Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909666
Siwei Zhang, Kimon Cokona, R. Pöhlmann, E. Staudinger, T. Wiedemann, A. Dammann
Swarm robotics has gained an increasing attention in applications like extraterrestrial exploration and disaster management, due to the ability of simultaneously observing at different locations and avoiding a single point of failure. In order to operate autonomously, robots in a swarm need to know their precise poses, including their positions, velocities and orientations. When external navigation infrastructures like the global navigation satellite systems (GNSS) are not ubiquitously accessible, the swarm of robots need to rely on internal measurements to estimate their poses. In this paper, we propose a cooperative 3D pose estimation framework, based on the insights of sensor characteristics that we gained from outdoor swarm navigation experiments. A decentralized particle filter (DPF) operates on each robot to estimate its pose via fusing radio-based ranging, inertial sensor data, control commands and the pose estimates of its neighbors. This framework is integrated in the swarm navigation ecosystem developed at the German Aerospace Center (DLR), and is unified for both simulations and experiments.
{"title":"Cooperative Pose Estimation in a Robotic Swarm: Framework, Simulation and Experimental Results","authors":"Siwei Zhang, Kimon Cokona, R. Pöhlmann, E. Staudinger, T. Wiedemann, A. Dammann","doi":"10.23919/eusipco55093.2022.9909666","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909666","url":null,"abstract":"Swarm robotics has gained an increasing attention in applications like extraterrestrial exploration and disaster management, due to the ability of simultaneously observing at different locations and avoiding a single point of failure. In order to operate autonomously, robots in a swarm need to know their precise poses, including their positions, velocities and orientations. When external navigation infrastructures like the global navigation satellite systems (GNSS) are not ubiquitously accessible, the swarm of robots need to rely on internal measurements to estimate their poses. In this paper, we propose a cooperative 3D pose estimation framework, based on the insights of sensor characteristics that we gained from outdoor swarm navigation experiments. A decentralized particle filter (DPF) operates on each robot to estimate its pose via fusing radio-based ranging, inertial sensor data, control commands and the pose estimates of its neighbors. This framework is integrated in the swarm navigation ecosystem developed at the German Aerospace Center (DLR), and is unified for both simulations and experiments.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121998378","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909584
Xianrui Wang, J. Benesty, Gongping Huang, Jingdong Chen
Spectral estimation is of significant practical importance in a wide range of applications. This paper proposes a minimum variance distortionless response (MVDR) method for spectral estimation based on the Kronecker product. Taking advantage of the particular structure of the Fourier vector, we decompose it as a Kronecker product of two shorter vectors. Then, we design the spectral estimation filters under the same structure, i.e., as a Kronecker product of two filters. Consequently, the conventional MVDR spectrum problem is transformed to one of estimating two filters of much shorter lengths. Since it has much fewer parameters to estimate, the proposed method is able to achieve better performance than its conventional counterpart, particularly when the number of available signal samples is small. Also presented in this paper is the generalization to the estimation of the cross-spectrum and coherence function.
{"title":"A Minimum Variance Distortionless Response Spectral Estimator with Kronecker Product Filters","authors":"Xianrui Wang, J. Benesty, Gongping Huang, Jingdong Chen","doi":"10.23919/eusipco55093.2022.9909584","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909584","url":null,"abstract":"Spectral estimation is of significant practical importance in a wide range of applications. This paper proposes a minimum variance distortionless response (MVDR) method for spectral estimation based on the Kronecker product. Taking advantage of the particular structure of the Fourier vector, we decompose it as a Kronecker product of two shorter vectors. Then, we design the spectral estimation filters under the same structure, i.e., as a Kronecker product of two filters. Consequently, the conventional MVDR spectrum problem is transformed to one of estimating two filters of much shorter lengths. Since it has much fewer parameters to estimate, the proposed method is able to achieve better performance than its conventional counterpart, particularly when the number of available signal samples is small. Also presented in this paper is the generalization to the estimation of the cross-spectrum and coherence function.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122629216","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909828
Albert Podusenko, B. V. Erp, Dmitry V. Bagaev, Ismail Senöz, B. Vries
The switching autoregressive model is a flexible model for signals generated by non-stationary processes. Unfortunately, evaluation of the exact posterior distributions of the latent variables for a switching autoregressive model is analytically intractable, and this limits the applicability of switching autoregressive models in practical signal processing tasks. In this paper we present a message passing-based approach for computing approximate posterior distributions in the switching autoregressive model. Our solution tracks approximate posterior distributions in a modular way and easily extends to more complicated model variations. The proposed message passing algorithm is verified and validated on synthetic and acoustic data sets respectively.
{"title":"Message Passing-based Inference in Switching Autoregressive Models","authors":"Albert Podusenko, B. V. Erp, Dmitry V. Bagaev, Ismail Senöz, B. Vries","doi":"10.23919/eusipco55093.2022.9909828","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909828","url":null,"abstract":"The switching autoregressive model is a flexible model for signals generated by non-stationary processes. Unfortunately, evaluation of the exact posterior distributions of the latent variables for a switching autoregressive model is analytically intractable, and this limits the applicability of switching autoregressive models in practical signal processing tasks. In this paper we present a message passing-based approach for computing approximate posterior distributions in the switching autoregressive model. Our solution tracks approximate posterior distributions in a modular way and easily extends to more complicated model variations. The proposed message passing algorithm is verified and validated on synthetic and acoustic data sets respectively.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"124 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116303850","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909579
Aditya Raikar, Meet H. Soni, Ashish Panda, S. Kopparapu
Acoustic environment plays a major role in the performance of a large-scale Automatic Speech Recognition (ASR) system. It becomes a lot more challenging when substantial amount of distortions, such as background noise and reverberations are present. Of late, it has been standard to use i-vectors for Acoustic Model (AM) adaptation. Embeddings from Single Task Learned (STL) neural network systems, such as x-vectors and r-vectors, have also been used to a varying degree of success. This paper proposes the use of Multi Task Learned (MTL) embeddings for large vocabulary hybrid acoustic model adaptation in reverberant environments. MTL embeddings are extracted from an affine layer of the deep neural network trained on multiple tasks such as speaker information and room information. Our experiments show that the proposed MTL embeddings outperform i-vectors, x-vectors and r-vectors for AM adaptation in reverberant conditions. Besides, it has been demonstrated that the proposed MTL-embeddings can be fused with i-vectors to provide further improvement. We provide results on artificially reverberated Librispeech data as well as real world reverberated HRRE data. On Librispeech database, the proposed method provides an improvement of 10.9% and 8.7% relative to i-vector in reverberated test-clean and test-other data respectively, while an improvement of 7% is observed relative to i-vector when the proposed system is tested on HRRE dataset.
{"title":"Acoustic Model Adaptation In Reverberant Conditions Using Multi-task Learned Embeddings","authors":"Aditya Raikar, Meet H. Soni, Ashish Panda, S. Kopparapu","doi":"10.23919/eusipco55093.2022.9909579","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909579","url":null,"abstract":"Acoustic environment plays a major role in the performance of a large-scale Automatic Speech Recognition (ASR) system. It becomes a lot more challenging when substantial amount of distortions, such as background noise and reverberations are present. Of late, it has been standard to use i-vectors for Acoustic Model (AM) adaptation. Embeddings from Single Task Learned (STL) neural network systems, such as x-vectors and r-vectors, have also been used to a varying degree of success. This paper proposes the use of Multi Task Learned (MTL) embeddings for large vocabulary hybrid acoustic model adaptation in reverberant environments. MTL embeddings are extracted from an affine layer of the deep neural network trained on multiple tasks such as speaker information and room information. Our experiments show that the proposed MTL embeddings outperform i-vectors, x-vectors and r-vectors for AM adaptation in reverberant conditions. Besides, it has been demonstrated that the proposed MTL-embeddings can be fused with i-vectors to provide further improvement. We provide results on artificially reverberated Librispeech data as well as real world reverberated HRRE data. On Librispeech database, the proposed method provides an improvement of 10.9% and 8.7% relative to i-vector in reverberated test-clean and test-other data respectively, while an improvement of 7% is observed relative to i-vector when the proposed system is tested on HRRE dataset.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126653473","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909555
Emilie D'Olne, Vincent W. Neo, P. Naylor
As the number of connected devices equipped with multiple microphones increases, scientific interest in distributed microphone array processing grows. Current beamforming methods heavily rely on estimating quantities related to array geometry, which is extremely challenging in real, non-stationary environments. Recent work on polynomial eigenvalue decomposition (PEVD) has shown promising results for speech enhancement in singular arrays without requiring the estimation of any array-related parameter [1]. This work extends these results to the realm of distributed microphone arrays, and further presents a novel framework for speech enhancement in distributed microphone arrays using PEVD. The proposed approach is shown to almost always outperform optimum beamformers located at arrays closest to the desired speaker. Moreover, the proposed approach exhibits very strong robustness to steering vector errors.
{"title":"Speech Enhancement in Distributed Microphone Arrays Using Polynomial Eigenvalue Decomposition","authors":"Emilie D'Olne, Vincent W. Neo, P. Naylor","doi":"10.23919/eusipco55093.2022.9909555","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909555","url":null,"abstract":"As the number of connected devices equipped with multiple microphones increases, scientific interest in distributed microphone array processing grows. Current beamforming methods heavily rely on estimating quantities related to array geometry, which is extremely challenging in real, non-stationary environments. Recent work on polynomial eigenvalue decomposition (PEVD) has shown promising results for speech enhancement in singular arrays without requiring the estimation of any array-related parameter [1]. This work extends these results to the realm of distributed microphone arrays, and further presents a novel framework for speech enhancement in distributed microphone arrays using PEVD. The proposed approach is shown to almost always outperform optimum beamformers located at arrays closest to the desired speaker. Moreover, the proposed approach exhibits very strong robustness to steering vector errors.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124141726","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909776
Srdan Lazendic, Jens Janssens, Shaoguang Huang, A. Pižurica
Despite their huge potential, deep learning-based models are still not trustful enough to warrant their adoption in clinical practice. The research on the interpretability and explainability of deep learning is currently attracting huge attention. Multilayer Convolutional Sparse Coding (ML-CSC) data model, provides a model-based explanation of convolutional neural networks (CNNs). In this article, we extend the ML-CSC framework towards multimodal data for medical image segmentation, and propose a merged joint feature extraction ML-CSC model. This work generalizes and improves upon our previous model, by deriving a more elegant approach that merges feature extraction and convolutional sparse coding in a unified framework. A segmentation study on a multimodal magnetic resonance imaging (MRI) dataset confirms the effectiveness of the proposed approach. We also supply an interpretability study regarding the involved model parameters.
{"title":"On Interpretability of CNNs for Multimodal Medical Image Segmentation","authors":"Srdan Lazendic, Jens Janssens, Shaoguang Huang, A. Pižurica","doi":"10.23919/eusipco55093.2022.9909776","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909776","url":null,"abstract":"Despite their huge potential, deep learning-based models are still not trustful enough to warrant their adoption in clinical practice. The research on the interpretability and explainability of deep learning is currently attracting huge attention. Multilayer Convolutional Sparse Coding (ML-CSC) data model, provides a model-based explanation of convolutional neural networks (CNNs). In this article, we extend the ML-CSC framework towards multimodal data for medical image segmentation, and propose a merged joint feature extraction ML-CSC model. This work generalizes and improves upon our previous model, by deriving a more elegant approach that merges feature extraction and convolutional sparse coding in a unified framework. A segmentation study on a multimodal magnetic resonance imaging (MRI) dataset confirms the effectiveness of the proposed approach. We also supply an interpretability study regarding the involved model parameters.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126414184","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909949
Marzieh Ajirak, Yuhao Liu, P. Djurić
In this paper, we address the classification and dimensionality reduction via ensembles of Gaussian Process Latent Variable Models (GPLVMs). The underlying idea is to have a diverse representation of latent spaces represented by an ensemble of GPLVMs. Each GPLVM of the ensemble has its own projections of the high dimensional observed data on a low dimensional latent space. These models are weighted using importance sampling. Since in practical settings, neither the kernel of the GPLVM nor the dimension of the latent space is known, it is logical to engage an ensemble of GPLVMs based on different kernels and for each of them estimate the dimension of the lower dimensional space. We demonstrate the advantage of working with ensembles for classification and show the performance of dimensionality reduction of our method with numerical simulations.
{"title":"Ensembles of Gaussian process latent variable models","authors":"Marzieh Ajirak, Yuhao Liu, P. Djurić","doi":"10.23919/eusipco55093.2022.9909949","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909949","url":null,"abstract":"In this paper, we address the classification and dimensionality reduction via ensembles of Gaussian Process Latent Variable Models (GPLVMs). The underlying idea is to have a diverse representation of latent spaces represented by an ensemble of GPLVMs. Each GPLVM of the ensemble has its own projections of the high dimensional observed data on a low dimensional latent space. These models are weighted using importance sampling. Since in practical settings, neither the kernel of the GPLVM nor the dimension of the latent space is known, it is logical to engage an ensemble of GPLVMs based on different kernels and for each of them estimate the dimension of the lower dimensional space. We demonstrate the advantage of working with ensembles for classification and show the performance of dimensionality reduction of our method with numerical simulations.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124418968","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909610
Marcelo Fiori, Bernardo Marenco, Federico Larroca, P. Bermolen, G. Mateos
The Random Dot Product Graph (RDPG) is a popular generative graph model for relational data. RDPGs postulate there exist latent positions for each node, and specifies the edge formation probabilities via the inner product of the corresponding latent vectors. The embedding task of estimating these latent positions from observed graphs is usually posed as a non-convex matrix factorization problem. The workhorse Adjacency Spectral Embedding offers an approximate solution obtained via the eigendecomposition of the adjacency matrix, which enjoys solid statistical guarantees but can be computationally intensive and is formally solving a surrogate problem. In this paper, we bring to bear recent non-convex optimization advances and demonstrate their impact to RDPG inference. We develop first-order gradient descent methods to better solve the original optimization problem, and to accommodate broader network embedding applications in an organic way. The effectiveness of the resulting graph representation learning framework is demonstrated on both synthetic and real data. We show the algorithms are scalable, robust to missing network data, and can track the latent positions over time when the graphs are acquired in a streaming fashion.
{"title":"Algorithmic Advances for the Adjacency Spectral Embedding","authors":"Marcelo Fiori, Bernardo Marenco, Federico Larroca, P. Bermolen, G. Mateos","doi":"10.23919/eusipco55093.2022.9909610","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909610","url":null,"abstract":"The Random Dot Product Graph (RDPG) is a popular generative graph model for relational data. RDPGs postulate there exist latent positions for each node, and specifies the edge formation probabilities via the inner product of the corresponding latent vectors. The embedding task of estimating these latent positions from observed graphs is usually posed as a non-convex matrix factorization problem. The workhorse Adjacency Spectral Embedding offers an approximate solution obtained via the eigendecomposition of the adjacency matrix, which enjoys solid statistical guarantees but can be computationally intensive and is formally solving a surrogate problem. In this paper, we bring to bear recent non-convex optimization advances and demonstrate their impact to RDPG inference. We develop first-order gradient descent methods to better solve the original optimization problem, and to accommodate broader network embedding applications in an organic way. The effectiveness of the resulting graph representation learning framework is demonstrated on both synthetic and real data. We show the algorithms are scalable, robust to missing network data, and can track the latent positions over time when the graphs are acquired in a streaming fashion.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127981128","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-29DOI: 10.48550/arXiv.2209.11342
Emmanuel Martinez, Edwin Vargas, H. Arguello
The abundant spatial and angular information from light fields has allowed the development of multiple disparity estimation approaches. However, the acquisition of light fields requires high storage and processing cost, limiting the use of this technology in practical applications. To overcome these drawbacks, the compressive sensing (CS) theory has allowed the development of optical architectures to acquire a single coded light field measurement. This measurement is decoded using an optimization algorithm or deep neural network that requires high computational costs. The traditional approach for disparity estimation from compressed light fields requires first recovering the entire light field and then a post-processing step, thus requiring long times. In contrast, this work proposes a fast disparity estimation from a single compressed measurement by omitting the recovery step required in traditional approaches. Specifically, we propose to jointly optimize an optical architecture for acquiring a single coded light field snapshot and a convolutional neural network (CNN) for estimating the disparity maps. Experimentally, the proposed method estimates disparity maps comparable with those obtained from light fields reconstructed using deep learning approaches. Furthermore, the proposed method is 20 times faster in training and inference than the best method that estimates the disparity from reconstructed light fields.
{"title":"Fast Disparity Estimation from a Single Compressed Light Field Measurement","authors":"Emmanuel Martinez, Edwin Vargas, H. Arguello","doi":"10.48550/arXiv.2209.11342","DOIUrl":"https://doi.org/10.48550/arXiv.2209.11342","url":null,"abstract":"The abundant spatial and angular information from light fields has allowed the development of multiple disparity estimation approaches. However, the acquisition of light fields requires high storage and processing cost, limiting the use of this technology in practical applications. To overcome these drawbacks, the compressive sensing (CS) theory has allowed the development of optical architectures to acquire a single coded light field measurement. This measurement is decoded using an optimization algorithm or deep neural network that requires high computational costs. The traditional approach for disparity estimation from compressed light fields requires first recovering the entire light field and then a post-processing step, thus requiring long times. In contrast, this work proposes a fast disparity estimation from a single compressed measurement by omitting the recovery step required in traditional approaches. Specifically, we propose to jointly optimize an optical architecture for acquiring a single coded light field snapshot and a convolutional neural network (CNN) for estimating the disparity maps. Experimentally, the proposed method estimates disparity maps comparable with those obtained from light fields reconstructed using deep learning approaches. Furthermore, the proposed method is 20 times faster in training and inference than the best method that estimates the disparity from reconstructed light fields.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127465257","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909715
Christoph Brauer, D. Lorenz, Lionel Tondji
The incorporation of prior knowledge into the ma-chine learning pipeline is subject of informed machine learning. Spatial invariances constitute a class of prior knowledge that can be taken into account especially in the design of model architectures or through virtual training examples. In this contribution, we investigate fully connected neural network architectures that are equivariant with respect to the dihedral group of order eight. This is practically motivated by the application of leakage detection in vacuum bagging which plays an important role in the manufacturing of fiber composite components. Our approach for the derivation of an equivariant architecture is constructive and transferable to other symmetry groups. It starts from a standard network architecture and results in a specific kind of weight sharing in each layer. In numerical experiments, we compare equivariant and standard networks on a novel leakage detection dataset. Our results indicate that group equivariant networks can capture the application specific prior knowledge much better than standard networks, even if the latter are trained on augmented data.
{"title":"Group equivariant networks for leakage detection in vacuum bagging","authors":"Christoph Brauer, D. Lorenz, Lionel Tondji","doi":"10.23919/eusipco55093.2022.9909715","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909715","url":null,"abstract":"The incorporation of prior knowledge into the ma-chine learning pipeline is subject of informed machine learning. Spatial invariances constitute a class of prior knowledge that can be taken into account especially in the design of model architectures or through virtual training examples. In this contribution, we investigate fully connected neural network architectures that are equivariant with respect to the dihedral group of order eight. This is practically motivated by the application of leakage detection in vacuum bagging which plays an important role in the manufacturing of fiber composite components. Our approach for the derivation of an equivariant architecture is constructive and transferable to other symmetry groups. It starts from a standard network architecture and results in a specific kind of weight sharing in each layer. In numerical experiments, we compare equivariant and standard networks on a novel leakage detection dataset. Our results indicate that group equivariant networks can capture the application specific prior knowledge much better than standard networks, even if the latter are trained on augmented data.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130457233","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：