Repeated self-distillation improves model performance significantly.
problem How much gain is possible by applying multiple steps of self-distillation?
method Investigated linear regression tasks, applied multiple steps of self-distillation, analyzed excess risk reduction.
result Multi-step self-distillation reduces excess risk by a factor as large as d, where d is the input dimension. Distillation speeds up classifier training and provides insights into its success.
problem Empirical success of knowledge distillation without theoretical explanation.
method Study of linear and deep linear classifiers, proving a generalization bound.
result Three key factors for distillation success: data geometry, optimization bias, strong monotonicity.
Improved few-shot learning with interpretable models.
problem Few-shot learning with limited data.
method Linear Distillation Learning using linear functions for each class.
result Better performance compared to other interpretable models.
Paper analyzes dataset distillation for efficient encoding of task-relevant information.
problem Efficiently encoding task-relevant information from gradient-based learning of non-linear tasks.
method Theoretical analysis of dataset distillation applied to two-layer neural networks with gradient-based training.
result Low-dimensional structure of the problem is efficiently encoded into distilled data, reproducing a model with high generalization ability.
SARD improves deep learning clinical prediction performance.
problem Deep learning models struggle to match linear models in healthcare predictions.
method Reverse Distillation to initialize deep models, combined with contextual and temporal embeddings.
result SARD outperforms state-of-the-art methods on clinical prediction outcomes.
Logit distance bounds representational similarity of models.
problem Approximating linear similarity when distributions are close.
method Defined a logit distance and proved its relationship to representational dissimilarity.
result Logit distance bounds representational similarity, providing nontrivial control in practice.
The paper analyzes knowledge distillation in wide neural networks, providing theoretical insights and practical implications.
problem Lack of theoretical understanding of knowledge distillation in wide neural networks.
method Theoretical analysis of knowledge distillation in a linearized model of a wide neural network, introducing a metric of task training difficulty.
result For a perfect teacher, a high ratio of teacher's soft labels can be beneficial. For imperfect teacher, hard labels can correct wrong predictions.
Optimal self-distillation improves generative models' velocity risk and mode recovery.
problem Improving generative models' velocity risk and mode recovery.
method Proved optimal self-distillation for rectified flow via linear probing, derived mixing coefficient, and provided validation tuning.
result Optimal self-distillation improves velocity risk and mode recovery.
Improved self-distillation reduces label noise and enhances model accuracy.
problem Label noise in multi-class classification.
method Label averaging and refined partial labels.
result Single-round self-distillation achieves comparable performance to multi-round distillation.
A method to automatically choose feature dimensions in linear attention for better approximation quality.
problem Choosing the feature dimension in linear attention to balance quality and efficiency.
method Statistical degrees of freedom for determining feature dimension, layer-wise training strategy.
result Our method achieves smaller approximation error compared to fixed dimensions and improves model performance.
SFAVEL distills features from language models for fact verification without annotations.
problem Fact verification with semantically meaningful and compact features.
method Self-supervised pretraining using contrastive loss with language models.
result Achieved state-of-the-art results on FB15k-237 and FEVER.
Recent variants improve knowledge distillation performance.
problem Improving the performance of knowledge distillation.
method Introducing additional components or changing the learning process.
result These variants have shown promising results.
Anti-Distillation improves reproducibility of deep networks by making ensemble predictions more diverse.
problem Deep networks are prone to high prediction differences, making them less reproducible.
method Anti-Distillation uses ensembles to force predictions to be more different and diverse.
result Anti-Distillation reduces prediction differences by making ensemble predictions more diverse.
A new method for distilling predictions from a teacher model to a student model without original training data.
problem Training without original labeled data.
method Prediction-only distillation scheme for linear and logistic regression.
result Prediction mixing can outperform both the teacher and pure-distilled models.
Develops efficient algorithm for distilling supervised and offline RL datasets.
problem Creating synthetic datasets for supervised and offline RL models to match training data performance.
method An efficient algorithm based on matching losses without model training, using randomly sampled regressors.
result Synthetic datasets derived with ildeO(d2) sampled regressors match MSE loss of bounded linear models on training data. This study improves knowledge distillation for RNN-T models with noisy labels.
problem Challenges in distilling knowledge from RNN-T models with variable quality teachers.
method Full-sum distillation and sequence-level knowledge distillation.
result Full-sum distillation outperforms other methods for RNN-T models, especially for bad teachers.
Improved dataset distillation for images and texts boosts model accuracy.
problem Reducing dataset size for faster and more energy-efficient model training.
method Simultaneous distillation of images and soft labels, extending to text datasets.
result 2-4% increase in accuracy for image classification tasks, 20% reduction in distilled samples.
Labels distilled from images improve model training efficiency and flexibility.
problem Creating synthetic labels for a small set of real images to train models effectively.
method Introduce a more robust and flexible meta-learning algorithm for distillation and an effective first-order strategy based on convex optimization layers.
result Label distillation leads to improved results and greater flexibility in neural architectures.
New method distills discrete diffusion models, maintaining quality and diversity.
problem Difficult to distill discrete diffusion models.
method Discrete Moment Matching Distillation (D-MMD)
result Maintains high quality and diversity in distilled models.
Estimates model performance from compute budget for distillation.
problem Risk mitigation in large-scale distillation.
method Distillation scaling law based on compute budget allocation.
result Maximizes student performance with compute-optimal allocation.
Real-time policy distillation speeds up and improves reinforcement learning.
problem Slow and inefficient policy distillation in reinforcement learning.
method Simultaneous training and distillation of a teacher and student model.
result Significantly reduced distillation time and improved small model performance.
Distillation affects some classes more than others, impacting fairness and bias.
problem Distillation affects some classes more than others, impacting fairness and bias.
method Examined class-wise accuracy and fairness metrics (DPD, EOD) on models trained with different datasets.
result Increasing the distillation temperature improves the distilled student model's fairness and individual fairness.
Survey of knowledge distillation for resource-limited devices.
problem Deploying large deep learning models on resource-limited devices.
method Knowledge distillation using a smaller model trained with information from a larger model.
result A new metric (distillation metric) for comparing different knowledge distillation algorithms.
Study shows optimal self-distillation improves model performance on noisy data.
problem Improving model performance on noisy Gaussian mixture data.
method Hyperparameter-tuned multi-stage self-distillation with a linear classifier, using replica method.
result Primary driver of SD's performance improvement is denoising through hard pseudo-labels.
This paper improves robustness in small neural networks through distillation.
problem Vulnerability of small neural networks to adversarial attacks.
method Adversarially Robust Distillation (ARD) to transfer robustness from teacher to student networks.
result ARD produces small models with superior robust accuracy compared to adversarially trained networks.
This work improves knowledge distillation by transferring full kernel matrices efficiently.
problem Efficiently transferring full pairwise similarity matrices for model compression in deep learning.
method The authors propose a method to transfer the full similarity matrix effectively using the Nyström method, decomposing it into partial matrices.
result The difference between the full kernel matrices of teacher and student can be well bounded by partial matrices, improving optimization efficiency.
The paper distills a weighted automaton from RNNs for language modeling.
problem Tackles the gap between deep learning and grammatical inference.
method Uses a spectral approach to infer a weighted automaton from a trained RNN.
result Extracted weighted automata are good approximations of the RNNs, validating the approach.
Study on ensemble, distillation, and self-distillation in deep learning models.
problem Improving test accuracy in deep learning models using ensemble and distillation methods.
method Formal study of ensemble and distillation, considering multi-view data structure.
result Proven that ensemble and distillation can improve test accuracy in deep learning models, and the superior performance can be distilled into a single model.
A new distillation method transfers channel information from teacher to student.
problem Transfer knowledge from teacher to student with fewer parameters and calculations.
method Channel Distillation (CD) and Guided Knowledge Distillation (GKD) with loss decay.
result Achieved 27.68% top-1 error on ImageNet with ResNet18, outperforming state-of-the-art methods.
OKDDip uses diverse peers to improve online knowledge distillation.
problem Early saturation in group-based distillation.
method Two-level distillation with multiple auxiliary peers and a group leader, using attention-based aggregation weights.
result OKDDip consistently gives better performance than state-of-the-art approaches.
RID framework quantifies and regularizes task-relevant knowledge in distillation.
problem Distilling irrelevant information can hinder student model performance.
method Partial Information Decomposition to quantify and regularize task-relevant knowledge.
result RID framework leads to more resilient distillation under nuisance teachers.
Distiller simplifies DNN compression research with a Python package.
problem Efficiently compressing deep neural networks.
method Open-source Python package with DNN compression algorithms.
result Facilitates new research and learning tasks in DNN compression.
A new method for training GNNs without a teacher model.
problem Training over-parameterized GNN models is difficult and inefficient.
method GNN Self-Distillation (GNN-SD) with NDR and ADR.
result Improves GNN performance with less training cost and better generalization.
FasterVoiceGrad speeds up VC by 6-7x with novel distillation.
problem Slow iterative sampling in diffusion-based VC models.
method Adversarial diffusion conversion distillation (ADCD) to create a faster one-step model.
result 6.6-6.9 and 1.8x faster on GPU and CPU, respectively.
This paper improves speech recognition by distilling knowledge from acoustic models.
problem Improving speech recognition accuracy using ensemble models.
method Proposes multi-teacher distillation strategies for joint CTC-attention end-to-end ASR systems, integrating error rate metric for optimization.
result Reports state-of-the-art error rates on various datasets and languages.
A new KD method distills dataset-based knowledge using MHA.
problem Distilling knowledge from large teacher networks to small student networks.
method Graph-based knowledge distillation by multi-head attention network.
result The method improves SN performance by 7.05% on CIFAR100.
A framework preserves uncertainty in ensemble distillation.
problem Preserving uncertainty decomposition in ensemble distillation.
method General framework for distilling both regression and classification ensembles, preserving natural uncertainty decomposition.
result Framework maintains decomposition of predictive uncertainty.
Novel framework learns efficient student models from teacher networks.
problem Model capacity gap between teacher and student networks.
method Neural architecture search and oracle knowledge distillation.
result Searched student models often outperform teacher models.
CAKD framework optimizes knowledge transfer by focusing on influential components of distillation.
problem Balancing and optimizing knowledge transfer in distillation models.
method Decouple KL divergence into BCD, SCD, and WCD; prioritize influential components.
result CAKD framework consistently outperforms baseline across diverse models and datasets.
Consistency distillation reduces memorization in diffusion models without harming sample quality.
problem Understanding how distillation affects memorization in diffusion models.
method Analysis of consistency distillation in diffusion models using a random feature neural network model.
result Consistency distillation reduces memorization in diffusion models without harming sample quality.
Self-distillation improves model performance but can lead to underfitting.
problem Understanding why self-distillation improves model performance and its limitations.
method Theoretical analysis of self-distillation in Hilbert space with ℓ2 regularization. result Self-distillation modifies regularization by limiting the number of basis functions, potentially leading to underfitting.
A new method distills datasets more efficiently and effectively.
problem Achieving competitive performance on test data with a small synthetic dataset.
method Tackles dataset distillation as a bilevel optimization problem, introduces RaT-BPTT to stabilize gradients and speed up optimization.
result Establishes new state-of-the-art performance across various benchmarks.
The transfer of knowledge from one policy to another is an important tool in Deep Reinforcement Learning. This process, referred to as distillation, has been used to great success, for example, by enhancing the optimisation of agents, leading to stronger performance faster, on harder domains [26, 32, 5, 8]. Despite the…
Improved generalization with iterative self-distillation using weighted ground-truth targets.
problem Improving generalization accuracy in neural networks.
method Iterative kernel regression with weighted ground-truth targets and ℓ2 regularization. result Closed-form solution for optimal weighting parameter and efficient estimation.
The paper explains knowledge distillation by analyzing visual concepts in DNNs.
problem Understanding how knowledge distillation affects the learning of visual concepts in deep neural networks.
method The paper proposes three hypotheses and designs mathematical metrics to evaluate feature representations of DNNs.
result The hypotheses were verified through experiments on various DNNs.
This work proposes splitting deep neural networks into smaller sub-networks for faster and more efficient distillation.
problem Challenges in training deep neural networks, including local optima, gradient issues, and computational demands.
method Proposes a non-end-to-end distillation approach by splitting networks into smaller, independent sub-networks (neighbourhoods).
result Independent training of smaller sub-networks can speed up distillation and improve efficiency in various applications.
SiD distills pretrained diffusion models into a fast one-step generator.
problem Efficiently distilling pretrained diffusion models into a fast generator.
method Reformulates forward diffusion processes as semi-implicit distributions and uses three score-related identities to create a loss mechanism.
result Achieves high FID performance and significantly reduces generation time.
Compressed Federated Distillation reduces communication in federated learning.
problem Communication constraints in Federated Learning.
method Compressed Federated Distillation (CFD) leverages soft labels and quantization techniques.
result Reduces communication by more than 4 orders of magnitude compared to Federated Averaging.