The more new features that are being added to smartphones, the harder it becomes for users to find them. This is because the feature names are usually short, and there are just too many to remember. In such a case, the users may want to ask contextual queries that describe the features they are looking for, but the standard term frequency-based search cannot process them. This paper presents a novel retrieval system for mobile features that accepts intuitive and contextual search queries. We trained a relevance model via contrastive learning from a pre-trained language model to perceive the contextual relevance between query embeddings and indexed mobile features. Also, to make it run efficiently on-device using minimal resources, we applied knowledge distillation to compress the model without degrading much performance. To verify the feasibility of our method, we collected test queries and conducted comparative experiments with the currently deployed search baselines. The results show that our system outperforms the others on contextual sentence queries and even on usual keyword-based queries.
We tackle the problem of novel class discovery, which aims to learn novel classes without supervision based on labeled data from known classes. A key challenge lies in transferring the knowledge in the known-class data to the learning of novel classes. Previous methods mainly focus on building a shared representation space for knowledge transfer and often ignore modeling class relations. To address this, we introduce a class relation representation for the novel classes based on the predicted class distribution of a model trained on known classes. Empirically, we find that such class relation becomes less informative during typical discovery training. To prevent such information loss, we propose a novel knowledge distillation framework, which utilizes our class-relation representation to regularize the learning of novel classes. In addition, to enable a flexible knowledge distillation scheme for each data point in novel classes, we develop a learnable weighting function for the regularization, which adaptively promotes knowledge transfer based on the semantic similarity between the novel and known classes. To validate the effectiveness and generalization of our method, we conduct extensive experiments on multiple benchmarks, including CIFAR100, Stanford Cars, CUB, and FGVC-Aircraft datasets. Our results demonstrate that the proposed method outperforms the previous state-of-the-art methods by a significant margin on almost all benchmarks. Code is available at \href{https://github.com/kleinzcy/Cr-KD-NCD}{here}.
Federated learning enables learning from decentralized data sources without compromising privacy, which makes it a crucial technique. However, it is vulnerable to model poisoning attacks, where malicious clients interfere with the training process. Previous defense mechanisms have focused on the server-side by using careful model aggregation, but this may not be effective when the data is not identically distributed or when attackers can access the information of benign clients. In this paper, we propose a new defense mechanism that focuses on the client-side, called FedDefender, to help benign clients train robust local models and avoid the adverse impact of malicious model updates from attackers, even when a server-side defense cannot identify or remove adversaries. Our method consists of two main components: (1) attack-tolerant local meta update and (2) attack-tolerant global knowledge distillation. These components are used to find noise-resilient model parameters while accurately extracting knowledge from a potentially corrupted global model. Our client-side defense strategy has a flexible structure and can work in conjunction with any existing server-side strategies. Evaluations of real-world scenarios across multiple datasets show that the proposed method enhances the robustness of federated learning against model poisoning attacks.
Federated Learning (FL) is a privacy-preserving distributed machine learning technique that enables individual clients (e.g., user participants, edge devices, or organizations) to train a model on their local data in a secure environment and then share the trained model with an aggregator to build a global model collaboratively. In this work, we propose FedDefender, a defense mechanism against targeted poisoning attacks in FL by leveraging differential testing. Our proposed method fingerprints the neuron activations of clients' models on the same input and uses differential testing to identify a potentially malicious client containing a backdoor. We evaluate FedDefender using MNIST and FashionMNIST datasets with 20 and 30 clients, and our results demonstrate that FedDefender effectively mitigates such attacks, reducing the attack success rate (ASR) to 10\% without deteriorating the global model performance.
End-to-end automatic speech translation (AST) relies on data that combines audio inputs with text translation outputs. Previous work used existing large parallel corpora of transcriptions and translations in a knowledge distillation (KD) setup to distill a neural machine translation (NMT) into an AST student model. While KD allows using larger pretrained models, the reliance of previous KD approaches on manual audio transcripts in the data pipeline restricts the applicability of this framework to AST. We present an imitation learning approach where a teacher NMT system corrects the errors of an AST student without relying on manual transcripts. We show that the NMT teacher can recover from errors in automatic transcriptions and is able to correct erroneous translations of the AST student, leading to improvements of about 4 BLEU points over the standard AST end-to-end baseline on the English-German CoVoST-2 and MuST-C datasets, respectively. Code and data are publicly available.\footnote{\url{https://github.com/HubReb/imitkd_ast/releases/tag/v1.1}}
Classifying player actions from soccer videos is a challenging problem, which has become increasingly important in sports analytics over the years. Most state-of-the-art methods employ highly complex offline networks, which makes it difficult to deploy such models in resource constrained scenarios. Here, in this paper we propose a novel end-to-end knowledge distillation based transfer learning network pre-trained on the Kinetics400 dataset and then perform extensive analysis on the learned framework by introducing a unique loss parameterization. We also introduce a new dataset named SoccerDB1 containing 448 videos and consisting of 4 diverse classes each of players playing soccer. Furthermore, we introduce an unique loss parameter that help us linearly weigh the extent to which the predictions of each network are utilized. Finally, we also perform a thorough performance study using various changed hyperparameters. We also benchmark the first classification results on the new SoccerDB1 dataset obtaining 67.20% validation accuracy. Apart from outperforming prior arts significantly, our model also generalizes to new datasets easily. The dataset has been made publicly available at: https://bit.ly/soccerdb1
Distilling knowledge from convolutional neural networks (CNNs) is a double-edged sword for vision transformers (ViTs). It boosts the performance since the image-friendly local-inductive bias of CNN helps ViT learn faster and better, but leading to two problems: (1) Network designs of CNN and ViT are completely different, which leads to different semantic levels of intermediate features, making spatial-wise knowledge transfer methods (e.g., feature mimicking) inefficient. (2) Distilling knowledge from CNN limits the network convergence in the later training period since ViT's capability of integrating global information is suppressed by CNN's local-inductive-bias supervision. To this end, we present Cumulative Spatial Knowledge Distillation (CSKD). CSKD distills spatial-wise knowledge to all patch tokens of ViT from the corresponding spatial responses of CNN, without introducing intermediate features. Furthermore, CSKD exploits a Cumulative Knowledge Fusion (CKF) module, which introduces the global response of CNN and increasingly emphasizes its importance during the training. Applying CKF leverages CNN's local inductive bias in the early training period and gives full play to ViT's global capability in the later one. Extensive experiments and analysis on ImageNet-1k and downstream datasets demonstrate the superiority of our CSKD. Code will be publicly available.
Knowledge distillation transfers knowledge from a large model to a small one via task and distillation losses. In this paper, we observe a trade-off between task and distillation losses, i.e., introducing distillation loss limits the convergence of task loss. We believe that the trade-off results from the insufficient optimization of distillation loss. The reason is: The teacher has a lower task loss than the student, and a lower distillation loss drives the student more similar to the teacher, then a better-converged task loss could be obtained. To break the trade-off, we propose the Distillation-Oriented Trainer (DOT). DOT separately considers gradients of task and distillation losses, then applies a larger momentum to distillation loss to accelerate its optimization. We empirically prove that DOT breaks the trade-off, i.e., both losses are sufficiently optimized. Extensive experiments validate the superiority of DOT. Notably, DOT achieves a +2.59% accuracy improvement on ImageNet-1k for the ResNet50-MobileNetV1 pair. Conclusively, DOT greatly benefits the student's optimization properties in terms of loss convergence and model generalization. Code will be made publicly available.
Recent years have seen a phenomenal rise in performance and applications of transformer neural networks. The family of transformer networks, including Bidirectional Encoder Representations from Transformer (BERT), Generative Pretrained Transformer (GPT) and Vision Transformer (ViT), have shown their effectiveness across Natural Language Processing (NLP) and Computer Vision (CV) domains. Transformer-based networks such as ChatGPT have impacted the lives of common men. However, the quest for high predictive performance has led to an exponential increase in transformers' memory and compute footprint. Researchers have proposed techniques to optimize transformer inference at all levels of abstraction. This paper presents a comprehensive survey of techniques for optimizing the inference phase of transformer networks. We survey techniques such as knowledge distillation, pruning, quantization, neural architecture search and lightweight network design at the algorithmic level. We further review hardware-level optimization techniques and the design of novel hardware accelerators for transformers. We summarize the quantitative results on the number of parameters/FLOPs and accuracy of several models/techniques to showcase the tradeoff exercised by them. We also outline future directions in this rapidly evolving field of research. We believe that this survey will educate both novice and seasoned researchers and also spark a plethora of research efforts in this field.
Reasoning in mathematical domains remains a significant challenge for relatively small language models (LMs). Many current methods focus on specializing LMs in mathematical reasoning and rely heavily on knowledge distillation from powerful but inefficient large LMs (LLMs). In this work, we explore a new direction that avoids over-reliance on LLM teachers, introducing a multi-view fine-tuning method that efficiently exploits existing mathematical problem datasets with diverse annotation styles. Our approach uniquely considers the various annotation formats as different "views" and leverages them in training the model. By postpending distinct instructions to input questions, models can learn to generate solutions in diverse formats in a flexible manner. Experimental results show that our strategy enables a LLaMA-7B model to outperform prior approaches that utilize knowledge distillation, as well as carefully established baselines. Additionally, the proposed method grants the models promising generalization ability across various views and datasets, and the capability to learn from inaccurate or incomplete noisy data. We hope our multi-view training paradigm could inspire future studies in other machine reasoning domains.
Large language models (LLMs) have demonstrated their ability to learn in-context, allowing them to perform various tasks based on a few input-output examples. However, the effectiveness of in-context learning is heavily reliant on the quality of the selected examples. In this paper, we propose a novel framework to iteratively train dense retrievers that can identify high-quality in-context examples for LLMs. Our framework initially trains a reward model based on LLM feedback to evaluate the quality of candidate examples, followed by knowledge distillation to train a bi-encoder based dense retriever. Our experiments on a suite of $30$ tasks demonstrate that our framework significantly enhances in-context learning performance. Furthermore, we show the generalization ability of our framework to unseen tasks during training. An in-depth analysis reveals that our model improves performance by retrieving examples with similar patterns, and the gains are consistent across LLMs of varying sizes. The code and data are available at https://github.com/microsoft/LMOps/tree/main/llm_retriever .
In this work, we introduce a self-supervised feature representation learning framework DreamTeacher that utilizes generative networks for pre-training downstream image backbones. We propose to distill knowledge from a trained generative model into standard image backbones that have been well engineered for specific perception tasks. We investigate two types of knowledge distillation: 1) distilling learned generative features onto target image backbones as an alternative to pretraining these backbones on large labeled datasets such as ImageNet, and 2) distilling labels obtained from generative networks with task heads onto logits of target backbones. We perform extensive analyses on multiple generative models, dense prediction benchmarks, and several pre-training regimes. We empirically find that our DreamTeacher significantly outperforms existing self-supervised representation learning approaches across the board. Unsupervised ImageNet pre-training with DreamTeacher leads to significant improvements over ImageNet classification pre-training on downstream datasets, showcasing generative models, and diffusion generative models specifically, as a promising approach to representation learning on large, diverse datasets without requiring manual annotation.
The focal point of egocentric video understanding is modelling hand-object interactions. Standard models, e.g. CNNs or Vision Transformers, which receive RGB frames as input perform well. However, their performance improves further by employing additional input modalities that provide complementary cues, such as object detections, optical flow, audio, etc. The added complexity of the modality-specific modules, on the other hand, makes these models impractical for deployment. The goal of this work is to retain the performance of such a multimodal approach, while using only the RGB frames as input at inference time. We demonstrate that for egocentric action recognition on the Epic-Kitchens and the Something-Something datasets, students which are taught by multimodal teachers tend to be more accurate and better calibrated than architecturally equivalent models trained on ground truth labels in a unimodal or multimodal fashion. We further adopt a principled multimodal knowledge distillation framework, allowing us to deal with issues which occur when applying multimodal knowledge distillation in a naive manner. Lastly, we demonstrate the achieved reduction in computational complexity, and show that our approach maintains higher performance with the reduction of the number of input views. We release our code at https://github.com/gorjanradevski/multimodal-distillation.
Several Deep Learning (DL) methods have recently been proposed for an automated identification of kidney stones during an ureteroscopy to enable rapid therapeutic decisions. Even if these DL approaches led to promising results, they are mainly appropriate for kidney stone types for which numerous labelled data are available. However, only few labelled images are available for some rare kidney stone types. This contribution exploits Deep Metric Learning (DML) methods i) to handle such classes with few samples, ii) to generalize well to out of distribution samples, and iii) to cope better with new classes which are added to the database. The proposed Guided Deep Metric Learning approach is based on a novel architecture which was designed to learn data representations in an improved way. The solution was inspired by Few-Shot Learning (FSL) and makes use of a teacher-student approach. The teacher model (GEMINI) generates a reduced hypothesis space based on prior knowledge from the labeled data, and is used it as a guide to a student model (i.e., ResNet50) through a Knowledge Distillation scheme. Extensive tests were first performed on two datasets separately used for the recognition, namely a set of images acquired for the surfaces of the kidney stone fragments, and a set of images of the fragment sections. The proposed DML-approach improved the identification accuracy by 10% and 12% in comparison to DL-methods and other DML-approaches, respectively. Moreover, model embeddings from the two dataset types were merged in an organized way through a multi-view scheme to simultaneously exploit the information of surface and section fragments. Test with the resulting mixed model improves the identification accuracy by at least 3% and up to 30% with respect to DL-models and shallow machine learning methods, respectively.
Knowledge distillation (KD) has shown great potential for transferring knowledge from a complex teacher model to a simple student model in which the heavy learning task can be accomplished efficiently and without losing too much prediction accuracy. Recently, many attempts have been made by applying the KD mechanism to the graph representation learning models such as graph neural networks (GNNs) to accelerate the model's inference speed via student models. However, many existing KD-based GNNs utilize MLP as a universal approximator in the student model to imitate the teacher model's process without considering the graph knowledge from the teacher model. In this work, we provide a KD-based framework on multi-scaled GNNs, known as graph framelet, and prove that by adequately utilizing the graph knowledge in a multi-scaled manner provided by graph framelet decomposition, the student model is capable of adapting both homophilic and heterophilic graphs and has the potential of alleviating the over-squashing issue with a simple yet effectively graph surgery. Furthermore, we show how the graph knowledge supplied by the teacher is learned and digested by the student model via both algebra and geometry. Comprehensive experiments show that our proposed model can generate learning accuracy identical to or even surpass the teacher model while maintaining the high speed of inference.
Ship orientation angle prediction (SOAP) with optical remote sensing images is an important image processing task, which often relies on deep convolutional neural networks (CNNs) to make accurate predictions. This paper proposes a novel framework to reduce the model sizes and computational costs of SOAP models without harming prediction accuracy. First, a new SOAP model called Mobile-SOAP is designed based on MobileNetV2, achieving state-of-the-art prediction accuracy. Four tiny SOAP models are also created by replacing the convolutional blocks in Mobile-SOAP with four small-scale networks, respectively. Then, to transfer knowledge from Mobile-SOAP to four lightweight models, we propose a novel knowledge distillation (KD) framework termed SOAP-KD consisting of a novel feature-based guidance loss and an optimized synthetic samples-based knowledge transfer mechanism. Lastly, extensive experiments on the FGSC-23 dataset confirm the superiority of Mobile-SOAP over existing models and also demonstrate the effectiveness of SOAP-KD in improving the prediction performance of four specially designed tiny models. Notably, by using SOAP-KD, the test mean absolute error of the ShuffleNetV2x1.0-based model is only 8% higher than that of Mobile-SOAP, but its number of parameters and multiply-accumulate operations (MACs) are respectively 61.6% and 60.8% less.
In the context of label-efficient learning on video data, the distillation method and the structural design of the teacher-student architecture have a significant impact on knowledge distillation. However, the relationship between these factors has been overlooked in previous research. To address this gap, we propose a new weakly supervised learning framework for knowledge distillation in video classification that is designed to improve the efficiency and accuracy of the student model. Our approach leverages the concept of substage-based learning to distill knowledge based on the combination of student substages and the correlation of corresponding substages. We also employ the progressive cascade training method to address the accuracy loss caused by the large capacity gap between the teacher and the student. Additionally, we propose a pseudo-label optimization strategy to improve the initial data label. To optimize the loss functions of different distillation substages during the training process, we introduce a new loss method based on feature distribution. We conduct extensive experiments on both real and simulated data sets, demonstrating that our proposed approach outperforms existing distillation methods in terms of knowledge distillation for video classification tasks. Our proposed substage-based distillation approach has the potential to inform future research on label-efficient learning for video data.
2D RGB images and 3D LIDAR point clouds provide complementary knowledge for the perception system of autonomous vehicles. Several 2D and 3D fusion methods have been explored for the LIDAR semantic segmentation task, but they suffer from different problems. 2D-to-3D fusion methods require strictly paired data during inference, which may not be available in real-world scenarios, while 3D-to-2D fusion methods cannot explicitly make full use of the 2D information. Therefore, we propose a Bidirectional Fusion Network with Cross-Modality Knowledge Distillation (CMDFusion) in this work. Our method has two contributions. First, our bidirectional fusion scheme explicitly and implicitly enhances the 3D feature via 2D-to-3D fusion and 3D-to-2D fusion, respectively, which surpasses either one of the single fusion schemes. Second, we distillate the 2D knowledge from a 2D network (Camera branch) to a 3D network (2D knowledge branch) so that the 3D network can generate 2D information even for those points not in the FOV (field of view) of the camera. In this way, RGB images are not required during inference anymore since the 2D knowledge branch provides 2D information according to the 3D LIDAR input. We show that our CMDFusion achieves the best performance among all fusion-based methods on SemanticKITTI and nuScenes datasets. The code will be released at https://github.com/Jun-CEN/CMDFusion.
Data-free knowledge distillation (DFKD) aims to obtain a lightweight student model without original training data. Existing works generally synthesize data from the pre-trained teacher model to replace the original training data for student learning. To more effectively train the student model, the synthetic data shall be customized to the current student learning ability. However, this is ignored in the existing DFKD methods and thus negatively affects the student training. To address this issue, we propose Customizing Synthetic Data for Data-Free Student Learning (CSD) in this paper, which achieves adaptive data synthesis using a self-supervised augmented auxiliary task to estimate the student learning ability. Specifically, data synthesis is dynamically adjusted to enlarge the cross entropy between the labels and the predictions from the self-supervised augmented task, thus generating hard samples for the student model. The experiments on various datasets and teacher-student models show the effectiveness of our proposed method. Code is available at: $\href{https://github.com/luoshiya/CSD}{https://github.com/luoshiya/CSD}$
For many real-world time series tasks, the computational complexity of prevalent deep leaning models often hinders the deployment on resource-limited environments (e.g., smartphones). Moreover, due to the inevitable domain shift between model training (source) and deploying (target) stages, compressing those deep models under cross-domain scenarios becomes more challenging. Although some of existing works have already explored cross-domain knowledge distillation for model compression, they are either biased to source data or heavily tangled between source and target data. To this end, we design a novel end-to-end framework called Universal and joint knowledge distillation (UNI-KD) for cross-domain model compression. In particular, we propose to transfer both the universal feature-level knowledge across source and target domains and the joint logit-level knowledge shared by both domains from the teacher to the student model via an adversarial learning scheme. More specifically, a feature-domain discriminator is employed to align teacher's and student's representations for universal knowledge transfer. A data-domain discriminator is utilized to prioritize the domain-shared samples for joint knowledge transfer. Extensive experimental results on four time series datasets demonstrate the superiority of our proposed method over state-of-the-art (SOTA) benchmarks.