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.
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.
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.
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.
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.
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.
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…
Dataset distillation is a method for reducing dataset sizes by learning a small number of synthetic samples containing all the information of a large dataset. This has several benefits like speeding up model training, reducing energy consumption, and reducing required storage space. Currently, each synthetic sample is …
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.
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. 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.
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.
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.
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.
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…
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.
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.
Knowledge distillation improves model accuracy by mimicking teacher model probabilities.
problem Improving model accuracy through model compression.
method Casting knowledge distillation as a semiparametric inference problem, deriving new guarantees, and developing enhancements.
result Enhancements improve student performance by mitigating teacher overfitting and underfitting.
New method distills cloud models into edge-friendly ones.
problem Cloud-to-edge model compression with limited data exchange.
method Two-step workflow of deprivatization and distillation.
result Outperforms previous state-of-the-art approaches on various benchmarks.
EM Distillation simplifies diffusion models to one-step generators.
problem Efficient sampling from complex diffusion models with minimal loss of quality.
method EM Distillation, a maximum likelihood approach based on Expectation-Maximization.
result EM Distillation outperforms existing one-step generative methods in FID scores.
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…
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.
Distillation is an effective knowledge-transfer technique that uses predicted distributions of a powerful teacher model as soft targets to train a less-parameterized student model. A pre-trained high capacity teacher, however, is not always available. Recently proposed online variants use the aggregated intermediate pr…
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.
Self-distillation optimally improves model performance in spiked covariance models.
problem Improving model performance in spiked covariance models.
method Developed spectral shrinkage estimators and analyzed self-distillation.
result Self-distillation achieves optimal performance among spectral shrinkage estimators for spiked covariance matrices.
Survey of knowledge distillation for model compression.
problem Deploying large deep models on resource-limited devices.
method Knowledge distillation from large teacher models to small student models.
result Effective model compression and acceleration.
Distillation improves simple models by approximating complex labels.
problem Why does distillation improve simple models?
method Statistical perspective on distillation, connecting to extreme multiclass retrieval.
result Distillation helps by approximating underlying class-probabilities, reducing bias and variance.
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.
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.
This paper explores explaining GBDT2NN predictions by its teacher model, improving distillation performance.
problem Explaining GBDT2NN predictions when the models have different structures.
method Empirical study on new approach to explain GBDT2NN predictions and use it as an auxiliary learning task.
result Proposed methods achieve better performance on both explanations and predictions.
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.
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.
RealUID distills matching models using real data without GANs.
problem Slow inference in matching models like diffusion and flow.
method RealUID is a universal distillation framework that incorporates real data into the distillation procedure without using GANs.
result RealUID offers a simple theoretical foundation that covers previous distillation methods for Flow Matching and Diffusion models.
Black-box risk scoring models permeate our lives, yet are typically proprietary or opaque. We propose Distill-and-Compare, a model distillation and comparison approach to audit such models. To gain insight into black-box models, we treat them as teachers, training transparent student models to mimic the risk scores ass…
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.
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.
Two approaches extend knowledge distillation to Gaussian Processes, showing relationships to existing methods.
problem Applying knowledge distillation to Gaussian Processes for regression and classification.
method Data-centric and distribution-centric approaches to extend distillation to GPR and GPC.
result Distribution-centric approach for GPC approximately corresponds to data duplication and scaling.
FED distills ensemble predictions efficiently, improving accuracy and uncertainty.
problem Efficiently distilling ensemble predictions from limited data.
method Proposes Functional Ensemble Distillation (FED) using mixup augmentation.
result Significantly boosts performance in accuracy and uncertainty estimation.
Optimal quantization improves dataset distillation for faster training.
problem Efficiently train models with synthetic data.
method Reformulate disentangled methods as optimal quantization problems.
result Better performance and generalization on ImageNet-1K and subsets.
Model distillation aims to distill the knowledge of a complex model into a simpler one. In this paper, we consider an alternative formulation called dataset distillation: we keep the model fixed and instead attempt to distill the knowledge from a large training dataset into a small one. The idea is to synthesize a smal…
Top-performing machine learning systems, such as deep neural networks, large ensembles and complex probabilistic graphical models, can be expensive to store, slow to evaluate and hard to integrate into larger systems. Ideally, we would like to replace such cumbersome models with simpler models that perform equally well…
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.
This paper investigates teacher hacking during language model distillation and proposes methods to mitigate it.
problem Teacher hacking during language model distillation, leading to suboptimal performance.
method A controlled experimental setup involving an oracle LM, teacher LM, and student LM, using fixed offline or online data generation techniques.
result Data diversity is the key factor in preventing teacher hacking during distillation.
Survey examines distillation methods for large language models.
problem Efficiently compress large language models while preserving their capabilities.
method Knowledge Distillation and Dataset Distillation techniques.
result Integrating KD and DD can produce more effective and scalable compression strategies.
Efficiently distills pretrained text-to-image models without real data, improving FID and CLIP scores.
problem Slow iterative refinement process of diffusion-based text-to-image models.
method Guided Score identity Distillation with Long and Short Classifier-Free Guidance.
result Achieves state-of-the-art FID performance with competitive CLIP score.
Enhances model compression with multi-teacher knowledge distillation.
problem Uncertainty evaluation and diverse teacher expertise in model deployment.
method Bayesian inference and teacher-informed prior with entropy-based weighting.
result Improved predictive accuracy and robust uncertainty quantification.
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.
Knowledge distillation introduced in the deep learning context is a method to transfer knowledge from one architecture to another. In particular, when the architectures are identical, this is called self-distillation. The idea is to feed in predictions of the trained model as new target values for retraining (and itera…