Pub Date : 2022-08-29DOI: 10.23919/eusipco55093.2022.9909813
V. Obradović, Ilija Rajak, M. Secujski, V. Delić
Online courses have had exponential growth during COVID-19 pandemic, and video lectures are also important for lifelong learning. However, lecturers experience a number of challenges in creating video lectures, related to both speech recording (microphone and noise; diction, articulation and intonation) and video recording (camera and light; consistency in appearance). It is particularly difficult to modify and update recorded content. The paper presents a solution for these problems based on the application of artificial intelligence in creating virtual speakers based on TTS synthesis and Wav2Lip GAN trained on a custom data set. A pilot project which included the evaluation and testing of the developed system by dozens of teachers will be presented in detail. The use of TTS overcomes the problems in achieving speaker consistency by providing high quality speech in different languages, while the attention and motivation of students is improved by using animated virtual speakers.
{"title":"Text driven virtual speakers","authors":"V. Obradović, Ilija Rajak, M. Secujski, V. Delić","doi":"10.23919/eusipco55093.2022.9909813","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909813","url":null,"abstract":"Online courses have had exponential growth during COVID-19 pandemic, and video lectures are also important for lifelong learning. However, lecturers experience a number of challenges in creating video lectures, related to both speech recording (microphone and noise; diction, articulation and intonation) and video recording (camera and light; consistency in appearance). It is particularly difficult to modify and update recorded content. The paper presents a solution for these problems based on the application of artificial intelligence in creating virtual speakers based on TTS synthesis and Wav2Lip GAN trained on a custom data set. A pilot project which included the evaluation and testing of the developed system by dozens of teachers will be presented in detail. The use of TTS overcomes the problems in achieving speaker consistency by providing high quality speech in different languages, while the attention and motivation of students is improved by using animated virtual speakers.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"11 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":"133625867","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/eusipco55093.2022.9909777
Paschalis Tsiaflakis, W. Coomans
Reinforcement learning (RL) is a powerful machine learning technique to learn optimal actions in a control system setup. An important drawback of RL algorithms is the need for balancing exploitation vs exploration. Exploration corresponds to taking randomized actions with the aim to learn from it and make better decisions in the future. However, these exploratory actions result in poor performance, and current RL algorithms have a slow convergence as one can only learn from a single action outcome per iteration. We propose a novel concept of Inference-based RL that is applicable to a specific class of RL problems, and that allows to eliminate the performance impact caused by traditional exploration strategies, thereby making RL performance more consistent and greatly improving the convergence speed. The specific RL problem class is a problem class in which the observation of the outcome of one action can be used to infer the outcome of other actions, without the need to actually perform them. We apply this novel concept to the use case of dynamic resource allocation, and show that the proposed algorithm outperforms existing RL algorithms, yielding a drastic increase in both convergence speed and performance.
{"title":"Inference-based Reinforcement Learning and its Application to Dynamic Resource Allocation","authors":"Paschalis Tsiaflakis, W. Coomans","doi":"10.23919/eusipco55093.2022.9909777","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909777","url":null,"abstract":"Reinforcement learning (RL) is a powerful machine learning technique to learn optimal actions in a control system setup. An important drawback of RL algorithms is the need for balancing exploitation vs exploration. Exploration corresponds to taking randomized actions with the aim to learn from it and make better decisions in the future. However, these exploratory actions result in poor performance, and current RL algorithms have a slow convergence as one can only learn from a single action outcome per iteration. We propose a novel concept of Inference-based RL that is applicable to a specific class of RL problems, and that allows to eliminate the performance impact caused by traditional exploration strategies, thereby making RL performance more consistent and greatly improving the convergence speed. The specific RL problem class is a problem class in which the observation of the outcome of one action can be used to infer the outcome of other actions, without the need to actually perform them. We apply this novel concept to the use case of dynamic resource allocation, and show that the proposed algorithm outperforms existing RL algorithms, yielding a drastic increase in both convergence speed and performance.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"23 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":"133662619","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/eusipco55093.2022.9909595
Vedran Mihal, B. Seifert, Markus Püschel
Directionality is an essential feature of many real-world networks, but problematic in graph signal processing (GSP) because there is no obvious choice of Fourier basis. In this work we investigate how to port GSP methods from undirected to directed graphs using recent work on graph signal denoising using trainable networks as a case study. We consider five notions of directed Fourier bases from the literature and different approaches for porting, from ad-hoc to conceptual. Our experimental results show that directionality does matter, the importance of a shift operator related to the chosen basis, and which directed Fourier basis may be best suited for applications. The best variant also provides a promising method for denoising signals on directed graphs.
{"title":"Porting Signal Processing from Undirected to Directed Graphs: Case Study Signal Denoising with Unrolling Networks","authors":"Vedran Mihal, B. Seifert, Markus Püschel","doi":"10.23919/eusipco55093.2022.9909595","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909595","url":null,"abstract":"Directionality is an essential feature of many real-world networks, but problematic in graph signal processing (GSP) because there is no obvious choice of Fourier basis. In this work we investigate how to port GSP methods from undirected to directed graphs using recent work on graph signal denoising using trainable networks as a case study. We consider five notions of directed Fourier bases from the literature and different approaches for porting, from ad-hoc to conceptual. Our experimental results show that directionality does matter, the importance of a shift operator related to the chosen basis, and which directed Fourier basis may be best suited for applications. The best variant also provides a promising method for denoising signals on directed graphs.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"52 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":"134526643","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/eusipco55093.2022.9909671
Daniela Dapena, D. Lau, G. Arce
Efficient sampling of graph signals is essential to graph signal processing. Recently, blue-noise was introduced as a sampling method that maximizes the separation between sampling nodes leading to high-frequency dominance patterns, and thus, to high-quality patterns. Despite the simple inter-pretation of the method, blue-noise sampling is restricted to approximately regular graphs. This study presents an extension of blue-noise sampling that allows the application of the method to irregular graphs. Before sampling with a blue-noise algorithm, the approach regularizes the weights of the edges such that the graph represents a regular structure. Then, the resulting pattern adapts the node's distribution to the local density of the nodes. This work also uses an approach that minimizes the strength of the high-frequency components to recover approximately bandlimited signals. The experimental results show that the proposed methods have superior performance compared to the state-of-the-art techniques.
{"title":"Density Aware Blue-Noise Sampling on Graphs","authors":"Daniela Dapena, D. Lau, G. Arce","doi":"10.23919/eusipco55093.2022.9909671","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909671","url":null,"abstract":"Efficient sampling of graph signals is essential to graph signal processing. Recently, blue-noise was introduced as a sampling method that maximizes the separation between sampling nodes leading to high-frequency dominance patterns, and thus, to high-quality patterns. Despite the simple inter-pretation of the method, blue-noise sampling is restricted to approximately regular graphs. This study presents an extension of blue-noise sampling that allows the application of the method to irregular graphs. Before sampling with a blue-noise algorithm, the approach regularizes the weights of the edges such that the graph represents a regular structure. Then, the resulting pattern adapts the node's distribution to the local density of the nodes. This work also uses an approach that minimizes the strength of the high-frequency components to recover approximately bandlimited signals. The experimental results show that the proposed methods have superior performance compared to the state-of-the-art techniques.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"73 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":"133083426","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/eusipco55093.2022.9909532
C. Herzet, Clément Elvira, H. Dang
We address the problem of safe screening for $ell_{1}$-penalized convex regression/classification problems, i.e., the identification of zero coordinates of the solutions. Unlike previous contributions of the literature, we propose a screening methodology which does not require the knowledge of a so-called “safe region”. Our approach does not rely on any other assumption than convexity (in particular, no strong-convexity hypothesis is needed) and therefore applies to a wide family of convex problems. When the Fenchel conjugate of the data-fidelity term is strongly convex, we show that the popular “GAP sphere test” proposed by Fercoq et al. can be recovered as a particular case of our methodology (up to a minor modification). We illustrate numerically the performance of our procedure on the “sparse support vector machine classification” problem.
{"title":"Region-free Safe Screening Tests for $ell_{1}$-penalized Convex Problems","authors":"C. Herzet, Clément Elvira, H. Dang","doi":"10.23919/eusipco55093.2022.9909532","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909532","url":null,"abstract":"We address the problem of safe screening for $ell_{1}$-penalized convex regression/classification problems, i.e., the identification of zero coordinates of the solutions. Unlike previous contributions of the literature, we propose a screening methodology which does not require the knowledge of a so-called “safe region”. Our approach does not rely on any other assumption than convexity (in particular, no strong-convexity hypothesis is needed) and therefore applies to a wide family of convex problems. When the Fenchel conjugate of the data-fidelity term is strongly convex, we show that the popular “GAP sphere test” proposed by Fercoq et al. can be recovered as a particular case of our methodology (up to a minor modification). We illustrate numerically the performance of our procedure on the “sparse support vector machine classification” problem.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"99 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":"133365478","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/eusipco55093.2022.9909585
A. Hanif, M. Doroslovački
Simultaneous wireless information and power transfer (SWIPT) has the potential to realize the envisioned ubiquity of the internet of things (IoT) by energizing them wirelessly whilst exchanging information. Recently, low-complexity receiver architectures for SWIPT are being considered for decoding information from amplitude modulated signals after rectification. However, less attention is paid towards improving the non-linear rectifier model prevalent in these architectures which is often truncated till fourth-order term in diode characteristic. In this paper, a novel, tractable analytical model for the rectenna non-linearity is presented which provides a theoretical upper bound to harvested DC power over the amplitude shift keying (ASK) constellation space corresponding to the entire diode non-linear region. Besides, the work also exposes the convexity of harvested DC power vis-à-vis incoming signal power thereby verifying the rate-energy (R-E) tradeoff in SWIPT for different choices of transmitted symbol amplitude distributions. Finally, the theoretical results presented using the adopted model are substantiated with the Monte Carlo circuit simulations allowing to conveniently evaluate and draw compromise in SWIPT performance against a choice of modulation scheme out of the ASK constellation space.
{"title":"Amplitude Shift Keying Constellation Space for Simultaneous Wireless Information and Power Transfer","authors":"A. Hanif, M. Doroslovački","doi":"10.23919/eusipco55093.2022.9909585","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909585","url":null,"abstract":"Simultaneous wireless information and power transfer (SWIPT) has the potential to realize the envisioned ubiquity of the internet of things (IoT) by energizing them wirelessly whilst exchanging information. Recently, low-complexity receiver architectures for SWIPT are being considered for decoding information from amplitude modulated signals after rectification. However, less attention is paid towards improving the non-linear rectifier model prevalent in these architectures which is often truncated till fourth-order term in diode characteristic. In this paper, a novel, tractable analytical model for the rectenna non-linearity is presented which provides a theoretical upper bound to harvested DC power over the amplitude shift keying (ASK) constellation space corresponding to the entire diode non-linear region. Besides, the work also exposes the convexity of harvested DC power vis-à-vis incoming signal power thereby verifying the rate-energy (R-E) tradeoff in SWIPT for different choices of transmitted symbol amplitude distributions. Finally, the theoretical results presented using the adopted model are substantiated with the Monte Carlo circuit simulations allowing to conveniently evaluate and draw compromise in SWIPT performance against a choice of modulation scheme out of the ASK constellation space.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"37 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":"134073512","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/eusipco55093.2022.9909641
Tanuj Hasija, Tim Marrinan
In many applications, components correlated across multiple data sets represent meaningful patterns and commonalities. Estimates of these patterns can be improved when the number of correlated components is known, but since data exploration often occurs in an unsupervised setting, the number of correlated components is generally not known. In this paper, we derive a generalized likelihood ratio test (GLRT) for estimating the number of components correlated across multiple data sets. In particular, we are concerned with the scenario where the number of available samples is small. As a result of the small sample support, correlation coefficients and other summary statistics are significantly overestimated by traditional methods. The proposed test combines linear dimensionality reduction with a GLRT based on a measure of multiset correlation referred as the generalized variance cost function (mCCA-GENVAR). By jointly estimating the rank of the dimensionality reduction and the number of correlated components, we are able to provide high-accuracy estimates in the challenging sample-poor setting. These advantages are illustrated in numerical experiments that compare and contrast the proposed method with existing techniques.
{"title":"A GLRT for estimating the number of correlated components in sample-poor mCCA","authors":"Tanuj Hasija, Tim Marrinan","doi":"10.23919/eusipco55093.2022.9909641","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909641","url":null,"abstract":"In many applications, components correlated across multiple data sets represent meaningful patterns and commonalities. Estimates of these patterns can be improved when the number of correlated components is known, but since data exploration often occurs in an unsupervised setting, the number of correlated components is generally not known. In this paper, we derive a generalized likelihood ratio test (GLRT) for estimating the number of components correlated across multiple data sets. In particular, we are concerned with the scenario where the number of available samples is small. As a result of the small sample support, correlation coefficients and other summary statistics are significantly overestimated by traditional methods. The proposed test combines linear dimensionality reduction with a GLRT based on a measure of multiset correlation referred as the generalized variance cost function (mCCA-GENVAR). By jointly estimating the rank of the dimensionality reduction and the number of correlated components, we are able to provide high-accuracy estimates in the challenging sample-poor setting. These advantages are illustrated in numerical experiments that compare and contrast the proposed method with existing techniques.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"29 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":"114984944","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/eusipco55093.2022.9909669
Genjia Liu, Maosen Li, Siheng Chen
Graph scattering transform (GST) is mathematically-designed graph convolutional model that iteratively applies graph filter banks to achieve comprehensive feature extraction from graph signals. While GST performs excessive decomposition of graph signals in the graph spectral domain, it does not explicitly achieve multiresolution in the graph vertex domain, causing potential failure in handling graphs with hierarchical structures. To address the limitation, this work proposes novel multiscale graph scattering transform (MGST) to achieve hierarchical representations along both graph vertex and spectral domains. With recursive partitioning a graph structure, we yield multiple subgraphs at various scales and then perform scattering frequency decomposition on each subgraph. MGST finally obtains a series of representations and each of them corresponds to a specific graph vertex-spectral subband, achieving multiresolution along both graph vertex and spectral domains. In the experiments, we validate the superior empirical performances of MGST and visualize each graph vertex-spectral subband.
{"title":"Multiscale Graph Scattering Transform","authors":"Genjia Liu, Maosen Li, Siheng Chen","doi":"10.23919/eusipco55093.2022.9909669","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909669","url":null,"abstract":"Graph scattering transform (GST) is mathematically-designed graph convolutional model that iteratively applies graph filter banks to achieve comprehensive feature extraction from graph signals. While GST performs excessive decomposition of graph signals in the graph spectral domain, it does not explicitly achieve multiresolution in the graph vertex domain, causing potential failure in handling graphs with hierarchical structures. To address the limitation, this work proposes novel multiscale graph scattering transform (MGST) to achieve hierarchical representations along both graph vertex and spectral domains. With recursive partitioning a graph structure, we yield multiple subgraphs at various scales and then perform scattering frequency decomposition on each subgraph. MGST finally obtains a series of representations and each of them corresponds to a specific graph vertex-spectral subband, achieving multiresolution along both graph vertex and spectral domains. In the experiments, we validate the superior empirical performances of MGST and visualize each graph vertex-spectral subband.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"9 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":"114994261","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/eusipco55093.2022.9909678
Erik Tegler, Martin Larsson, M. Oskarsson, Kalle Åström
Recent advances in simultaneous estimation of both receiver and sender positions in ad-hoc sensor networks have made it possible to automatically calibrate node positions - a prerequisite for many applications. In man-made environments there are often large planar reflective surfaces that give significant reverberations. In this paper, we study geometric problems of receiver-sender node calibration in the presence of such reflective planes. We establish a rank-1 factorization problem that can be used to simplify the estimation. We also show how to estimate offsets, in the Time difference of arrival case, using only the rank constraint. Finally, we present a new solver for the minimal cases of sender-receiver position estimation. These contributions result in a powerful stratified approach for the node calibration problem, given a reflective plane. The methods are verified with both synthetic and real data.
{"title":"Sensor node calibration in presence of a dominant reflective plane","authors":"Erik Tegler, Martin Larsson, M. Oskarsson, Kalle Åström","doi":"10.23919/eusipco55093.2022.9909678","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909678","url":null,"abstract":"Recent advances in simultaneous estimation of both receiver and sender positions in ad-hoc sensor networks have made it possible to automatically calibrate node positions - a prerequisite for many applications. In man-made environments there are often large planar reflective surfaces that give significant reverberations. In this paper, we study geometric problems of receiver-sender node calibration in the presence of such reflective planes. We establish a rank-1 factorization problem that can be used to simplify the estimation. We also show how to estimate offsets, in the Time difference of arrival case, using only the rank constraint. Finally, we present a new solver for the minimal cases of sender-receiver position estimation. These contributions result in a powerful stratified approach for the node calibration problem, given a reflective plane. The methods are verified with both synthetic and real data.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"111 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":"123434220","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/eusipco55093.2022.9909832
Aylin Tastan, Michael Muma, A. Zoubir
Block diagonal structure of the affinity matrix is advantageous, e.g. in graph-based cluster analysis, where each block corresponds to a cluster. However, constructing block diagonal affinity matrices may be challenging and computationally demanding. We propose a new eigenvalue-based block diagonal representation (EBDR) method. The idea is to estimate a block diagonal affinity matrix by finding an approximation to a vector of target eigenvalues. The target eigenvalues, which follow the ideal block-diagonal model, are efficiently determined based on a vector derived from the graph Laplacian that represents the blocks as a piece-wise linear function. The proposed EBDR shows promising performance compared to four optimally tuned state-of-the-art methods in terms of clustering accuracy and computation time using real-data examples.
{"title":"Eigenvalue-Based Block Diagonal Representation and Application to p-Nearest Neighbor Graphs","authors":"Aylin Tastan, Michael Muma, A. Zoubir","doi":"10.23919/eusipco55093.2022.9909832","DOIUrl":"https://doi.org/10.23919/eusipco55093.2022.9909832","url":null,"abstract":"Block diagonal structure of the affinity matrix is advantageous, e.g. in graph-based cluster analysis, where each block corresponds to a cluster. However, constructing block diagonal affinity matrices may be challenging and computationally demanding. We propose a new eigenvalue-based block diagonal representation (EBDR) method. The idea is to estimate a block diagonal affinity matrix by finding an approximation to a vector of target eigenvalues. The target eigenvalues, which follow the ideal block-diagonal model, are efficiently determined based on a vector derived from the graph Laplacian that represents the blocks as a piece-wise linear function. The proposed EBDR shows promising performance compared to four optimally tuned state-of-the-art methods in terms of clustering accuracy and computation time using real-data examples.","PeriodicalId":231263,"journal":{"name":"2022 30th European Signal Processing Conference (EUSIPCO)","volume":"19 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":"124948598","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：