Knowledge Distillation is becoming one of the primary trends among neural network compression algorithms to improve the generalization performance of a smaller student model with guidance from a larger teacher model. This momentous rise in applications of knowledge distillation is accompanied by the introduction of numerous algorithms for distilling the knowledge such as soft targets and hint layers. Despite this advancement in different techniques for distilling the knowledge, the aggregation of different paths for distillation has not been studied comprehensively. This is of particular significance, not only because different paths have different importance, but also due to the fact that some paths might have negative effects on the generalization performance of the student model. Hence, we need to adaptively adjust the importance of each path to maximize the impact of distillation on the student model. In this paper, we explore different approaches for aggregating these different paths and introduce our proposed adaptive approach based on multitask learning methods. We empirically demonstrate the effectiveness of the proposed approach over other baselines on the applications of knowledge distillation in classification, semantic segmentation, and object detection tasks.
Graph Neural Networks (GNNs) are popular for graph machine learning and have shown great results on wide node classification tasks. Yet, they are less popular for practical deployments in the industry owing to their scalability challenges incurred by data dependency. Namely, GNN inference depends on neighbor nodes multiple hops away from the target, and fetching them burdens latency-constrained applications. Existing inference acceleration methods like pruning and quantization can speed up GNNs by reducing Multiplication-and-ACcumulation (MAC) operations, but the improvements are limited given the data dependency is not resolved. Conversely, multi-layer perceptrons (MLPs) have no graph dependency and infer much faster than GNNs, even though they are less accurate than GNNs for node classification in general. Motivated by these complementary strengths and weaknesses, we bring GNNs and MLPs together via knowledge distillation (KD). Our work shows that the performance of MLPs can be improved by large margins with GNN KD. We call the distilled MLPs Graph-less Neural Networks (GLNNs) as they have no inference graph dependency. We show that GLNNs with competitive accuracy infer faster than GNNs by 146X-273X and faster than other acceleration methods by 14X-27X. Under a production setting involving both transductive and inductive predictions across 7 datasets, GLNN accuracies improve over stand-alone MLPs by 12.36% on average and match GNNs on 6/7 datasets. Comprehensive analysis shows when and why GLNNs can achieve competitive accuracies to GNNs and suggests GLNN as a handy choice for latency-constrained applications.
On many natural language processing tasks, large pre-trained language models (PLMs) have shown overwhelming performances compared with traditional neural network methods. Nevertheless, their huge model size and low inference speed have hindered the deployment on resource-limited devices in practice. In this paper, we target to compress PLMs with knowledge distillation, and propose a hierarchical relational knowledge distillation (HRKD) method to capture both hierarchical and domain relational information. Specifically, to enhance the model capability and transferability, we leverage the idea of meta-learning and set up domain-relational graphs to capture the relational information across different domains. And to dynamically select the most representative prototypes for each domain, we propose a hierarchical compare-aggregate mechanism to capture hierarchical relationships. Extensive experiments on public multi-domain datasets demonstrate the superior performance of our HRKD method as well as its strong few-shot learning ability. For reproducibility, we release the code at https://github.com/cheneydon/hrkd.
The common practice for training commonsense models has gone from-human-to-corpus-to-machine: humans author commonsense knowledge graphs in order to train commonsense models. In this work, we investigate an alternative, from-machine-to-corpus-to-machine: general language models author these commonsense knowledge graphs to train commonsense models. Our study leads to a new framework, Symbolic Knowledge Distillation. As with prior art in Knowledge Distillation (Hinton et al., 2015), our approach uses larger models to teach smaller models. A key difference is that we distill knowledge symbolically-as text-in addition to the neural model. We also distill only one aspect-the commonsense of a general language model teacher, allowing the student to be a different type, a commonsense model. Altogether, we show that careful prompt engineering and a separately trained critic model allow us to selectively distill high-quality causal commonsense from GPT-3, a general language model. Empirical results demonstrate that, for the first time, a human-authored commonsense knowledge graph is surpassed by our automatically distilled variant in all three criteria: quantity, quality, and diversity. In addition, it results in a neural commonsense model that surpasses the teacher model's commonsense capabilities despite its 100x smaller size. We apply this to the ATOMIC resource, and share our new symbolic knowledge graph and commonsense models.
Nowadays, most classification networks use one-hot encoding to represent categorical data because of its simplicity. However, one-hot encoding may affect the generalization ability as it neglects inter-class correlations. We observe that, even when a neural network trained with one-hot labels produces incorrect predictions, it still pays attention to the target image region and reveals which classes confuse the network. Inspired by this observation, we propose a confusion-focusing mechanism to address the class-confusion issue. Our confusion-focusing mechanism is implemented by a two-branch network architecture. Its baseline branch generates confusing classes, and its FocusNet branch, whose architecture is flexible, discriminates correct labels from these confusing classes. We also introduce a novel focus-picking loss function to improve classification accuracy by encouraging FocusNet to focus on the most confusing classes. The experimental results validate that our FocusNet is effective for image classification on common datasets, and that our focus-picking loss function can also benefit the current neural networks in improving their classification accuracy.
Neural Architecture Search (NAS) has been pivotal in finding optimal network configurations for Convolution Neural Networks (CNNs). While many methods explore NAS from a global search-space perspective, the employed optimization schemes typically require heavy computational resources. This work introduces a method that is efficient in computationally constrained environments by examining the micro-search space of channel size. In tackling channel-size optimization, we design an automated algorithm to extract the dependencies within different connected layers of the network. In addition, we introduce the idea of knowledge distillation, which enables preservation of trained weights, admist trials where the channel sizes are changing. Further, since the standard performance indicators (accuracy, loss) fail to capture the performance of individual network components (providing an overall network evaluation), we introduce a novel metric that highly correlates with test accuracy and enables analysis of individual network layers. Combining dependency extraction, metrics, and knowledge distillation, we introduce an efficient searching algorithm, with simulated annealing inspired stochasticity, and demonstrate its effectiveness in finding optimal architectures that outperform baselines by a large margin.
3D object detection often involves complicated training and testing pipelines, which require substantial domain knowledge about individual datasets. Inspired by recent non-maximum suppression-free 2D object detection models, we propose a 3D object detection architecture on point clouds. Our method models 3D object detection as message passing on a dynamic graph, generalizing the DGCNN framework to predict a set of objects. In our construction, we remove the necessity of post-processing via object confidence aggregation or non-maximum suppression. To facilitate object detection from sparse point clouds, we also propose a set-to-set distillation approach customized to 3D detection. This approach aligns the outputs of the teacher model and the student model in a permutation-invariant fashion, significantly simplifying knowledge distillation for the 3D detection task. Our method achieves state-of-the-art performance on autonomous driving benchmarks. We also provide abundant analysis of the detection model and distillation framework.
In this work, we investigate the unexplored intersection of domain generalization (DG) and data-free learning. In particular, we address the question: How can knowledge contained in models trained on different source domains be merged into a single model that generalizes well to unseen target domains, in the absence of source and target domain data? Machine learning models that can cope with domain shift are essential for real-world scenarios with often changing data distributions. Prior DG methods typically rely on using source domain data, making them unsuitable for private decentralized data. We define the novel problem of Data-Free Domain Generalization (DFDG), a practical setting where models trained on the source domains separately are available instead of the original datasets, and investigate how to effectively solve the domain generalization problem in that case. We propose DEKAN, an approach that extracts and fuses domain-specific knowledge from the available teacher models into a student model robust to domain shift. Our empirical evaluation demonstrates the effectiveness of our method which achieves first state-of-the-art results in DFDG by significantly outperforming data-free knowledge distillation and ensemble baselines.
Accurate and automated super-resolution image synthesis is highly desired since it has the great potential to circumvent the need for acquiring high-cost medical scans and a time-consuming preprocessing pipeline of neuroimaging data. However, existing deep learning frameworks are solely designed to predict high-resolution (HR) image from a low-resolution (LR) one, which limits their generalization ability to brain graphs (i.e., connectomes). A small body of works has focused on superresolving brain graphs where the goal is to predict a HR graph from a single LR graph. Although promising, existing works mainly focus on superresolving graphs belonging to the same domain (e.g., functional), overlooking the domain fracture existing between multimodal brain data distributions (e.g., morphological and structural). To this aim, we propose a novel inter-domain adaptation framework namely, Learn to SuperResolve Brain Graphs with Knowledge Distillation Network (L2S-KDnet), which adopts a teacher-student paradigm to superresolve brain graphs. Our teacher network is a graph encoder-decoder that firstly learns the LR brain graph embeddings, and secondly learns how to align the resulting latent representations to the HR ground truth data distribution using an adversarial regularization. Ultimately, it decodes the HR graphs from the aligned embeddings. Next, our student network learns the knowledge of the aligned brain graphs as well as the topological structure of the predicted HR graphs transferred from the teacher. We further leverage the decoder of the teacher to optimize the student network. L2S-KDnet presents the first TS architecture tailored for brain graph super-resolution synthesis that is based on inter-domain alignment. Our experimental results demonstrate substantial performance gains over benchmark methods.
Human behavioral monitoring during sleep is essential for various medical applications. Majority of the contactless human pose estimation algorithms are based on RGB modality, causing ineffectiveness in in-bed pose estimation due to occlusions by blankets and varying illumination conditions. Long-wavelength infrared (LWIR) modality based pose estimation algorithms overcome the aforementioned challenges; however, ground truth pose generations by a human annotator under such conditions are not feasible. A feasible solution to address this issue is to transfer the knowledge learned from images with pose labels and no occlusions, and adapt it towards real world conditions (occlusions due to blankets). In this paper, we propose a novel learning strategy comprises of two-fold data augmentation to reduce the cross-domain discrepancy and knowledge distillation to learn the distribution of unlabeled images in real world conditions. Our experiments and analysis show the effectiveness of our approach over multiple standard human pose estimation baselines.
We propose a new approach, Knowledge Distillation using Optimal Transport (KNOT), to distill the natural language semantic knowledge from multiple teacher networks to a student network. KNOT aims to train a (global) student model by learning to minimize the optimal transport cost of its assigned probability distribution over the labels to the weighted sum of probabilities predicted by the (local) teacher models, under the constraints, that the student model does not have access to teacher models' parameters or training data. To evaluate the quality of knowledge transfer, we introduce a new metric, Semantic Distance (SD), that measures semantic closeness between the predicted and ground truth label distributions. The proposed method shows improvements in the global model's SD performance over the baseline across three NLP tasks while performing on par with Entropy-based distillation on standard accuracy and F1 metrics. The implementation pertaining to this work is publicly available at: https://github.com/declare-lab/KNOT.
Knowledge distillation transfers the knowledge from a cumbersome teacher to a small student. Recent results suggest that the student-friendly teacher is more appropriate to distill since it provides more transferable knowledge. In this work, we propose the novel framework, "prune, then distill," that prunes the model first to make it more transferrable and then distill it to the student. We provide several exploratory examples where the pruned teacher teaches better than the original unpruned networks. We further show theoretically that the pruned teacher plays the role of regularizer in distillation, which reduces the generalization error. Based on this result, we propose a novel neural network compression scheme where the student network is formed based on the pruned teacher and then apply the "prune, then distill" strategy. The code is available at https://github.com/ososos888/prune-then-distill
We study multilingual AMR parsing from the perspective of knowledge distillation, where the aim is to learn and improve a multilingual AMR parser by using an existing English parser as its teacher. We constrain our exploration in a strict multilingual setting: there is but one model to parse all different languages including English. We identify that noisy input and precise output are the key to successful distillation. Together with extensive pre-training, we obtain an AMR parser whose performances surpass all previously published results on four different foreign languages, including German, Spanish, Italian, and Chinese, by large margins (up to 18.8 \textsc{Smatch} points on Chinese and on average 11.3 \textsc{Smatch} points). Our parser also achieves comparable performance on English to the latest state-of-the-art English-only parser.
Contemporary question answering (QA) systems, including transformer-based architectures, suffer from increasing computational and model complexity which render them inefficient for real-world applications with limited resources. Further, training or even fine-tuning such models requires a vast amount of labeled data which is often not available for the task at hand. In this manuscript, we conduct a comprehensive analysis of the mentioned challenges and introduce suitable countermeasures. We propose a novel knowledge distillation (KD) approach to reduce the parameter and model complexity of a pre-trained BERT system and utilize multiple active learning (AL) strategies for immense reduction in annotation efforts. In particular, we demonstrate that our model achieves the performance of a 6-layer TinyBERT and DistilBERT, whilst using only 2% of their total parameters. Finally, by the integration of our AL approaches into the BERT framework, we show that state-of-the-art results on the SQuAD dataset can be achieved when we only use 20% of the training data.
Effectively structuring deep knowledge plays a pivotal role in transfer from teacher to student, especially in semantic vision tasks. In this paper, we present a simple knowledge structure to exploit and encode information inside the detection system to facilitate detector knowledge distillation. Specifically, aiming at solving the feature imbalance problem while further excavating the missing relation inside semantic instances, we design a graph whose nodes correspond to instance proposal-level features and edges represent the relation between nodes. To further refine this graph, we design an adaptive background loss weight to reduce node noise and background samples mining to prune trivial edges. We transfer the entire graph as encoded knowledge representation from teacher to student, capturing local and global information simultaneously. We achieve new state-of-the-art results on the challenging COCO object detection task with diverse student-teacher pairs on both one- and two-stage detectors. We also experiment with instance segmentation to demonstrate robustness of our method. It is notable that distilled Faster R-CNN with ResNet18-FPN and ResNet50-FPN yields 38.68 and 41.82 Box AP respectively on the COCO benchmark, Faster R-CNN with ResNet101-FPN significantly achieves 43.38 AP, which outperforms ResNet152-FPN teacher about 0.7 AP. Code: https://github.com/dvlab-research/Dsig.
The training of generative adversarial networks (GANs) is usually vulnerable to mode collapse and vanishing gradients. The evolutionary generative adversarial network (E-GAN) attempts to alleviate these issues by optimizing the learning strategy with multiple loss functions. It uses a learning-based evolutionary framework, which develops new mutation operators specifically for general deep neural networks. However, the evaluation mechanism in the fitness function of E-GAN cannot truly reflect the adaptability of individuals to their environment, leading to an inaccurate assessment of the diversity of individuals. Moreover, the evolution step of E-GAN only contains mutation operators without considering the crossover operator jointly, isolating the superior characteristics among individuals. To address these issues, we propose an improved E-GAN framework called IE-GAN, which introduces a new fitness function and a generic crossover operator. In particular, the proposed fitness function, from an objective perspective, can model the evolutionary process of individuals more accurately. The crossover operator, which has been commonly adopted in evolutionary algorithms, can enable offspring to imitate the superior gene expression of their parents through knowledge distillation. Experiments on various datasets demonstrate the effectiveness of our proposed IE-GAN in terms of the quality of the generated samples and time efficiency.
Knowledge distillation~(KD) has been proved effective for compressing large-scale pre-trained language models. However, existing methods conduct KD statically, e.g., the student model aligns its output distribution to that of a selected teacher model on the pre-defined training dataset. In this paper, we explore whether a dynamic knowledge distillation that empowers the student to adjust the learning procedure according to its competency, regarding the student performance and learning efficiency. We explore the dynamical adjustments on three aspects: teacher model adoption, data selection, and KD objective adaptation. Experimental results show that (1) proper selection of teacher model can boost the performance of student model; (2) conducting KD with 10% informative instances achieves comparable performance while greatly accelerates the training; (3) the student performance can be boosted by adjusting the supervision contribution of different alignment objective. We find dynamic knowledge distillation is promising and provide discussions on potential future directions towards more efficient KD methods. Our code is available at https://github.com/lancopku/DynamicKD.
Despite achieving promising results in a breadth of medical image segmentation tasks, deep neural networks require large training datasets with pixel-wise annotations. Obtaining these curated datasets is a cumbersome process which limits the applicability in scenarios. Mixed supervision is an appealing alternative for mitigating this obstacle. In this work, we propose a dual-branch architecture, where the upper branch (teacher) receives strong annotations, while the bottom one (student) is driven by limited supervision and guided by the upper branch. Combined with a standard cross-entropy loss over the labeled pixels, our novel formulation integrates two important terms: (i) a Shannon entropy loss defined over the less-supervised images, which encourages confident student predictions in the bottom branch; and (ii) a KL divergence term, which transfers the knowledge (i.e., predictions) of the strongly supervised branch to the less-supervised branch and guides the entropy (student-confidence) term to avoid trivial solutions. We show that the synergy between the entropy and KL divergence yields substantial improvements in performance. We also discuss an interesting link between Shannon-entropy minimization and standard pseudo-mask generation, and argue that the former should be preferred over the latter for leveraging information from unlabeled pixels. We evaluate the effectiveness of the proposed formulation through a series of quantitative and qualitative experiments using two publicly available datasets. Results demonstrate that our method significantly outperforms other strategies for semantic segmentation within a mixed-supervision framework, as well as recent semi-supervised approaches. Our code is publicly available: https://github.com/by-liu/ConfKD.
With the increasing popularity of graph neural networks (GNNs) in several sensitive applications like healthcare and medicine, concerns have been raised over the privacy aspects of trained GNNs. More notably, GNNs are vulnerable to privacy attacks, such as membership inference attacks, even if only black-box access to the trained model is granted. We propose PrivGNN, a privacy-preserving framework for releasing GNN models in a centralized setting. Assuming an access to a public unlabeled graph, PrivGNN provides a framework to release GNN models trained explicitly on public data along with knowledge obtained from the private data in a privacy preserving manner. PrivGNN combines the knowledge-distillation framework with the two noise mechanisms, random subsampling, and noisy labeling, to ensure rigorous privacy guarantees. We theoretically analyze our approach in the Renyi differential privacy framework. Besides, we show the solid experimental performance of our method compared to several baselines adapted for graph-structured data. Our code is available at https://github.com/iyempissy/privGnn.
A computationally expensive and memory intensive neural network lies behind the recent success of language representation learning. Knowledge distillation, a major technique for deploying such a vast language model in resource-scarce environments, transfers the knowledge on individual word representations learned without restrictions. In this paper, inspired by the recent observations that language representations are relatively positioned and have more semantic knowledge as a whole, we present a new knowledge distillation objective for language representation learning that transfers the contextual knowledge via two types of relationships across representations: Word Relation and Layer Transforming Relation. Unlike other recent distillation techniques for the language models, our contextual distillation does not have any restrictions on architectural changes between teacher and student. We validate the effectiveness of our method on challenging benchmarks of language understanding tasks, not only in architectures of various sizes, but also in combination with DynaBERT, the recently proposed adaptive size pruning method.