An improved residual network model for image recognition using a combination of snapshot ensembles and the cutout technique

Termritthikun, Chakkrit; Jamtsho, Yeshi; Muneesawang, Paisarn

doi:10.1007/s11042-019-08332-3

Cited by 12 publications

(6 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our system only needs to update the distribution of labels in corresponding environment, which is easier than the re-training or transferring of models. We performed our system and existing methods on the samples of CIFAR-10 [12][13][14], CIFAR-100 [15][16][17] and Mini-ImageNet [18][19][20]. All of these evaluations proved the effectiveness of our system.…”

Section: Figure 1: How the Distribution Of Labels Increases The Accur...mentioning

confidence: 83%

See 1 more Smart Citation

Environment Adaptive Deep Learning Classification System Based on One-shot Guidance

Jin¹,

Pei²,

Zhao³

et al. 2022

Computers, Materials &Amp; Continua

View full text Add to dashboard Cite

When utilizing the deep learning models in some real applications, the distribution of the labels in the environment can be used to increase the accuracy. Generally, to compute this distribution, there should be the validation set that is labeled by the ground truths. On the other side, the dependency of ground truths limits the utilization of the distribution in various environments. In this paper, we carried out a novel system for the deep learning-based classification to solve this problem. Firstly, our system only uses one validation set with ground truths to compute some hyper parameters, which is named as one-shot guidance. Secondly, in an environment, our system builds the validation set and labels this by the prediction results, which does not need any guidance by the ground truths. Thirdly, the computed distribution of labels by the validation set selectively cooperates with the probability of labels by the output of models, which is to increase the accuracy of predict results on testing samples. We selected six popular deep learning models on three real datasets for the evaluation. The experimental results show that our system can achieve higher accuracy than state-of-art methods while reducing the dependency of labeled validation set.

show abstract

Section: Figure 1: How the Distribution Of Labels Increases The Accur...mentioning

confidence: 83%

“…CIFAR-100 has 100 classes and each class contains 600 images [15][16][17]. We separate each class of the dataset to 500 training samples and 100 testing ones.…”

Section: Random Case On Cifar-100 and Mini-imagenetmentioning

confidence: 99%

Environment Adaptive Deep Learning Classification System Based on One-shot Guidance

Jin¹,

Pei²,

Zhao³

et al. 2022

Computers, Materials &Amp; Continua

View full text Add to dashboard Cite

show abstract

“…CIFAR-100 is similar to the CIFAR-10. This set has 100 classes and each class has 600 images [14][15][16]. 50000 training samples are to train the models.…”

Section: The Evaluation On Cifar-10 and Cifar-100mentioning

confidence: 99%

“…By using fusion method, the classification accuracy may be increased. We evaluated the methods on some real datasets that include CIFAR-10 [11-13] and CIFAR-100 [14][15][16]. Furthermore, we also collected a real dataset where the samples have different resolution and each label has different number of samples.…”

Section: Introductionmentioning

confidence: 99%

Solving the Feature Diversity Problem Based on Multi-Model Scheme

Jin¹,

Zhao²,

Pei³

et al. 2021

Journal on Artificial Intelligence

View full text Add to dashboard Cite

Generally, the performance of deep learning models is related to the captured features of training samples. When the training samples belong to different domains, the diverse features may increase the difficulty of training high performance models. In this paper, we built a new framework that generates multiple models on the organized samples to increase the accuracy of classification. Firstly, our framework selects some existing models and trains each of them on organized training sets to get multiple trained models. Secondly, we select some of them based on a validation set. Finally, we use some fusion method on the outputs of the selected models to get more accurate results. The experimental results show that our framework achieved higher accuracy than the existing methods. Our framework can be an option for the deep learning system to increase the classification accuracy.

show abstract

“…Model T-IM SENNs [18] 78.5% ResNet-34 [29] 52.0% DE-CapsNet [30] 93.0% VGG 16 [31] 52.2% VGG-19 [32] 93.4% VGG 16 + aug [31] 56.4% ResNet-110 [33] 93.6% IRRCNN [34] 52.2% DenseNet [35] 94.8% ResNet-110 [36] 56.6% WRN-28-2 [37] 94.9% WRN-40-20 [38] 63.8% 2) and ( 5)). Further, E(x) denotes that a learnt embedding was used as input to the model (see sec.…”

Section: Model C10mentioning

confidence: 99%

Optimising for Interpretability: Convolutional Dynamic Alignment Networks

Böhle¹,

Fritz²,

Schiele³

2021

Preprint

View full text Add to dashboard Cite

We introduce a new family of neural network models called Convolutional Dynamic Alignment Networks (CoDA Nets), which are performant classifiers with a high degree of inherent interpretability. Their core building blocks are Dynamic Alignment Units (DAUs), which are optimised to transform their inputs with dynamically computed weight vectors that align with task-relevant patterns. As a result, CoDA Nets model the classification prediction through a series of input-dependent linear transformations, allowing for linear decomposition of the output into individual input contributions. Given the alignment of the DAUs, the resulting contribution maps align with discriminative input patterns. These model-inherent decompositions are of high visual quality and outperform existing attribution methods under quantitative metrics. Further, CoDA Nets constitute performant classifiers, achieving on par results to ResNet and VGG models on e.g. CIFAR-10 and TinyImagenet. Lastly, CoDA Nets can be combined with conventional neural network models to yield powerful classifiers that more easily scale to complex datasets such as Imagenet whilst exhibiting an increased interpretable depth, i.e., the output can be explained well in terms of contributions from intermediate layers within the network.

show abstract

An improved residual network model for image recognition using a combination of snapshot ensembles and the cutout technique

Cited by 12 publications

References 20 publications

Environment Adaptive Deep Learning Classification System Based on One-shot Guidance

Environment Adaptive Deep Learning Classification System Based on One-shot Guidance

Solving the Feature Diversity Problem Based on Multi-Model Scheme

Optimising for Interpretability: Convolutional Dynamic Alignment Networks

Contact Info

Product

Resources

About