the simple yet effective designs, the trained DiM can be directly applied to different distillation ratios and large architectures without extra cost. We validate the proposed DiM across 4 datasets and achieve state-of-the-art results on all of them. To the best of our knowledge, we are the first to achieve higher accuracy on complex architectures than simple ones, such as 75.1\% with ResNet-18 and 72.6\% with ConvNet-3 on ten images per class of CIFAR-10. Besides, DiM outperforms previous methods with 10\% $\sim$ 22\% when images per class are 1 and 10 on the SVHN dataset.
Despite Federated Learning (FL)'s trend for learning machine learning models in a distributed manner, it is susceptible to performance drops when training on heterogeneous data. In addition, FL inevitability faces the challenges of synchronization, efficiency, and privacy. Recently, dataset distillation has been explored in order to improve the efficiency and scalability of FL by creating a smaller, synthetic dataset that retains the performance of a model trained on the local private datasets. We discover that using distilled local datasets can amplify the heterogeneity issue in FL. To address this, we propose a new method, called Federated Virtual Learning on Heterogeneous Data with Local-Global Distillation (FedLGD), which trains FL using a smaller synthetic dataset (referred as virtual data) created through a combination of local and global dataset distillation. Specifically, to handle synchronization and class imbalance, we propose iterative distribution matching to allow clients to have the same amount of balanced local virtual data; to harmonize the domain shifts, we use federated gradient matching to distill global virtual data that are shared with clients without hindering data privacy to rectify heterogeneous local training via enforcing local-global feature similarity. We experiment on both benchmark and real-world datasets that contain heterogeneous data from different sources, and further scale up to an FL scenario that contains large number of clients with heterogeneous and class imbalance data. Our method outperforms state-of-the-art heterogeneous FL algorithms under various settings with a very limited amount of distilled virtual data.
Dataset distillation aims to synthesize small datasets with little information loss from original large-scale ones for reducing storage and training costs. Recent state-of-the-art methods mainly constrain the sample synthesis process by matching synthetic images and the original ones regarding gradients, embedding distributions, or training trajectories. Although there are various matching objectives, currently the strategy for selecting original images is limited to naive random sampling. We argue that random sampling overlooks the evenness of the selected sample distribution, which may result in noisy or biased matching targets. Besides, the sample diversity is also not constrained by random sampling. These factors together lead to optimization instability in the distilling process and degrade the training efficiency. Accordingly, we propose a novel matching strategy named as \textbf{D}ataset distillation by \textbf{RE}present\textbf{A}tive \textbf{M}atching (DREAM), where only representative original images are selected for matching. DREAM is able to be easily plugged into popular dataset distillation frameworks and reduce the distilling iterations by more than 8 times without performance drop. Given sufficient training time, DREAM further provides significant improvements and achieves state-of-the-art performances.
Active Learning (AL) has remained relatively unexplored for LiDAR perception tasks in autonomous driving datasets. In this study we evaluate Bayesian active learning methods applied to the task of dataset distillation or core subset selection (subset with near equivalent performance as full dataset). We also study the effect of application of data augmentation (DA) within Bayesian AL based dataset distillation. We perform these experiments on the full Semantic-KITTI dataset. We extend our study over our existing work only on 1/4th of the same dataset. Addition of DA and BALD have a negative impact over the labeling efficiency and thus the capacity to distill datasets. We demonstrate key issues in designing a functional AL framework and finally conclude with a review of challenges in real world active learning.
Autonomous driving (AD) perception today relies heavily on deep learning based architectures requiring large scale annotated datasets with their associated costs for curation and annotation. The 3D semantic data are useful for core perception tasks such as obstacle detection and ego-vehicle localization. We propose a new dataset, Navya 3D Segmentation (Navya3DSeg), with a diverse label space corresponding to a large scale production grade operational domain, including rural, urban, industrial sites and universities from 13 countries. It contains 23 labeled sequences and 25 supplementary sequences without labels, designed to explore self-supervised and semi-supervised semantic segmentation benchmarks on point clouds. We also propose a novel method for sequential dataset split generation based on iterative multi-label stratification, and demonstrated to achieve a +1.2% mIoU improvement over the original split proposed by SemanticKITTI dataset. A complete benchmark for semantic segmentation task was performed, with state of the art methods. Finally, we demonstrate an Active Learning (AL) based dataset distillation framework. We introduce a novel heuristic-free sampling method called ego-pose distance based sampling in the context of AL. A detailed presentation on the dataset is available here https://www.youtube.com/watch?v=5m6ALIs-s20.
We propose a new dataset distillation algorithm using reparameterization and convexification of implicit gradients (RCIG), that substantially improves the state-of-the-art. To this end, we first formulate dataset distillation as a bi-level optimization problem. Then, we show how implicit gradients can be effectively used to compute meta-gradient updates. We further equip the algorithm with a convexified approximation that corresponds to learning on top of a frozen finite-width neural tangent kernel. Finally, we improve bias in implicit gradients by parameterizing the neural network to enable analytical computation of final-layer parameters given the body parameters. RCIG establishes the new state-of-the-art on a diverse series of dataset distillation tasks. Notably, with one image per class, on resized ImageNet, RCIG sees on average a 108\% improvement over the previous state-of-the-art distillation algorithm. Similarly, we observed a 66\% gain over SOTA on Tiny-ImageNet and 37\% on CIFAR-100.
Modern deep learning requires large volumes of data, which could contain sensitive or private information that cannot be leaked. Recent work has shown for homogeneous neural networks a large portion of this training data could be reconstructed with only access to the trained network parameters. While the attack was shown to work empirically, there exists little formal understanding of its effective regime which datapoints are susceptible to reconstruction. In this work, we first build a stronger version of the dataset reconstruction attack and show how it can provably recover the \emph{entire training set} in the infinite width regime. We then empirically study the characteristics of this attack on two-layer networks and reveal that its success heavily depends on deviations from the frozen infinite-width Neural Tangent Kernel limit. Next, we study the nature of easily-reconstructed images. We show that both theoretically and empirically, reconstructed images tend to "outliers" in the dataset, and that these reconstruction attacks can be used for \textit{dataset distillation}, that is, we can retrain on reconstructed images and obtain high predictive accuracy.
Recent success of deep learning is largely attributed to the sheer amount of data used for training deep neural networks.Despite the unprecedented success, the massive data, unfortunately, significantly increases the burden on storage and transmission and further gives rise to a cumbersome model training process. Besides, relying on the raw data for training \emph{per se} yields concerns about privacy and copyright. To alleviate these shortcomings, dataset distillation~(DD), also known as dataset condensation (DC), was introduced and has recently attracted much research attention in the community. Given an original dataset, DD aims to derive a much smaller dataset containing synthetic samples, based on which the trained models yield performance comparable with those trained on the original dataset. In this paper, we give a comprehensive review and summary of recent advances in DD and its application. We first introduce the task formally and propose an overall algorithmic framework followed by all existing DD methods. Next, we provide a systematic taxonomy of current methodologies in this area, and discuss their theoretical interconnections. We also present current challenges in DD through extensive experiments and envision possible directions for future works.
Deep learning technology has developed unprecedentedly in the last decade and has become the primary choice in many application domains. This progress is mainly attributed to a systematic collaboration in which rapidly growing computing resources encourage advanced algorithms to deal with massive data. However, it has gradually become challenging to handle the unlimited growth of data with limited computing power. To this end, diverse approaches are proposed to improve data processing efficiency. Dataset distillation, a dataset reduction method, addresses this problem by synthesizing a small typical dataset from substantial data and has attracted much attention from the deep learning community. Existing dataset distillation methods can be taxonomized into meta-learning and data matching frameworks according to whether they explicitly mimic the performance of target data. Although dataset distillation has shown surprising performance in compressing datasets, there are still several limitations such as distilling high-resolution data or data with complex label spaces. This paper provides a holistic understanding of dataset distillation from multiple aspects, including distillation frameworks and algorithms, factorized dataset distillation, performance comparison, and applications. Finally, we discuss challenges and promising directions to further promote future studies on dataset distillation.
Dataset distillation has emerged as a prominent technique to improve data efficiency when training machine learning models. It encapsulates the knowledge from a large dataset into a smaller synthetic dataset. A model trained on this smaller distilled dataset can attain comparable performance to a model trained on the original training dataset. However, the existing dataset distillation techniques mainly aim at achieving the best trade-off between resource usage efficiency and model utility. The security risks stemming from them have not been explored. This study performs the first backdoor attack against the models trained on the data distilled by dataset distillation models in the image domain. Concretely, we inject triggers into the synthetic data during the distillation procedure rather than during the model training stage, where all previous attacks are performed. We propose two types of backdoor attacks, namely NAIVEATTACK and DOORPING. NAIVEATTACK simply adds triggers to the raw data at the initial distillation phase, while DOORPING iteratively updates the triggers during the entire distillation procedure. We conduct extensive evaluations on multiple datasets, architectures, and dataset distillation techniques. Empirical evaluation shows that NAIVEATTACK achieves decent attack success rate (ASR) scores in some cases, while DOORPING reaches higher ASR scores (close to 1.0) in all cases. Furthermore, we conduct a comprehensive ablation study to analyze the factors that may affect the attack performance. Finally, we evaluate multiple defense mechanisms against our backdoor attacks and show that our attacks can practically circumvent these defense mechanisms.