Harmonic Networks with Limited Training Samples

Uličný, Matej; Krylov, Vladimir A.; Dahyot, Rozenn

doi:10.23919/eusipco.2019.8902831

Cited by 13 publications

(12 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This enables efficient model compression by parameter truncation with only minor degradation in the model performance. This has been shown to be particularly useful for tasks with limited training samples [6].…”

Section: Discussionmentioning

confidence: 99%

“…Email: vladimir.krylov@dcu.ie particular basis functions [3], [4] have mostly focused on ability to compress the network, we show furthermore that prior information coming from well chosen filter basis can not only be used for compression but can also speed up training convergence and improve performance. This paper presents a comprehensive study based on our earlier works [5], [6] on harmonic blocks 1 . In this work we introduce further analysis and substantially expand the experimental validation to include edge and object detection cases that broader demonstrate the benefits of harmonic networks.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Harmonic Convolutional Networks based on Discrete Cosine Transform

Uličný¹,

Krylov²,

Dahyot³

2020

Preprint

Self Cite

View full text Add to dashboard Cite

Convolutional neural networks (CNNs) learn filters in order to capture local correlation patterns in feature space. In this paper we propose to revert to learning combinations of preset spectral filters by switching to CNNs with harmonic blocks. We rely on the use of the Discrete Cosine Transform (DCT) filters which have excellent energy compaction properties and are widely used for image compression. The proposed harmonic blocks rely on DCT-modeling and replace conventional convolutional layers to produce partially or fully harmonic versions of new or existing CNN architectures. We demonstrate how the harmonic networks can be efficiently compressed in a straightforward manner by truncating high-frequency information in harmonic blocks which is possible due to the redundancies in the spectral domain. We report extensive experimental validation demonstrating the benefits of the introduction of harmonic blocks into state-of-the-art CNN models in image classification, segmentation and edge detection applications.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Harmonic Convolutional Networks based on Discrete Cosine Transform

Uličný¹,

Krylov²,

Dahyot³

2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…The research community is actively working to solve this issue, proposing solutions that can be adopted on tasks that involve small data or low resources. Examples of such approaches are pre-trained models (e.g., transfer-learning [73]), unlabeled data (e.g., semisupervised learning [19]), custom training paradigms, or novel blocks for specific architectures [76]. The first category includes transfer learning techniques and few-shot learning.…”

Section: Image Classificationmentioning

confidence: 99%

A self-interpretable module for deep image classification on small data

2022

View full text Add to dashboard Cite

Deep neural networks are the driving force of the recent explosion of machine learning applications in everyday life. However, they usually require a lot of training data to work well, and they act as black-boxes, making predictions without any explanation about them. This paper presents Memory Wrap, a module (i.e, a set of layers) that can be added to deep learning models to improve their performance and interpretability in settings where few data are available. Memory Wrap adopts a sparse content-attention mechanism between the input and some memories of past training samples. We show that adding Memory Wrap to standard deep neural networks improves their performance when they learn from a limited set of data, and allows them to reach comparable performance when they learn from the full dataset. We discuss how the analysis of its structure and content-attention weights helps to get insights about its decision process and makes their predictions more interpretable, compared to the same networks without Memory Wrap. We test our approach on image classification tasks using several networks on three different datasets, namely CIFAR10, SVHN, and CINIC10.

show abstract

“…Ulicny and Dahyot [20] used a CNN to classify images in the DCT domain. Ulicny et al [21,22] designed the so-called harmonic blocks to replace the traditional convolutions. These blocks generate features by learning combinations of spectral filters defined by DCT.…”

Section: Related Workmentioning

confidence: 99%

Exploring Semantic Segmentation on the DCT Representation

Hang

2019

Proceedings of the ACM Multimedia Asia

View full text Add to dashboard Cite

Typical convolutional networks are trained and conducted on RGB images. However, images are often compressed for memory savings and efficient transmission in real-world applications. In this paper, we explore methods for performing semantic segmentation on the discrete cosine transform (DCT) representation defined by the JPEG standard. We first rearrange the DCT coefficients to form a preferred input type, then we tailor an existing network to the DCT inputs. The proposed method has an accuracy close to the RGB model at about the same network complexity. Moreover, we investigate the impact of selecting different DCT components on segmentation performance. With a proper selection, one can achieve the same level accuracy using only 36% of the DCT coefficients. We further show the robustness of our method under quantization errors. To our knowledge, this paper is the first to explore semantic segmentation on the DCT representation.

show abstract

Harmonic Networks with Limited Training Samples

Cited by 13 publications

References 24 publications

Harmonic Convolutional Networks based on Discrete Cosine Transform

Harmonic Convolutional Networks based on Discrete Cosine Transform

A self-interpretable module for deep image classification on small data

Exploring Semantic Segmentation on the DCT Representation

Contact Info

Product

Resources

About