In this work, we propose a disentangled latent optimization-based method for parameterizing grouped deforming 3D objects into shape and deformation factors in an unsupervised manner. Our approach involves the joint optimization of a generator network along with the shape and deformation factors, supported by specific regularization techniques. For efficient amortized inference of disentangled shape and deformation codes, we train two order-invariant PointNet-based encoder networks in the second stage of our method. We demonstrate several significant downstream applications of our method, including unsupervised deformation transfer, deformation classification, and explainability analyses. Extensive experiments conducted on 3D human, animal, and facial expression datasets demonstrate that our simple approach is highly effective in these downstream tasks, comparable or superior to existing methods with much higher complexity.

Abstract With the refinement of 3D imaging modality, cryo-electron tomography (cryo-ET) has emerged as a powerful technique for the structural analysis of macromolecular complexes at near-atomic resolution. Recent advancements in volumetric segmentation methods applied to cryo-ET datasets have garnered significant attention within the biomedical sector. However, existing methods rely heavily on manually labeled data, which demands highly specialized expertise, making fully supervised approaches less feasible for cryo-ET images. To address this, a number of unsupervised domain adaptation (UDA) techniques have been developed to improve segmentation network performance using unlabeled data. Nevertheless, directly applying these methods to cryo-ET image segmentation presents two major challenges: 1) the source dataset, usually obtained through simulation, contains a fixed level of noise, while the target dataset, being directly collected from raw-data from the real-world scenario, has unpredictable noise levels; 2) the source data used for training typically consists of known macromolecules, in contrast, the target domain data are often unknown, causing the model to be biased towards those known macromolecules, leading to a domain shift problem. To address such challenges, in this paper, we introduce a voxel-wise unsupervised domain adaptation approach, termed Vox-UDA, specifically for cryo-ET subtomogram segmentation. Vox-UDA incorporates a noise generation module to simulate target-like noises in the source dataset for cross-noise level adaptation, and a denoised pseudo-labeling strategy based on the improved bilateral filter to alleviate the domain shift problem. Additionally, we further consider a scenario that is more in line with the real world, where the target (experimental) dataset might not be accessible during training or has a very small sample size, and we present a voxel-wise zero-shot domain adaptation (ZSDA) approach, named Vox-ZSDA. In Vox-ZSDA, we introduce a self-supervised graph learning strategy to eliminate any dependency on the target data, accompanied by a dynamic graph contrastive learning technique to enhance the model’s sensitivity to macromolecular structures to boost the segmentation performance. More importantly, we construct the first UDA and ZSDA cryo-ET subtomogram segmentation benchmark on three experimental datasets. Extensive experimental results on multiple benchmarks and newly curated real-world datasets demonstrate the superiority of our proposed approach compared to state-of-the-art UDA and ZSDA methods.

Teacher-student frameworks have emerged as a leading approach in semi-supervised medical image segmentation, demonstrating strong performance across various tasks. However, the learning effects are still limited by the strong correlation and unreliable knowledge transfer process between teacher and student networks. To overcome this limitation, we introduce a novel switching Dual-Student architecture that strategically selects the most reliable student at each iteration to enhance dual-student collaboration and prevent error reinforcement. We also introduce a strategy of Loss-Aware Exponential Moving Average to dynamically ensure that the teacher absorbs meaningful information from students, improving the quality of pseudo-labels. Our plug-and-play framework is extensively evaluated on 3D medical image segmentation datasets, where it outperforms state-of-the-art semi-supervised methods, demonstrating its effectiveness in improving segmentation accuracy under limited supervision.

Cryo-Electron Tomography (cryo-ET) is a 3D imaging technology that facilitates the study of macromolecular structures at near-atomic resolution. Recent volumetric segmentation approaches on cryo-ET images have drawn widespread interest in the biological sector. However, existing methods heavily rely on manually labeled data, which requires highly professional skills, thereby hindering the adoption of fully-supervised approaches for cryo-ET images. Some unsupervised domain adaptation (UDA) approaches have been designed to enhance the segmentation network performance using unlabeled data. However, applying these methods directly to cryo-ET image segmentation tasks remains challenging due to two main issues: 1) the source dataset, usually obtained through simulation, contains a fixed level of noise, while the target dataset, directly collected from raw-data from the real-world scenario, have unpredictable noise levels. 2) the source data used for training typically consists of known macromoleculars. In contrast, the target domain data are often unknown, causing the model to be biased towards those known macromolecules, leading to a domain shift problem. To address such challenges, in this work, we introduce a voxel-wise unsupervised domain adaptation approach, termed Vox-UDA, specifically for cryo-ET subtomogram segmentation. Vox-UDA incorporates a noise generation module to simulate target-like noises in the source dataset for cross-noise level adaptation. Additionally, we propose a denoised pseudo-labeling strategy based on the improved Bilateral Filter to alleviate the domain shift problem. More importantly, we construct the first UDA cryo-ET subtomogram segmentation benchmark on three experimental datasets. Extensive experimental results on multiple benchmarks and newly curated real-world datasets demonstrate the superiority of our proposed approach compared to state-of-the-art UDA methods.

Subtomogram alignment is a critical task in cryo-electron tomography (cryo-ET) analysis, essential for achieving high-resolution reconstructions of macromolecular complexes. However, learning effective positional representations remains challenging due to limited labels and high noise levels inherent in cryo-ET data. In this work, we address this challenge by proposing a self-supervised learning approach that leverages intrinsic geometric transformations as implicit supervisory signals, enabling robust representation learning despite data scarcity. We introduce BOE-ViT, the first Vision Transformer (ViT) framework for 3D subtomogram alignment. Recognizing that traditional ViTs lack equivariance and are therefore suboptimal for orientation estimation, we enhance the model with two innovative modules that introduce equivariance include 1) the Polyshift module for improved shift estimation and 2) Multi-Axis Rotation Encoding (MARE) for enhanced rotation estimation. Experimental results demonstrate that BOE-ViT significantly outperforms state-of-the-art methods. Notably, at SNR 0.01 dataset, our approach achieves a 77.3% reduction in rotation estimation error and a 62.5% reduction in translation estimation error, effectively overcoming the challenges in cryo-ET subtomogram alignment.

In recent years, the interpretability of Deep Neural Networks (DNNs) has garnered significant attention, particularly due to their widespread deployment in critical domains like healthcare, finance, and autonomous systems. To address the challenge of understanding how DNNs make decisions, Explainable AI (XAI) methods, such as saliency maps, have been developed to provide insights into the inner workings of these models. This paper introduces DiffCAM, a novel XAI method designed to overcome limitations in existing Class Activation Map (CAM)-based techniques, which often rely on decision boundary gradients to estimate feature importance. DiffCAM differentiates itself by considering the actual data distribution of the reference class, identifying feature importance based on how a target example differs from reference examples. This approach captures the most discriminative features without relying on decision boundaries or prediction results, making DiffCAM applicable to a broader range of models, including foundation models. Through extensive experiments, we demonstrate the superior performance and flexibility of DiffCAM in providing meaningful explanations across diverse datasets and scenarios.

Visual Prompt Tuning (VPT) has emerged as a parameter-efficient fine-tuning paradigm for vision transformers, with conventional approaches utilizing dataset-level prompts that remain the same across all input instances. We observe that this strategy results in sub-optimal performance due to high variance in downstream datasets. To address this challenge, we propose Visual Instance-aware Prompt Tuning (ViaPT), which generates instance-aware prompts based on each individual input and fuses them with dataset-level prompts, leveraging Principal Component Analysis (PCA) to retain important prompting information. Moreover, we reveal that VPT-Deep and VPT-Shallow represent two corner cases based on a conceptual understanding, in which they fail to effectively capture instance-specific information, while random dimension reduction on prompts only yields performance between the two extremes. Instead, ViaPT overcomes these limitations by balancing dataset-level and instance-level knowledge, while reducing the amount of learnable parameters compared to VPT-Deep. Extensive experiments across 34 diverse datasets demonstrate that our method consistently outperforms state-of-the-art baselines, establishing a new paradigm for analyzing and optimizing visual prompts for vision transformers.

Cryo-electron tomography (cryo-ET) is confronted with the intricate task of unveiling novel structures. General class discovery (GCD) seeks to identify new classes by learning a model that can pseudo-label unannotated (novel) instances solely using supervision from labeled (base) classes. While 2D GCD for image data has made strides, its 3D counterpart remains unexplored. Traditional methods encounter challenges due to model bias and limited feature transferability when clustering unlabeled 2D images into known and potentially novel categories based on labeled data. To address this limitation and extend GCD to 3D structures, we propose an innovative approach that harnesses a pretrained 2D transformer, enriched by an effective weight inflation strategy tailored for 3D adaptation, followed by a decoupled prototypical network. Incorporating the power of pretrained weight-inflated Transformers, we further integrate CLIP, a vision-language model to incorporate textual information. Our method synergizes a graph convolutional network with CLIP's frozen text encoder, preserving class neighborhood structure. In order to effectively represent unlabeled samples, we devise semantic distance distributions, by formulating a bipartite matching problem for category prototypes using a decoupled prototypical network. Empirical results unequivocally highlight our method's potential in unveiling hitherto unknown structures in cryo-ET. By bridging the gap between 2D GCD and the distinctive challenges of 3D cryo-ET data, our approach paves novel avenues for exploration and discovery in this domain.

• We propose MVGFormer, a multi-view fusion framework with a dual-stream encoder guided by a visual graph. • We design two decoder variants: MF for hierarchical feature fusion and P3DA for multi-scale representation. • We introduce a view-masked self-supervised strategy that reconstructs masked views from remaining ones. • MVGFormer achieves superior performance over existing state-of-the-art cryo-ET segmentation methods. Cryo-Electron Tomography (cryo-ET) is a cutting-edge 3D imaging technology that enables detailed examination of biological macromolecular structures at near-atomic resolution. Recent deep learning applications on cryo-ET, such as cryo-ET segmentation, have drawn widespread interest for their potential to improve particle alignment, classification, and other tasks. However, current methods heavily rely on convolutional architectures, which prioritize local information while neglecting the global structural information inherent in cryo-ET data. Transformer-based models, known for their large receptive field, have become the de-facto design for 2D vision tasks due to their ability to effectively capture global information. This approach is also well-suited for 3D tasks, given the complex nature of 3D objects. Based on this, we extend 2D vision transformers into 3D and propose a novel transformer-based framework for cryo-ET segmentation, named MVGFormer. MVGFormer introduces a multi-view perspective fusion transformer encoder, which captures rich global structural information from multiple perspectives using unique positional embeddings. To enhance contextual awareness, we design a parallel context encoder that builds a visual graph to guide attention. We further introduce two complementary 3D decoders: multi-level feature fusion (MF) and parallel atrous convolutions (P3DA), which together capture multi-scale structural cues for precise segmentation. Furthermore, we introduce a view-masked self-supervised learning strategy to reinforce the effectiveness of the multi-view design and improve the model’s representation capability. To our knowledge, MVGFormer is the first transformer-based model for cryo-ET segmentation. We empirically evaluate MVGFormer on six cryo-ET datasets across three different tasks. Extensive experimental results demonstrate its superiority over state-of-the-art 3D segmentation methods.

The advent of large-scale pretrained language models (PLMs) has contributed greatly to the progress in natural language processing (NLP). Despite its recent success and wide adoption, fine-tuning a PLM often suffers from overfitting, which leads to poor generalizability due to the extremely high complexity of the model and the limited training samples from downstream tasks. To address this problem, we propose a novel and effective fine-tuning framework, named layerwise noise stability regularization (LNSR). Specifically, our method perturbs the input of neural networks with the standard Gaussian or in-manifold noise in the representation space and regularizes each layer's output of the language model. We provide theoretical and experimental analyses to prove the effectiveness of our method. The empirical results show that our proposed method outperforms several state-of-the-art algorithms, such as norm and start point (L2-SP), Mixout, FreeLB, and smoothness inducing adversarial regularization and Bregman proximal point optimization (SMART). In addition to evaluating the proposed method on relatively simple text classification tasks, similar to the prior works, we further evaluate the effectiveness of our method on more challenging question-answering (QA) tasks. These tasks present a higher level of difficulty, and they provide a larger amount of training examples for tuning a well-generalized model. Furthermore, the empirical results indicate that our proposed method can improve the ability of language models to domain generalization.

To explain the prediction result of a Deep Neural Network (DNN) model based on a given sample, LIME [1] and its derivatives have been proposed to approximate the local behavior of the DNN model around the data point via linear surrogates. Though these algorithms interpret the DNN by finding the key features used for classification, the random interpolations used by LIME would perturb the explanation result and cause the instability and inconsistency between repetitions of LIME computations. To tackle this issue, we propose G-LIME that extends the vanilla LIME through high-dimensional Bayesian linear regression using the sparsity and informative global priors. Specifically, with a dataset representing the population of samples (e.g., the training set), G-LIME first pursues the global explanation of the DNN model using the whole dataset. Then, with a new data point, -LIME incorporates an modified estimator of ElasticNet-alike to refine the local explanation result through balancing the distance to the global explanation and the sparsity/feature selection in the explanation. Finally, G-LIME uses Least Angle Regression (LARS) and retrieves the solution path of a modified ElasticNet under varying -regularization, to screen and rank the importance of features [2] as the explanation result. Through extensive experiments on real world tasks, we show that the proposed method yields more stable, consistent, and accurate results compared to LIME.

Uncertainty estimation for unlabeled data is crucial to active learning. With a deep neural network employed as the backbone model, the data selection process is highly challenging due to the potential over-confidence of the model inference. Existing methods resort to special learning fashions (e.g. adversarial) or auxiliary models to address this challenge. This tends to result in complex and inefficient pipelines, which would render the methods impractical. In this work, we propose a novel algorithm that leverages noise stability to estimate data uncertainty. The key idea is to measure the output derivation from the original observation when the model parameters are randomly perturbed by noise. We provide theoretical analyses by leveraging the small Gaussian noise theory and demonstrate that our method favors a subset with large and diverse gradients. Our method is generally applicable in various tasks, including computer vision, natural language processing, and structural data analysis. It achieves competitive performance compared against state-of-the-art active learning baselines.

MOTIVATION: Deep learning models have achieved remarkable success in a wide range of natural-world tasks, such as vision, language, and speech recognition. These accomplishments are largely attributed to the availability of open-source large-scale datasets. More importantly, pre-trained foundational modellearnings exhibit a surprising degree of transferability to downstream tasks, enabling efficient learning even with limited training examples. However, the application of such natural-domain models to the domain of tiny Cryo-Electron Tomography (Cryo-ET) images has been a relatively unexplored frontier. This research is motivated by the intuition that 3D Cryo-ET voxel data can be conceptually viewed as a sequence of progressively evolving video frames. RESULTS: Leveraging the above insight, we propose a novel approach that involves the utilization of 3D models pre-trained on large-scale video datasets to enhance Cryo-ET subtomogram classification. Our experiments, conducted on both simulated and real Cryo-ET datasets, reveal compelling results. The use of video initialization not only demonstrates improvements in classification accuracy but also substantially reduces training costs. Further analyses provide additional evidence of the value of video initialization in enhancing subtomogram feature extraction. Additionally, we observe that video initialization yields similar positive effects when applied to medical 3D classification tasks, underscoring the potential of cross-domain knowledge transfer from video-based models to advance the state-of-the-art in a wide range of biological and medical data types. AVAILABILITY AND IMPLEMENTATION: https://github.com/xulabs/aitom.

Abstract Fine-tuning deep neural networks pre-trained on large scale datasets is one of the most practical transfer learning paradigm given limited quantity of training samples. To obtain better generalization, using the starting point as the reference (SPAR), either through weights or features, has been successfully applied to transfer learning as a regularizer. However, due to the domain discrepancy between the source and target task, there exists obvious risk of negative transfer in a straightforward manner of knowledge preserving. In this paper, we propose a novel transfer learning algorithm, introducing the idea of Target-awareness REpresentation Disentanglement ( $$\textrm{TRED}$$ TRED ), where the relevant knowledge with respect to the target task is disentangled from the original source model and used as a regularizer during fine-tuning the target model. Two alternative approaches, maximizing Maximum Mean Discrepancy (Max-MMD) and minimizing mutual information (Min-MI) are introduced to achieve the desired disentanglement. Experiments on various real world datasets show that our method stably improves the standard fine-tuning by more than 2% in average. $$\textrm{TRED}$$ TRED also outperforms related state-of-the-art transfer learning regularizers such as $$\mathrm {L^2\text {-}SP}$$ L 2 - SP , $$\textrm{AT}$$ AT , $$\textrm{DELTA}$$ DELTA , and $$\textrm{BSS}$$ BSS . Moreover, our solution is compatible with different choices of disentangling strategies. While the combination of Max-MMD and Min-MI typically achieves higher accuracy, only using Max-MMD can be a preferred choice in applications with low resource budgets.

Fine-tuning pre-trained language models, such as BERT, has shown enormous success among various NLP tasks. Though simple and effective, the process of fine-tuning has been found unstable, which often leads to unexpected poor performance. To increase stability and generalizability, most existing works resort to maintaining the parameters or representations of pre-trained models during fine-tuning. Nevertheless, very little work explores mining the reliable part of pre-learned information that can help to stabilize fine-tuning. To address this challenge, we introduce a novel solution in which we fine-tune BERT with stabilized cross-layer mutual information. Our method aims to preserve the reliable behaviors of cross-layer information propagation, instead of preserving the information itself, of the pre-trained model. Therefore, our method circumvents the domain conflicts between pre-trained and target tasks. We conduct extensive experiments with popular pre-trained BERT variants on NLP datasets, demonstrating the universal effectiveness and robustness of our method.

Transfer learning has been a popular learning paradigm in the deep learning era, especially in annotation-insufficient scenarios. Better ImageNet pre-trained models have been demonstrated, from the perspective of architecture, by previous research to have better transferability to downstream tasks[26]. However, in this paper, we find that during the same pre-training process, models at middle epochs, which are inadequately pre-trained, can outperform fully trained models when used as feature extractors (FE), while the fine-tuning (FT) performance still grows with the source performance. This reveals that there is not a solid positive correlation between top-1 accuracy on ImageNet and the transferring result on target data. Based on the contradictory phenomenon between FE and FT that a better feature extractor fails to be fine-tuned better accordingly, we conduct comprehensive analyses on features before the softmax layer to provide insightful explanations. Our discoveries suggest that, during pre-training, models tend to first learn spectral components corresponding to large singular values and the residual components contribute more when fine-tuning.

Abstract Although more layers and more parameters generally improve the accuracy of the models, such big models generally have high computational complexity and require big memory, which exceed the capacity of small devices for inference and incurs long training time. In addition, it is difficult to afford long training time and inference time of big models even in high performance servers, as well. As an efficient approach to compress a large deep model (a teacher model) to a compact model (a student model), knowledge distillation emerges as a promising approach to deal with the big models. Existing knowledge distillation methods cannot exploit the elastic available computing resources and correspond to low efficiency. In this paper, we propose an Elastic Deep Learning framework for knowledge Distillation, that is, EDL‐Dist. The advantages of EDL‐Dist are threefold. First, the inference and the training process is separated. Second, elastic available computing resources can be utilized to improve the efficiency. Third, fault‐tolerance of the training and inference processes is supported. We take extensive experimentation to show that the throughput of EDL‐Dist is up to 3.125 times faster than the baseline method (online knowledge distillation) while the accuracy is similar or higher.

Central to active learning (AL) is what data should be selected for annotation. Existing works attempt to select highly uncertain or informative data for annotation. Nevertheless, it remains unclear how selected data impacts the test performance of the task model used in AL. In this work, we explore such an impact by theoretically proving that selecting unlabeled data of higher gradient norm leads to a lower upper-bound of test loss, resulting in a better test performance. However, due to the lack of label information, directly computing gradient norm for unlabeled data is infeasible. To address this challenge, we propose two schemes, namely expected-gradnorm and entropy-gradnorm. The former computes the gradient norm by constructing an expected empirical loss while the latter constructs an unsupervised loss with entropy. Furthermore, we integrate the two schemes in a universal AL framework. We evaluate our method on classical image classification and semantic segmentation tasks. To demonstrate its competency in domain applications and its robustness to noise, we also validate our method on a cellular imaging analysis task, namely cryo-Electron Tomography subtomogram classification. Results demonstrate that our method achieves superior performance against the state of the art. We refer readers to https://arxiv.org/pdf/2112.05683.pdf for the full version of this paper which includes the appendix and source code link.

Techniques to explain the predictions of deep neural networks (DNNs) have been largely required for gaining insights into the black boxes. We introduce InterpretDL, a toolkit of explanation algorithms based on PaddlePaddle, with uniformed programming interfaces and "plug-and-play" designs. A few lines of codes are needed to obtain the explanation results without modifying the structure of the model. InterpretDL currently contains 16 algorithms, explaining training phases, datasets, global and local behaviors of post-trained deep models. InterpretDL also provides a number of tutorial examples and showcases to demonstrate the capability of InterpretDL working on a wide range of deep learning models, e.g., Convolutional Neural Networks (CNNs), Multi-Layer Preceptors (MLPs), Transformers, etc., for various tasks in both Computer Vision (CV) and Natural Language Processing (NLP). Furthermore, InterpretDL modularizes the implementations, making efforts to support the compatibility across frameworks. The project is available at https://github.com/PaddlePaddle/InterpretDL.

Transfer learning through fine-tuning a pre-trained neural network with an extremely large dataset, such as ImageNet, can significantly improve and accelerate training while the accuracy is frequently bottlenecked by the limited dataset size of the new target task. To solve the problem, some regularization methods, constraining the outer layer weights of the target network using the starting point as references (SPAR), have been studied. In this article, we propose a novel regularized transfer learning framework \operatorname{DELTA} , namely DE ep L earning T ransfer using Feature Map with A ttention . Instead of constraining the weights of neural network, \operatorname{DELTA} aims at preserving the outer layer outputs of the source network. Specifically, in addition to minimizing the empirical loss, \operatorname{DELTA} aligns the outer layer outputs of two networks, through constraining a subset of feature maps that are precisely selected by attention that has been learned in a supervised learning manner. We evaluate \operatorname{DELTA} with the state-of-the-art algorithms, including L^2 and \emph {L}^2\text{-}SP . The experiment results show that our method outperforms these baselines with higher accuracy for new tasks. Code has been made publicly available. 1

Regularization that incorporates the linear combination of empirical loss and explicit regularization terms as the loss function has been frequently used for many machine learning tasks. The explicit regularization term is designed in different types, depending on its applications. While regularized learning often boost the performance with higher accuracy and faster convergence, the regularization would sometimes hurt the empirical loss minimization and lead to poor performance. To deal with such issues in this work, we propose a novel strategy, namely Gr adients O rthogonal D ecomposition ( GrOD ), that improves the training procedure of regularized deep learning. Instead of linearly combining gradients of the two terms, GrOD re-estimates a new direction for iteration that does not hurt the empirical loss minimization while preserving the regularization affects, through orthogonal decomposition. We have performed extensive experiments to use GrOD improving the commonly used algorithms of transfer learning [ 2 ], knowledge distillation [ 3 ], and adversarial learning [ 4 ]. The experiment results based on large datasets, including Caltech 256 [ 5 ], MIT indoor 67 [ 6 ], CIFAR-10 [ 7 ], and ImageNet [ 8 ], show significant improvement made by GrOD for all three algorithms in all cases.

Temporal relational modeling in video is essential for human action understanding, such as action recognition and action segmentation. Although Graph Convolution Networks (GCNs) have shown promising advantages in relation reasoning on many tasks, it is still a challenge to apply graph convolution networks on long video sequences effectively. The main reason is that large number of nodes (i.e., video frames) makes GCNs hard to capture and model temporal relations in videos. To tackle this problem, in this paper, we introduce an effective GCN module, Dilated Temporal Graph Reasoning Module (DTGRM), designed to model temporal relations and dependencies between video frames at various time spans. In particular, we capture and model temporal relations via constructing multi-level dilated temporal graphs where the nodes represent frames from different moments in video. Moreover, to enhance temporal reasoning ability of the proposed model, an auxiliary self-supervised task is proposed to encourage the dilated temporal graph reasoning module to find and correct wrong temporal relations in videos. Our DTGRM model outperforms state-of-the-art action segmentation models on three challenging datasets: 50Salads, Georgia Tech Egocentric Activities (GTEA), and the Breakfast dataset. The code is available at https://github.com/redwang/DTGRM.

While recent studies on semi-supervised learning have shown remarkable progress in leveraging both labeled and unlabeled data, most of them presume a basic setting of the model is randomly initialized. In this work, we consider semi-supervised learning and transfer learning jointly, leading to a more practical and competitive paradigm that can utilize both powerful pre-trained models from source domain as well as labeled/unlabeled data in the target domain. To better exploit the value of both pre-trained weights and unlabeled target examples, we introduce adaptive consistency regularization that consists of two complementary components: Adaptive Knowledge Consistency (AKC) on the examples between the source and target model, and Adaptive Representation Consistency (ARC) on the target model between labeled and unlabeled examples. Examples involved in the consistency regularization are adaptively selected according to their potential contributions to the target task. We conduct extensive experiments on popular benchmarks including CIFAR-10, CUB-200, and MURA, by fine-tuning the ImageNet pre-trained ResNet-50 model. Results show that our proposed adaptive consistency regularization outperforms state-of-the-art semi-supervised learning techniques such as Pseudo Label, Mean Teacher, and FixMatch. Moreover, our algorithm is orthogonal to existing methods and thus able to gain additional improvements on top of MixMatch and FixMatch. Our code is available at https://github.com/Walleclipse/Semi-Supervised-Transfer-Learning-Paddle.

Hang Hua, Xingjian Li, Dejing Dou, Chengzhong Xu, Jiebo Luo. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Ensemble learning is a widely used technique to train deep convolutional neural networks (CNNs) for improved robustness and accuracy. While existing algorithms usually first train multiple diversified networks and then assemble these networks as an aggregated classifier, we propose a novel learning paradigm, namely, “In-Network Ensemble” ( INE ) that incorporates the diversity of multiple models through training a SINGLE deep neural network. Specifically, INE segments the outputs of the CNN into multiple independent classifiers, where each classifier is further fine-tuned with better accuracy through a so-called diversified knowledge distillation process . We then aggregate the fine-tuned independent classifiers using an Averaging-and-Softmax operator to obtain the final ensemble classifier. Note that, in the supervised learning settings, INE starts the CNN training from random, while, under the transfer learning settings, it also could start with a pre-trained model to incorporate the knowledge learned from additional datasets. Extensive experiments have been done using eight large-scale real-world datasets, including CIFAR, ImageNet, and Stanford Cars, among others, as well as common deep network architectures such as VGG, ResNet, and Wide ResNet. We have evaluated the method under two tasks: supervised learning and transfer learning. The results show that INE outperforms the state-of-the-art algorithms for deep ensemble learning with improved accuracy.

As it is difficult to directly share data among different organizations, data federation brings new opportunities to the data-related cooperation among different organizations by providing abstract data interfaces. With the development of Cloud computing, organizations store data on the Cloud to achieve elasticity and scalability for data processing. The existing data placement approaches generally only consider one aspect, which is either communication cost or time cost, and do not consider the features of jobs that process the data. In this paper, we propose an approach to enable secure data processing on the Cloud with the data from different organizations. The approach consists of a data federation platform for secure data processing on the Cloud named FedCube and a greedy data placement algorithm that creates a plan to store data on the Cloud in order to achieve multiple objectives based on a cost model. The cost model is composed of two objectives, i.e., reducing both monetary cost and execution time. We present an experimental evaluation by comparing our data placement algorithm with the existing methods based on the data federation platform. The experiments show that our proposed algorithm significantly reduce the total cost (up to 69.8%).

Fine-tuning the deep convolution neural network (CNN) using a pre-trained model helps transfer knowledge learned from larger datasets to the target task. While the accuracy could be largely improved even when the training dataset is small, the transfer learning outcome is similar with the pre-trained one with closed CNN weights[17], as the backpropagation here brings less updates to deeper CNN layers. In this work, we propose RIFLE - a simple yet effective strategy that deepens backpropagation in transfer learning settings, through periodically ReInitializing the Fully-connected LayEr with random scratch during the fine-tuning procedure. RIFLE brings significant perturbation to the backpropagation process and leads to deep CNN weights update, while the affects of perturbation can be easily converged throughout the overall learning procedure. The experiments show that the use of RIFLE significantly improves deep transfer learning accuracy on a wide range of datasets, outperforming known tricks for the similar purpose, such as dropout, dropconnect, stochastic depth, and cyclic learning rate, under the same settings with 0.5%-2% higher testing accuracy. Empirical cases and ablation studies further indicate RIFLE brings meaningful updates to deep CNN layers with accuracy improved.

Transfer learning have been frequently used to improve deep neural network training through incorporating weights of pre-trained networks as the starting-point of optimization for regularization. While deep transfer learning can usually boost the performance with better accuracy and faster convergence, transferring weights from inappropriate networks hurts training procedure and may lead to even lower accuracy. In this paper, we consider deep transfer learning as minimizing a linear combination of empirical loss and regularizer based on pre-trained weights, where the regularizer would restrict the training procedure from lowering the empirical loss, with conflicted descent directions (e.g., derivatives). Following the view, we propose a novel strategy making regularization-based Deep Transfer learning Never Hurt (DTNH) that, for each iteration of training procedure, computes the derivatives of the two terms separately, then re-estimates a new descent direction that does not hurt the empirical loss minimization while preserving the regularization affects from the pre-trained weights. Extensive experiments have been done using common transfer learning regularizers, such as L2-SP and knowledge distillation, on top of a wide range of deep transfer learning benchmarks including Caltech, MIT indoor 67, CIFAR-10 and ImageNet. The empirical results show that the proposed descent direction estimation strategy DTNH can always improve the performance of deep transfer learning tasks based on all above regularizers, even when transferring pre-trained weights from inappropriate networks. All in all, DTNH strategy can improve state-of-the-art regularizers in all cases with 0.1%-7% higher accuracy in all experiments.

In heterogeneous systems that include CPUs and GPUs, the data transfers between these components play a critical role in determining the performance of applications. Software pipelining is a common approach to mitigate the overheads of those transfers. In this paper we investigate advanced software-pipelining optimizations for the double-precision general matrix multiplication (DGEMM) algorithm running on a heterogeneous system that includes ATI GPUs. Our approach decomposes the DGEMM workload to a finer detail and hides the latency of CPU-GPU data transfers to a higher degree than previous approaches in literature. We implement our approach in a five-stage software pipelined DGEMM and analyze its performance on a platform including x86 multi-core CPUs and an ATI Radeon™ HD5970 GPU that has two Cypress GPU chips on board. Our implementation delivers 758 GFLOPS (82% floating-point efficiency) when it uses only the GPU, and 844 GFLOPS (80% efficiency) when it distributes the workload on both CPU and GPU. We analyze the performance of our optimized DGEMM as the number of GPU chips employed grows from one to two, and the results show that resource contention on the PCIe bus and on the host memory are limiting factors.

Over the past decades, we noticed huge advances in FPGA technologies. The topic of floating-point accelerator on FPGA has gained renewed interests due to the increased device size and the emergence of fast hardware floating-point library. The popularity of FFT makes it easier to justify spending lots of effort doing detailed optimization. However, the ever increasing data size in some compelling application domains remains beyond the capability of existing FFT accelerators. The demand for more performance remains an active research topic. In this paper, leveraging structured description of FFT algorithms, we propose a FPGA-based FFT core generation framework, which emits Verilog HDL code given high-level algorithmic description and can handle radix-2 as well as prime-radix problem size. In particular, the proposed framework is optimized for 2D FFT and real FFT. The performance of our implementation is comparable with a commercial FFT IP. When compared with the latest results on GPU and CPU, measured in peak floating-point performance and energy efficiency, it shows that GPUs have outperformed FPGAs for FFT acceleration. However, we consider that FPGAs still have advantage in some situations.

Heterogeneous architecture is becoming an important way to build a massive parallel computer system, i.e. the CPU-GPU heterogeneous systems ranked in Top500 list. However, it is a challenge to efficiently utilize massive parallelism of both applications and architectures on such heterogeneous systems. In this paper we present a practice on how to exploit and orchestrate parallelism at algorithm level to take advantage of underlying parallelism at architecture level. A potential Petaflops application -- cryo-EM 3D reconstruction is selected as an example. We exploit all possible parallelism in cryo-EM 3D reconstruction, and leverage a self-adaptive dynamic scheduling algorithm to create a proper parallelism mapping between the application and architecture. The parallelized programs are evaluated on a subsystem of Dawning Nebulae supercomputer, whose node is composed of two Intel six-core Xeon CPUs and one Nvidia Fermi GPU. The experiment confirms that hierarchical parallelism is an efficient pattern of parallel programming to utilize capabilities of both CPU and GPU in a heterogeneous system. The CUDA kernels run more than 3 times faster than the OpenMP parallelized ones using 12 cores (threads). Based on the GPU-only version, the hybrid CPU-GPU program further improves the whole application's performance by 30% on the average.

Application-level protocol specifications are helpful for network security management, including intrusion detection and intrusion prevention which rely on monitoring technologies such as deep packet inspection. Moreover, detailed knowledge of protocol specifications is also an effective way of detecting malicious code. However, current methods for obtaining unknown and proprietary protocol message formats (i.e., no publicly available protocol specification), especially binary protocols, highly rely on manual operations, such as reverse engineering which is time-consuming and laborious. In this paper, we propose Biprominer, a tool that can automatically extract binary protocol message formats of an application from its real-world network trace. In addition, we present a transition probability model for a better description of the protocol. The chief feature of Biprominer is that it does not need to have any priori knowledge of protocol formats, because Biprominer is based on the statistical nature of the protocol format. We evaluate the efficacy of Biprominer over three binary protocols, with an average precision more than 99% and a recall better than 96.7%.