Extracting a curriculum from a teacher network improves distillation efficiency.
problem Efficiently training a small network using a large teacher network's output.
method Random projection of teacher network's hidden representations to progressively train the student network.
result Extracted curriculum significantly outperforms one-shot distillation and achieves similar performance to progressive distillation.
RFAD uses random features to speed up dataset distillation.
problem Efficiently compress large datasets for reduced storage and computation.
method Random feature approximation of the Neural Network Gaussian Process kernel.
result At least 100-fold speedup over KIP with competitive accuracy.
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.
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.
FAST-DAD distills complex ensemble models into faster, more accurate individual models.
problem Deploying complex AutoML ensemble predictors on tabular data is slow, large, and opaque.
method Data augmentation strategy based on Gibbs sampling from a self-attention pseudolikelihood estimator.
result FAST-DAD distillation produces significantly better individual models than standard training.
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.
Our work is a simple extension of the paper "Exploration by Random Network Distillation". More in detail, we show how to efficiently combine Intrinsic Rewards with Experience Replay in order to achieve more efficient and robust exploration (with respect to PPO/RND) and consequently better results in terms of agent perf…
This paper connects RND, deep ensembles, and Bayesian inference, providing a unified theoretical perspective.
problem Uncertainty quantification in deep learning models.
method Analysis of Random Network Distillation (RND) within the neural tangent kernel framework.
result The uncertainty signal from RND is equivalent to the predictive variance of a deep ensemble and can be made to mirror the centered posterior predictive distribution of Bayesian inference.
SPEQ improves quantized neural networks by stochastic precision sharing and cosine similarity loss.
problem Improving quantized deep neural networks for edge devices.
method SPEQ combines stochastic precision sharing and cosine similarity loss for knowledge distillation.
result SPEQ outperforms existing methods in various tasks.
We introduce an exploration bonus for deep reinforcement learning methods that is easy to implement and adds minimal overhead to the computation performed. The bonus is the error of a neural network predicting features of the observations given by a fixed randomly initialized neural network. We also introduce a method …
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.
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.
Knowledge distillation is effective for producing small, high-performance neural networks for classification, but these small networks are vulnerable to adversarial attacks. This paper studies how adversarial robustness transfers from teacher to student during knowledge distillation. We find that a large amount of robu…
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.
Often we wish to transfer representational knowledge from one neural network to another. Examples include distilling a large network into a smaller one, transferring knowledge from one sensory modality to a second, or ensembling a collection of models into a single estimator. Knowledge distillation, the standard approa…
This paper distills Bayesian posterior expectations for deep neural networks.
problem Improving deep neural network performance and uncertainty quantification.
method Develops a framework for distilling expectations from Bayesian posterior distributions using Monte Carlo samples.
result The framework successfully distills posterior predictive distribution and expected entropy.
We present a novel framework of knowledge distillation that is capable of learning powerful and efficient student models from ensemble teacher networks. Our approach addresses the inherent model capacity issue between teacher and student and aims to maximize benefit from teacher models during distillation by reducing t…
Knowledge distillation (KD) is a technique to derive optimal performance from a small student network (SN) by distilling knowledge of a large teacher network (TN) and transferring the distilled knowledge to the small SN. Since a role of convolutional neural network (CNN) in KD is to embed a dataset so as to perform a g…
This paper presents the philosophy, design and feature-set of Neural Network Distiller, an open-source Python package for DNN compression research. Distiller is a library of DNN compression algorithms implementations, with tools, tutorials and sample applications for various learning tasks. Its target users are both en…
New methods distill data for deep networks efficiently.
problem Reduction of training data cost and inconvenience.
method Generative teaching networks, gradient matching, Implicit Function Theorem.
result New methods are computationally more efficient and improve model performance.
Policy distillation in deep reinforcement learning provides an effective way to transfer control policies from a larger network to a smaller untrained network without a significant degradation in performance. However, policy distillation is underexplored in deep reinforcement learning, and existing approaches are compu…
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.
PS-KD distills a model's own knowledge to soften hard targets during training.
problem Improving generalization of deep neural networks by softening hard targets.
method Progressive self-knowledge distillation (PS-KD) that progressively distills a model's own knowledge to soften hard targets.
result PS-KD improves accuracy and provides high quality of confidence estimates in terms of calibration and ordinal ranking.
Distillation is a method to transfer knowledge from one model to another and often achieves higher accuracy with the same capacity. In this paper, we aim to provide a theoretical understanding on what mainly helps with the distillation. Our answer is "early stopping". Assuming that the teacher network is overparameteri…
While noise is commonly considered a nuisance in computing systems, a number of studies in neuroscience have shown several benefits of noise in the nervous system from enabling the brain to carry out computations such as probabilistic inference as well as carrying additional information about the stimuli. Similarly, no…
Convolutional neural networks have been widely deployed in various application scenarios. In order to extend the applications' boundaries to some accuracy-crucial domains, researchers have been investigating approaches to boost accuracy through either deeper or wider network structures, which brings with them the expon…
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.
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.
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.
ESPD improves learning efficiency in sparse reward reinforcement learning.
problem Sparse reward reinforcement learning challenges.
method Evolutionary Stochastic Policy Distillation (ESPD) based on drifted random walk insight.
result High learning efficiency demonstrated in MuJoCo robotics control suite experiments.
New method regularizes deep networks by distilling self-knowledge.
problem Overfitting in deep neural networks.
method Self-knowledge distillation to regularize class-wise predictions.
result Significant improvement in generalization and calibration.
UVU simplifies value uncertainty quantification in RL.
problem Estimating epistemic uncertainty in value functions for reinforcement learning.
method UVU uses squared prediction errors between an online learner and a fixed, randomly initialized target network, incorporating policy-conditional value uncertainty.
result UVU achieves equal performance to large ensembles on challenging offline RL settings, with computational savings.
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.
Bayesian neural networks (BNNs) allow us to reason about uncertainty in a principled way. Stochastic Gradient Langevin Dynamics (SGLD) enables efficient BNN learning by drawing samples from the BNN posterior using mini-batches. However, SGLD and its extensions require storage of many copies of the model parameters, a p…
Ensembles of neural networks have been shown to give better performance than single networks, both in terms of predictions and uncertainty estimation. Additionally, ensembles allow the uncertainty to be decomposed into aleatoric (data) and epistemic (model) components, giving a more complete picture of the predictive u…
Few-step distillation improves T2I models without real images or CFG trade-offs.
problem Challenges in accelerating T2I models with high-resolution and CFG.
method Score identity distillation (SiD) for few-step generation, with adversarial loss and new guidance strategies.
result State-of-the-art performance on SDXL at 1024x1024 resolution, robust to real images absence.
Distilled models often fail to match teacher models, despite improving generalization.
problem The discrepancy between teacher and student predictive distributions remains large.
method Investigated the optimization difficulties and dataset details affecting student performance.
result Optimizing for matching the teacher does not always lead to better generalization.
This work improves certifiably robust models by distilling knowledge from adversarially robust teachers.
problem Certifiably robust models suffer from poor standard performance.
method Knowledge distillation from adversarially robust teachers to improve standard performance.
result Distillation from adversarially robust teachers consistently improves certified training performance.
Many engineers wish to deploy modern neural networks in memory-limited settings; but the development of flexible methods for reducing memory use is in its infancy, and there is little knowledge of the resulting cost-benefit. We propose structural model distillation for memory reduction using a strategy that produces a …
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.
KD2M unifies feature matching in neural networks.
problem Matching neural network activations for knowledge transfer.
method Formalizes distribution matching strategy.
result New theoretical results for KD.
This paper distills financial indicators into neural networks to reduce noise and improve accuracy.
problem Reduction of non-stationary noise in financial time series data.
method Co-distillation of smaller networks trained on indicators to transfer prior knowledge and reduce overfitting.
result The proposed method outperforms traditional methods in terms of speed and accuracy on real financial datasets.
Generative framework improves causal estimation from observational data.
problem Estimating individualized treatment effects from non-randomized data.
method Importance-Weighted Diffusion Distillation (IWDD) combining diffusion models and IPW.
result IWDD achieves state-of-the-art prediction performance and significantly improves causal estimation.
Sobolev training helps neural nets fit function values and derivatives.
problem Training neural nets to match function values and derivatives accurately.
method Using Sobolev loss with gradient flow for overparameterized networks.
result Gradient flow from random initialization can fit any function and its derivatives.
Model compression has been widely adopted to obtain light-weighted deep neural networks. Most prevalent methods, however, require fine-tuning with sufficient training data to ensure accuracy, which could be challenged by privacy and security issues. As a compromise between privacy and performance, in this paper we inve…
PLD distills knowledge using choice-theoretic Plackett-Luce model.
problem Model compression and knowledge transfer between large and small networks.
method PLD uses a weighted list-wise ranking loss based on the Plackett-Luce model.
result PLD achieves consistent gains across diverse architectures and distillation methods.
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.
FairDTD improves fairness in GNNs by distilling dual teacher knowledge, balancing utility and bias.
problem Bias in GNN predictions due to sensitive attributes.
method Dual-Teacher Distillation with a causal graph model, feature and structure teachers, and graph-level distillation.
result Achieves optimal fairness while preserving high model utility.