Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks259 from public data
- 2,465
- Ensembling neural networks: Many could be better than all2,019
The bias-variance decomposition of the error is provided in this paper, which shows that the success of GASEN may lie in that it can significantly reduce the bias as well as the variance.
- ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression1,895
ThiNet is proposed, an efficient and unified framework to simultaneously accelerate and compress CNN models in both training and inference stages, and it is revealed that it needs to prune filters based on statistics information computed from its next layer, not the current layer, which differentiates ThiNet from existing methods.
- 784
- ActionFormer: Localizing Moments of Actions with Transformers577
This work presents ActionFormer -- a simple yet powerful model to identify actions in time and recognize their categories in a single shot, without using action proposals or relying on pre-defined anchor windows.
- Deep Label Distribution Learning With Label Ambiguity511
The proposed deep label distribution learning (DLDL) method effectively utilizes the label ambiguity in both feature learning and classifier learning, which help prevent the network from overfitting even when the training set is small.
- Probabilistic End-To-End Noise Correction for Learning With Noisy Labels498
An end-to-end framework called PENCIL, which can update both network parameters and label estimations as label distributions and is more general and robust than existing methods and is easy to apply.
- Selective Convolutional Descriptor Aggregation for Fine-Grained Image Retrieval464
The selective convolutional descriptor aggregation (SCDA) method is proposed, which is unsupervised, using no image label or bounding box annotation, and achieves comparable retrieval results with the state-of-the-art general image retrieval approaches.
- 398
- Mask-CNN: Localizing parts and selecting descriptors for fine-grained bird species categorization392
It is proved that selecting useful deep descriptors contributes well to fine-grained image recognition, and a novel Mask-CNN model without the fully connected layers is proposed, which has a small feature dimensionality and efficient inference speed by comparing with other fine- grained approaches.
- Minimal gated unit for recurrent neural networks375
This work proposes a gated unit for RNN, named as minimal gated units (MGU), since it only contains one gate, which is a minimal design among all gated hidden units.
- 277
- Face Detection274
This chapter begins with a review of recent work in modeling face-specific information, including appearance-based methods used by sliding window classifiers, concepts from learning and local interest-point descriptors, and then focuses on representing the contextual information shared by faces with the surrounding scene.
- 264
- AutoPruner: An end-to-end trainable filter pruning method for efficient deep model inference258
This paper proposes an efficient channel selection layer, namely AutoPruner, to find less important filters automatically in a joint training manner and empirically demonstrates that the gradient information of this channel selectionlayer is also helpful for the whole model training.
- Face recognition with one training image per person257
This paper generalize and further enhance (PC) 2 A along two directions, which combines the original image with its second-order projections as well as its first-order projection in order to acquire more information from the original face, and applies principal component analysis (PCA) to such a set of the combined images.
- 230
- Residual Attention: A Simple but Effective Method for Multi-Label Recognition206
This work proposes an embarrassingly simple module, named class-specific residual attention (CSRA), which generates class- specific features for every category by proposing a simple spatial attention score, and then combines it with the class-agnostic average pooling feature.
- Age Estimation Using Expectation of Label Distribution Learning195
A lightweight network architecture is designed and a unified framework which can jointly learn age distribution and regress age is proposed which can achieve comparable results as the state-of-the-art even though model parameters are further reduced to 0.9M~(3.8MB disk storage).
- 192
- Exploit Bounding Box Annotations for Multi-Label Object Recognition185
This paper first extracts object proposals from each image, then proposes to make use of ground-truth bounding box annotations (strong labels) to add another level of local information by using nearest-neighbor relationships of local regions to form a multi-view pipeline.
- PBN184
This work combines the sensing power of on-body wireless sensors with the additional sensing power, computational resources, and user-friendly interface of an Android smartphone, and provides an accurate and efficient classification approach through the use of ensemble learning.
- On the Design of Cascades of Boosted Ensembles for Face Detection180
A novel method is proposed for making decisions about how many hypotheses to include in an ensemble and the appropriate balance of detection and false positive rates in the individual stages, which exploits the shape of the stage ROC curves in ways that have been previously ignored.
- Person Re-Identification with Correspondence Structure Learning175
A boosting-based approach to learn a correspondence structure which indicates the patch-wise matching probabilities between images from a target camera pair is introduced, which integrates a global matching constraint over the learned correspondence structure to exclude cross-view misalignments during the image patch matching process, hence achieving a more reliable matching score between images.
- Bag of Tricks for Long-Tailed Visual Recognition with Deep Convolutional Neural Networks168
This paper collects existing tricks in long-tailed visual recognition and performs extensive and systematic experiments in order to give a detailed experimental guideline and obtain an effective combination of these tricks, and proposes a novel data augmentation approach based on class activation maps for long-tail recognition.
- Piecewise Classifier Mappings: Learning Fine-Grained Learners for Novel Categories With Few Examples159
An end-to-end trainable deep network inspired by the state-of-the-art fine-grained recognition model and is tailored for the FSFG task is proposed, which generates the decision boundary via learning a set of more attainable sub-classifiers in a more parameter-economic way.
- Neural Network Pruning With Residual-Connections and Limited-Data148
CURL significantly outperforms previous state-of-the-art methods on ImageNet, and when pruning on small datasets, CURL achieves comparable or much better performance than fine-tuning a pretrained small model.
- 145
- Learning a Rare Event Detection Cascade by Direct Feature Selection145
A novel cascade learning algorithm based on forward feature selection which is two orders of magnitude faster than the Viola-Jones approach and yields classifiers of equivalent quality could be used for more demanding classification tasks, such as on-line learning.
- 143
- 139
- Genetic algorithm based selective neural network ensemble138
An approach named GASEN is proposed, which trains several individual neural networks and then employs genetic algorithm to select an optimum subset of individual networks to constitute an ensemble, which has preferable performance in generating ensembles with strong generalization ability in relatively small computational cost.
- Distilling Virtual Examples for Long-tailed Recognition136
It is proved that distilling from these virtual examples is equivalent to label distribution learning under certain constraints, and it is shown that when the virtual example distribution becomes flatter than the original input distribution, the under-represented tail classes will receive significant improvements, which is crucial in long-tailed recognition.
- 136
- 131
- 122
- 122
- Efficient face candidates selector for face detection119
An efficient face candidates selector is proposed for face detection tasks in still gray level images and is robust in that it can deal with illumination changes and moderate rotations.
- Rethinking the Route Towards Weakly Supervised Object Localization115
This paper proposes the pseudo supervised object localization (PSOL) method, which has good transferability across different datasets without fine-tuning and achieves 58.00% localization accuracy on ImageNet and 74.74% localization Accuracy on CUB-200, which have a large edge over previous models.
- Unsupervised object discovery and co-localization by deep descriptor transformation110
Deep Descriptor Transformation (DDT) is proposed, a simple yet effective method for evaluating the correlations of descriptors and then obtaining the category-consistent regions, which can accurately locate the common object in a set of unlabeled images, i.e., object co-localization.
- Key Instance Detection in Multi-Instance Learning107
This paper forms the Key Instance Detection (KID) problem, and proposes a voting framework (VF) solution to KID, represented by a citer kNN graph, which outperforms state-of-the-art MIL approaches on the performance of bag label prediction.
- Deep Bimodal Regression for Apparent Personality Analysis101
The Deep Bimodal Regression framework is proposed and a solution for the Apparent Personality Analysis competition track in the ChaLearn Looking at People challenge in association with ECCV 2016 is come up with.
- 98
- Research Advances on the Damage Mechanism of Skin Glycation and Related Inhibitors97
This article focuses on the damage mechanism of glucose and its glycation products on the skin by summarizing the biochemical characteristics, compositions, as well as processes of the production and elimination of AGEs.
- 97
- 95
- 95
- 87
- 85
- 85
- A unified pruning framework for vision transformers77
UP-ViTs achieve state-of-the-art results when pruning various ViT backbones and also outperform original ViTs, and is extended into NLP tasks and obtained more efficient transformer models.
- Deep Discriminative CNN with Temporal Ensembling for Ambiguously-Labeled Image Classification77
This paper innovatively employs the deep convolutional neural networks for ambiguously-labeled image classification, in which the well-known ResNet is adopted as the authors' backbone, and designs an entropy-based regularizer to enhance the discrimination ability.
- Webly Supervised Fine-Grained Recognition: Benchmark Datasets and An Approach76
Comprehensive experimental results and analyses on two new benchmark datasets demonstrate that the proposed method achieves superior performance over the competing baseline models and states-of-the-art.
- 75
- Compressing Transformers: Features Are Low-Rank, but Weights Are Not!74
AAFM is proposed, which adaptively determines the compressed model structure and locally compresses each linear layer's output features rather than the model weights, and is successfully applied to the language modeling task in NLP.
- 72
- 70
- Distilling Knowledge by Mimicking Features67
This paper argues that it is more advantageous to make the student mimic the teacher’s features in the penultimate layer of knowledge distillation, and proposes a loss term based on locality-sensitive hashing (LSH) that mimics feature directions more accurately, relaxes constraints on feature magnitudes, and achieves state-of-the-art distillation accuracy.
- Training Vision Transformers with only 2040 Images65
This paper investigates how to train ViTs with limited data and gives theoretical analyses that the method (based on parametric instance discrimination) is superior to other methods in that it can capture both feature alignment and instance similarities.
- A Diffusion and Clustering-Based Approach for Finding Coherent Motions and Understanding Crowd Scenes63
This paper addresses the problem of detecting coherent motions in crowd scenes and presents its two applications in crowd scene understanding: semantic region detection and recurrent activity mining and introduces a cluster-and-merge process.
- Learning Correspondence Structures for Person Re-Identification62
This paper addresses the problem of handling spatial misalignments due to camera-view changes or human-pose variations in person re-identification by introducing a boosting-based approach to learn a correspondence structure, which indicates the patchwise matching probabilities between images from a target camera pair and introduces a global constraint-based matching process.
- A Heat-Map-Based Algorithm for Recognizing Group Activities in Videos62
A new heat-map-based algorithm that first models human trajectories as series of heat sources and then applies a thermal diffusion process to create a heat map (HM) for representing the group activities for activity recognition is proposed.
- A Tube-and-Droplet-Based Approach for Representing and Analyzing Motion Trajectories59
This paper addresses the problem of representing motion trajectories in a highly informative way, and utilizes the droplet-based process for analyzing trajectories and applies it to trajectory analysis applications including trajectory clustering, trajectory classification & abnormality detection, and 3D action recognition.
- Linear Asymmetric Classifier for cascade detectors56
Experimental results on face detection show that LAC can improve the detection performance in comparison to standard methods, and it is shown that Fisher Discriminant Analysis on the features selected by AdaBoost yields better performance than AdaBoost itself.
- 55
- A Fast Dual Method for HIK SVM Learning54
ICD is proposed, a deterministic and highly scalable dual space HIK SVM solver that achieves high accuracies using its default parameters in many datasets and empirically shows that ICD is not sensitive to the C parameter in SVM.
- Corrigendum to “Ensembling neural networks: Many could be better than all” [Artificial Intelligence 137 (1–2) (2002) 239–263]53
The main contributions of the paper—the subset search strategy (GASEN) introduced in Section 3 after Eqs.
- 52
- 51
- Finding Coherent Motions and Semantic Regions in Crowd Scenes: A Diffusion and Clustering Approach48
A coarse-to-fine thermal-diffusion-based approach to detect coherent motions in crowd scenes and subsequently constructing semantic regions for activity recognition, and introduces a two-step clustering process to construct stable semantic regions from the extracted time-varying coherent motions.
- Coarse-to-Fine: A RNN-Based Hierarchical Attention Model for Vehicle Re-identification47
This work proposes an end-to-end RNN-based Hierarchical Attention (RNN-HA) classification model for vehicle re-identification and demonstrates that the proposed model achieves superior performance over state-of-the-art methods.
- 42
- Efficient and Effective Visual Codebook Generation Using Additive Kernels42
It is demonstrated that HIK can be used in an unsupervised manner to significantly improve the generation of visual codebooks and a one-class SVM formulation to create more effective visual code words is proposed.
- 41
- Deep Descriptor Transforming for Image Co-Localization39
This paper proposes a simple but effective method, named Deep Descriptor Transforming (DDT), for evaluating the correlations of descriptors and then obtaining the category-consistent regions, which can accurately locate the common object in a set of images.
- 39
- Improving CNN linear layers with power mean non-linearity38
This paper investigates the power mean function, which is proved effective and efficient in SVM kernel learning, and proposes Power Mean Transformation, which nonlinearizes both convolution and FC layers and can improve classification accuracy, bring generalization ability and add different non-linearity to CNN models.
- Action Recognition With Coarse-to-Fine Deep Feature Integration and Asynchronous Fusion38
This paper introduces a coarse-to-fine network which extracts shared deep features at different action class granularities and progressively integrates them to obtain a more accurate feature representation for input actions, and introduces an asynchronous fusion network which fuses information from different streams by asynchronously integrating stream-wise features in different time points.
- Adaptive Feeding: Achieving Fast and Accurate Detections by Adaptively Combining Object Detectors38
Experimental results on the PASCAL VOC, MS COCO and Caltech Pedestrian datasets confirm that AF has the ability to achieve comparable speed as the fast detector and comparable accuracy as the accurate one at the same time.
- A Random CNN Sees Objects: One Inductive Bias of CNN and Its Applications36
Experimental results show that the proposed Tobias significantly improves downstream tasks, especially for object detection, and has consistent improvements on training sets of different sizes, and is more resilient to changes in image augmentations.
- Repetitive Reprediction Deep Decipher for Semi-Supervised Learning34
A principled end-to-end framework named deep decipher (D2) for SSL is proposed, it is proved that pseudo-labels are related to network predictions by an exponential link function, which gives a theoretical support for using predictions as pseudo-Labels and it is demonstrated that updating pseudo- Labels by network predictions will make them uncertain.
- 33
- 32
- COMBINING REGRESSION ESTIMATORS: GA-BASED SELECTIVE NEURAL NETWORK ENSEMBLE32
The relationship between the generalization ability of the neural network ensemble and the correlation of the individual neural networks constituting the ensemble is analyzed in the context of combining neural regression estimators, which reveals that ensembling a selective subset of trained networks is superior to ensembled all the trained networks in some cases.
- 31
- 31
- Quantized Feature Distillation for Network Quantization29
Quantitative results show that QFD is more flexible and effective than previous quantization methods, and surpasses existing methods by a noticeable margin on not only image classification but also object detection, albeit being much simpler.
- Multi-Label Self-Supervised Learning with Scene Images29
Visualizations qualitatively show that clearly the pseudo-labels by MLS can automatically find semantically similar pseudo-positive pairs across different images to facilitate contrastive learning.
- A New Network-Based Algorithm for Human Activity Recognition in Videos28
A new network-transmission-based (NTB) algorithm is proposed for human activity recognition in videos that model people in the scene as packages, while human activities can be modeled as the process of package transmission in the network.
- Salvage of Supervision in Weakly Supervised Object Detection27
This paper proposes new approaches to utilize these weak and noisy signals effectively, and shows that each type of supervisory signal brings in notable improvements, and outperforms existing WSOD methods (which mainly use only the weak labels) by large margins.
- 27
- No One Left Behind: Improving the Worst Categories in Long-Tailed Learning26
By simply retraining the classifier of an existing pretrained model with the proposed loss function and using an optional ensemble trick that combines the predictions of the two classifiers, this work achieves a more uniform distribution of recall values across categories, which leads to a higher harmonic mean accuracy while the (arithmetic) average accuracy is still high.
- Optimizing Ranking Measures for Compact Binary Code Learning26
This work presents a general framework that allows one to directly optimize multivariate performance measures such as the AUC and NDCG and demonstrates the generality of StructHash by applying it to ranking prediction and image retrieval, and shows that it outperforms a few state-of-the-art hashing methods.
- 25
- 24
- DOA estimation for ULA by spectral Capon rooting method24
A direction of arrival (DOA) estimation method based on the spectral Capon rooting algorithm that realises DOA estimation by real polynomial rooting technique rather than a grid search and provides a lower computation complexity and a better performance compared to the spectralCapon searching algorithm.
- Mixup Without Hesitation23
It is found that mixup constantly explores the representation space, and inspired by the exploration-exploitation dilemma in reinforcement learning, this paper proposes mixup Without hesitation (mWh), a concise, effective, and easy-to-use training algorithm.
- A clinical study of the coronal plane deformity in Parkinson disease23
The present study indicated that the postural deformities in the coronal plane were related to the morphology of the spinal level, especially the position of the Cobb angle, which should be performed as early as possible to benefit the Parkinson disease patients with PS.
- 22
- Facial expression cloning with elastic and muscle models22
A novel algorithm is proposed for facial expression cloning that introduces a new elastic model to balance the global and local warping effects, such that the impacts from facial feature diversity among people can be minimized, and thus more effective geometric warping results can be achieved.
- ECML: An Ensemble Cascade Metric-Learning Mechanism Toward Face Verification21
The experimental results demonstrate that EC-RMML is superior to state-of-the-art metric-learning methods for face verification, and the proposed ECML mechanism is also applicable to other metric- learning approaches.
- Ensemble of GA based Selective Neural Network Ensembles21
E-GASEN, a two-layer neural network ensemble architecture is proposed, in which the base learners of the final ensemble are also ensembles, and experimental results show that e-GasEN generalizes better than a popular ensemble method.
- Practical Network Acceleration with Tiny Sets20
It is revealed that dropping blocks is a fundamentally superior approach in this scenario and enjoys a higher acceleration ratio and results in a better latency-accuracy performance under the few-shot setting, and a new concept namely recoverability to measure the difficulty of recovering the compressed network.
- 20
- CodeAttention: translating source code to comments by exploiting the code constructs19
A new attention mechanism called CodeAttention is proposed to translate code to comments, which is able to utilize the code constructs, such as critical statements, symbols and keywords, which could understand the semantic meanings of code better than previous methods.
- Compressing Models with Few Samples: Mimicking then Replacing17
This paper proposes a new framework named Mimicking then Replacing (MiR) for few-sample compression, which firstly urges the pruned model to output the same features as the teacher's in the penultimate layer, and then replaces teacher's layers before penultimate with a well-tuned compact one.
- A New Network-Based Algorithm for Human Group Activity Recognition in Videos17
A new network-transmission-based (NTB) algorithm is proposed for human activity recognition in videos that model people in the scene as packages, while human activities can be modeled as the process of package transmission in the network.
- 17
- 15
- Probabilistic classifiers with a generalized Gaussian scale mixture prior15
A flexible probabilistic model using a generalized Gaussian scale mixture (GGSM) prior that can provide an appropriate degree of sparsity for its model parameters, and yield either sparse or non-sparse estimates according to the intrinsic sparsity of features in a dataset is proposed.
- 14
- 14
- Explorating the mechanism of Huangqin Tang against skin lipid accumulation through network pharmacology and experimental validation13
Investigating the mechanisms of HQT in the treatment of skin lipid accumulation by network pharmacology and validating the results via in vitro experiments disclosed that HQT ameliorates lipogenesis in PA-induced SZ95 sebocytes partially through the AMPK signaling pathway.
- Can Partial Strong Labels Boost Multi-label Object Recognition?13
A multi-view multi-instance framework that transforms the multi-label classification problem into a multi-class multi- instance learning problem by extracting object proposals from images and holds the generalization ability to boost the performance of unseen categories by available strong labels.
- A new heat-map-based algorithm for human group activity recognition13
A new heat-map-based (HMB) algorithm is proposed for human group activity recognition that first models people trajectories as series of "heat sources" and applies a thermal diffusion process to create a heat map (HM) for representing the group activities.
- Advances in Multimedia Information Processing – PCM 201213
This paper introduces an incremental learning approach to video-based face recognition, which efficiently exploits the spatiotemporal information in videos and achieves a 100% recognition rate performed on the Honda/UCSD database, and gives near realtime feedback.
- 13
- Point Calculation Detection System Based on In Suit Convolution Transistor for Clinical Diagnosis of Heart Disease12
A convolution organic transistor (COT) is proposed for monitoring the ECG with CNNs that can complete the convolution calculation of ECG signals ≈20 000 times per second in theory and reduce the number of devices by 83% compared to conventional arrays.
- Weakly supervised foreground learning for weakly supervised localization and detection12
This paper motivate and propose the weakly supervised foreground learning (WSFL) task by showing that both WSOL and WSOD can be greatly improved if groundtruth foreground masks are available, and proposes a complete WSFL pipeline with low computational cost, which generates pseudo boxes, learns foreground masks, and does not need any localization annotations.
- 12
- Practical large scale classification with additive kernels12
This paper proposes the PmSVM-LUT algorithm that employs Look-Up Tables to boost the training and testing speed and save memory usage of additive kernel SVM classication, in order to meet the needs of large scale problems.
- Memristor-based in-situ convolutional strategy for accurate braille recognition11
A memristor-based in-situ convolutional strategy that uses the dynamic changes in the conductive wire, doping area, and polarization area of memristors as the process of convolutional operations, and uses the time required for conductance switching of a single device as the computation result, embodying convolutional computation through the unique spiked digital signal of the memristor.
- 11
- Sunrise or Sunset: Selective Comparison Learning for Subtle Attribute Recognition11
This paper proposes a new pairwise learning strategy to learn features from selective pairs of images that surpasses baseline methods by a large margin and achieves better results even compared with humans.
- 11
- Visual place categorization11
The VPC system achieves promising results for this challenging problem, especially for important categories such as bedroom, bathroom, and kitchen, and the proposed techniques achieved higher accuracies than competing descriptors and visual codebook generation methods.
- 11
- 10
- 9
- 9
- Structured Learning of Binary Codes with Column Generation for Optimizing Ranking Measures9
This work proposes a column generation based binary code learning framework for data-dependent hash function learning and demonstrates the generality of the method by applying it to ranking prediction and image retrieval, and shows that it outperforms several state-of-the-art hashing methods.
- 9
- 8
- DiffuLT: How to Make Diffusion Model Useful for Long-tail Recognition8
This paper proposes a new pipeline for long-tail (LT) recognition that utilizes the long-tailed dataset itself to generate a balanced proxy that can be optimized through cross-entropy (CE), and employs a randomly initialized diffusion model, trained exclusively on the long-tailed dataset, to synthesize new samples for underrepresented classes.
- 8
- Simultaneous 3D hand detection and pose estimation using single depth images8
This paper proposed to simultaneously detect the hand location and estimate its 3D pose in a multi-task learning framework using 3D region proposal, which searches possible hand locations in the 3D space.
- 8
- 8
- 7
- 7
- Neural random subspace7
Neural Random Subspace is proposed, a novel deep learning based random subspace method that enjoys the benefits of end-to-end, data-driven representation learning, as well as pervasive support from deep learning software and hardware platforms, hence achieving faster inference speed and higher accuracy.
- Sparse multiple instance learning as document classification7
The proposed methods achieve significantly higher accuracies and AUC than the state-of-the-art in a large number of sparse MIL problems, and the document classification analogy explains their efficacy in sparse Mil problems.
- 7
- 6
- Resource-constrained deep learning: challenges and practices6
This study first analyzes deep learnings thirst for various types of resources and the challenges they lead to and thereafter briefly introduces research progress from three aspects: data, label, and computation resources.
- Representing Sets of Instances for Visual Recognition6
The proposed D3 method effectively compares two sets as two distributions, and proposes a directional total variation distance (DTVD) to measure their dissimilarity.
- 6
- 6
- Neuromorphic parallel computing hardware based on quantum dots for 12-lead electrocardiogram monitoring5
This study proposes a neuromorphic parallel computing hardware architecture based on quantum dot synaptic transistors that enables 12-lead ECG monitoring within a single array circuit, eliminating the need for twelve separate circuits.
- QwT-v2: Practical, Effective and Efficient Post-Training Quantization5
This paper proposes QwT-v2, which not only enjoys all advantages of but also resolves major defects of QwT, and introduces significantly less extra parameters and computations compared to QwT, and at the same time matches or even outperforms QwT in accuracy.
- 5
- Supervised Contrastive Representation Embedding Based on Transformer for Few-Shot Classification5
This work employs Swin Transformer as the backbone to replace CNN architecture in order to explore the huge potential of transformer-based backbone for the field of few-shot learning and introduces supervised contrastive loss to meta learning to take good advantage of extremely limited relations for the first time.
- In Defense of Feature Mimicking for Knowledge Distillation.5
This paper argues that it is more advantageous to make the student mimic the teacher's features in the penultimate layer of knowledge distillation, and proposes a loss term based on locality-sensitive hashing (LSH) that mimics feature directions more accurately, relaxes constraints on feature magnitudes, and achieves state-of-the-art distillation accuracy.
- 5
- 5
- 5
- Balance Support Vector Machines Locally Using the Structural Similarity Kernel5
Experimental results on several benchmark datasets show that structural similarity can help the linear and the histogram intersection kernel to match or surpass the performance of the RBF kernel in SVM learning, and can significantly improve imbalanced learning results.
- 5
- Practical Network Acceleration With Tiny Sets: Hypothesis, Theory, and Algorithm4
This paper proposes the finetune convexity hypothesis to explain why recent few-shot compression algorithms do not suffer from overfitting problems, and claims dropping blocks is a fundamentally superior few-shot compression scheme in terms of more convex optimization and a higher acceleration ratio.
- Editorial of the Special Issue on Multi-instance Learning in Pattern Recognition and Vision4
This special issue aims at presenting recent advancements in both algorithmic and application-driven research related to multiinstance learning, and accepted 5 papers after a strict review process.
- Training data selection method for adaptive beamforming4
A method for adaptively selecting training data is proposed to improve the performance of adaptive beamforming using the sparse iterative covariance-based estimation technique, and the homogeneity of training samples can be improved significantly.
- Fast realization of root MUSIC using multi-taper real polynomial rooting4
An efficient implementation of root MUSIC using the multi-taper rooting technique is developed, which decomposes the (2N-2)-order complex polynomial as 2N-1 groups of low order real polynomials by the multi thetaper fast Fourier transform (FFT).
- Representing And Recognizing Motion Trajectories4
This paper first proposes to derive scene-related equipotential lines for points in a motion trajectory and concatenate them to construct a 3D tube for representing the trajectory, and a droplet-based method is further proposed which derives a "water droplet" from the3D tube and recognizes trajectory activities accordingly.
- Multi-class Cosegmentation with Pairwise Active Learning4
A novel dimensionality reduction method is proposed to produce an affinity matrix which reflects both the intrinsic structure of data and the constraints, and correlation clustering is applied on the newly generated affinity matrix to acquire refined results.
- Object Templates for Visual Place Categorization4
This paper proposes a local objects classifier that can automatically and effectively select key local objects of a semantic category from randomly sampled patches by the structural similarity support vector machine; and further classify the test frames with the Local Naive Bayes Nearest Neighbors algorithm.
- 4
- 4
- Minimal Interaction Separated Tuning: A New Paradigm for Visual Adaptation3
Minimal Interaction Separated Tuning, or MIST, is proposed, which reveals that the sum of intermediate features from pretrained models not only has minimal information transfer but also has high adaptation capability.
- 3
- 3
- 3
- 3
- 2
- 2
- 2
- 2
- Image categorization with resource constraints: introduction, challenges and advances2
This paper provides a comprehensive and in-depth introduction of recent developments of the research in image categorization with resource constraints, and makes an investigation into the recently developments of deep neural networks, with a focus on resource constrained deep nets.
- 2
- 2
- Exclusive Visual Descriptor Quantization2
It is shown that ExVQ is much faster than state-of-the-art ANN methods in BOV models while maintaining almost the same classification accuracy, and even with the VQ error rate as high as 30%, the classification accuracy of some ANN methods is similar to that of exhaustive search.
- 2
- 2
- 1
- 1
- 1
- 1
- 1
- Unified Low-rank Compression Framework for Click-through Rate Prediction1
A unified low-rank compression framework that can be applied to embedding tables and MLP layers in various CTR prediction models and finds that even with the most classic matrix decomposition SVD method, this framework can achieve better performance than the original model.
- Random Subspace Sampling for Classification with Missing Data1
This paper proposes a random subspace sampling method, RSS, by sampling missing items from the corresponding feature histogram distributions in random subspaces, which is effective and efficient at different levels of missing data.
- On Improving the Algorithm-, Model-, and Data- Efficiency of Self-Supervised Learning1
This paper proposes an efficient single-branch SSL method based on non-parametric instance discrimination based on a novel self-distillation loss that minimizes the KL divergence between the probability distribution and its square root version.
- 1
- Anesthesia and Intubation of Preadolescent Mouse Pups for Cardiothoracic Surgery1
D dosage regimens of ketamine/xylazine/atropine in 10-day-old C57BL/6J mouse pups that allow endotracheal intubation, while minimizing animal mortality are presented and cardiothoracic surgery in 10 day-old pups be restricted to pups weighing at least 5.5 g.
- 1
- 1
- 1
- 1
- Distance Metrics and Data Transformations1
Distance metrics and similarity measures, the mean as a similarity measure, and power mean kernel: a review of the literature and some examples.
- 1
- Overview of a Pattern Recognition System1
This chapter introduces a few components that are common to most (if not all) pattern recognition or machine learning systems, including a few key concepts and building blocks, and uses the nearest neighbor classifier as a simple solution to the face recognition problem.
- 1
- 1
- Learning with Augmented Multi-Instance View1
This paper proposes the Augmented Multi-Instance View (AMIV) framework to construct a better model by exploiting augmented information for abstract screening tasks, and proposes the AMIV-lss approach by establishing a latent semantic subspace between the two views.
- 1
- 1
- 1
- 1
- 1
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Research on practice teaching system of embedded system based on the cultivation of innovative and entrepreneurial ability–
It is necessary to strengthen the analysis of the practice teaching system of embedded system based on the cultivation of innovative and entrepreneurial ability in order to ensure the teaching effect of embedded system.
- –
- Diffusion Product Quantization–
This work proposes a method to compress the codebook by evaluating the importance of each vector and removing redundancy, ensuring the model size remaining within the desired range, and applies this method to the DiT model on ImageNet, demonstrating competitive generative performance.
- –
- –
- –
- –
- Video Static Foreground Detection Method for Desktop Robotic Arm–
Compared with the traditional foreground detection method, this paper presents the 3-frames threshold method to solve the problem of multi-texture background environment appears many noise and the difference degree to describe pixels is presented.
- –
- The Basic Idea behind Expectation-Maximization–
The general EM derivation, the E& M-steps, and the EM algorithm: will EM converge?
- –
- –
- –
- –
- –
- –
- Sparse and Misaligned Data–
In the learning and recognition methods introduced till now, many types of data in the real world exhibit strong characteristics that cannot be ignored, and some data properties are adversarial—they will be helpful to obtain better representation and accuracy if they are properly utilized.
- –
- –
- –
- –
- –
- –
- –
- Neural Forest Learning.–
Neural Forest Learning (NFL), a novel deep learning based random-forest-like method, enjoys the benefits of end-to-end, data-driven representation learning, as well as pervasive support from deep learning software and hardware platforms, hence achieving faster inference speed and higher accuracy than previous forest methods.
- –
- Content-Based Image Recovery–
A multi-scale architecture and a spatial pooling layer are presented on the base of recent deep convolutional neural networks (CNNs) to overcome the problem of recovering from aspect-ratio distorted images based on their contents.
- –
- Investigation of Motion Balance of Children with Spastic Diplegia–
Overall, in order to fulfill the balance control, strategies of foot valgus, larger deviation in medial-lateral (ML) direction, longer contact time and higher velocity in AP and ML directions were adopted by CP children.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Towards the Optimal Training of Cascades of Boosted Ensembles–
A probabilistic model for cascade performance and its use in a fully-automated training algorithm is described.
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-14. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.