Pub Date : 2023-07-02DOI: 10.1109/SSP53291.2023.10208006
Maxime Leiber, Y. Marnissi, A. Barrau, M. Mohamed el Badaoui
In this paper, we propose a differentiable version of the short-time Fourier transform (STFT) that allows for gradient-based optimization of the hop length or the frame temporal position by making these parameters continuous. Our approach provides improved control over the temporal positioning of frames, as the continuous nature of the hop length allows for a more finely-tuned optimization. Furthermore, our contribution enables the use of optimization methods such as gradient descent, which are more computationally efficient than conventional discrete optimization methods. Our differentiable STFT can also be easily integrated into existing algorithms and neural networks. We present a simulated illustration to demonstrate the efficacy of our approach and to garner interest from the research community.
{"title":"Differentiable short-time Fourier transform with respect to the hop length","authors":"Maxime Leiber, Y. Marnissi, A. Barrau, M. Mohamed el Badaoui","doi":"10.1109/SSP53291.2023.10208006","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10208006","url":null,"abstract":"In this paper, we propose a differentiable version of the short-time Fourier transform (STFT) that allows for gradient-based optimization of the hop length or the frame temporal position by making these parameters continuous. Our approach provides improved control over the temporal positioning of frames, as the continuous nature of the hop length allows for a more finely-tuned optimization. Furthermore, our contribution enables the use of optimization methods such as gradient descent, which are more computationally efficient than conventional discrete optimization methods. Our differentiable STFT can also be easily integrated into existing algorithms and neural networks. We present a simulated illustration to demonstrate the efficacy of our approach and to garner interest from the research community.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123894448","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 : 2023-07-02DOI: 10.1109/SSP53291.2023.10207938
Alex X. Wang, S. Chukova, Colin Simpson, Binh P. Nguyen
The growth of information technology and advancements in artificial intelligence (AI) have made data creation and usage more prevalent. AI research can be grouped into two categories: model-centric and data-centric. Model-centric AI focuses on using the same data and making changes to model hyper-parameters, architectures, and other configurations. Data-centric AI, on the other hand, prioritizes improving existing data or incorporating new data to improve the performance of machine learning (ML) algorithms. Data-centric AI can greatly improve the performance of machine learning models by improving data quality, increasing data diversity, and using advanced data augmentation methods. The use of ML for early detection of mental health issues is vital due to its ability to identify issues early, provide personalized treatments, detect patterns, and increase accessibility to mental health services. While there have been numerous mental illness detection studies using model-centric approaches, there is a lack of research from a data-centric AI perspective. This study aims to address this gap by comparing established tabular data synthesis methods to explore the impact of synthetic data and data-centric AI on the early detection of mental health issues.
{"title":"Data-centric AI to Improve Early Detection of Mental Illness","authors":"Alex X. Wang, S. Chukova, Colin Simpson, Binh P. Nguyen","doi":"10.1109/SSP53291.2023.10207938","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10207938","url":null,"abstract":"The growth of information technology and advancements in artificial intelligence (AI) have made data creation and usage more prevalent. AI research can be grouped into two categories: model-centric and data-centric. Model-centric AI focuses on using the same data and making changes to model hyper-parameters, architectures, and other configurations. Data-centric AI, on the other hand, prioritizes improving existing data or incorporating new data to improve the performance of machine learning (ML) algorithms. Data-centric AI can greatly improve the performance of machine learning models by improving data quality, increasing data diversity, and using advanced data augmentation methods. The use of ML for early detection of mental health issues is vital due to its ability to identify issues early, provide personalized treatments, detect patterns, and increase accessibility to mental health services. While there have been numerous mental illness detection studies using model-centric approaches, there is a lack of research from a data-centric AI perspective. This study aims to address this gap by comparing established tabular data synthesis methods to explore the impact of synthetic data and data-centric AI on the early detection of mental health issues.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117136516","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 : 2023-07-02DOI: 10.1109/SSP53291.2023.10208019
Jianwei Zhou, Wenjie Wang, Xi Hong, Ming Yang, Chenhao Zhang
Recent decades have seen substantial research into analytical performance analysis of direction-of-arrival (DOA) estimation algorithms, revealing various statistical properties. However, many analyses fail to fully uncover insights into performance even for specific cases. This paper presents additional performance analysis of several subspace-based DOA estimation algorithms, using highly compact and simplified mean squared error (MSE) formulas for different algorithms, including an extension to the spatial smoothing scheme. All statistics are expressed in terms of physical parameters, fully revealing the relationships between array structures and DOA estimation performance.
{"title":"Performance Analysis of DOA Estimation Algorithms Using Physical Parameters in Specific Cases","authors":"Jianwei Zhou, Wenjie Wang, Xi Hong, Ming Yang, Chenhao Zhang","doi":"10.1109/SSP53291.2023.10208019","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10208019","url":null,"abstract":"Recent decades have seen substantial research into analytical performance analysis of direction-of-arrival (DOA) estimation algorithms, revealing various statistical properties. However, many analyses fail to fully uncover insights into performance even for specific cases. This paper presents additional performance analysis of several subspace-based DOA estimation algorithms, using highly compact and simplified mean squared error (MSE) formulas for different algorithms, including an extension to the spatial smoothing scheme. All statistics are expressed in terms of physical parameters, fully revealing the relationships between array structures and DOA estimation performance.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124678269","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 : 2023-07-02DOI: 10.1109/SSP53291.2023.10207933
Yongbin Park, Minchul Kim, Jiwon Yoon
The Laplace mechanism is a commonly employed approach that offers privacy guarantees within the framework of differential privacy. Nevertheless, the Laplace mechanism exhibits two limitations. Firstly, the privacy leakage of data can be exacerbated when the general differential private mechanism is accessed repeatedly with the same input owing to the sequential property of differential privacy. Secondly, the Laplace mechanism may not be suitable for some applications that solely involve positive samples as it can yield unwanted negative samples from the Laplace distribution.We address these issues by utilizing the Gamma distribution to handle database entries that must be consist of positive values ranging from 0 to infinity. In our approach, the epsilon parameter of our mechanism is determined by the value with noise according to the definition of differential privacy. Notably, the range of the noise is unbounded on the right thereby epsilon to approach infinity as the value with noise increases. To mitigate this, we impose constraints on the range of the noise in order to reasonably restrict the epsilon value of the mechanism. However, it should be noted that these constraints may impact the probability of ensuring epsilon-differential privacy and necessitate the imposition of a minimum boundary on the values of dataset. Additionally, we propose a new noise parameter that can be used to adjust the probability of ensuring differential privacy for a fixed epsilon.
{"title":"Differential privacy using Gamma distribution","authors":"Yongbin Park, Minchul Kim, Jiwon Yoon","doi":"10.1109/SSP53291.2023.10207933","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10207933","url":null,"abstract":"The Laplace mechanism is a commonly employed approach that offers privacy guarantees within the framework of differential privacy. Nevertheless, the Laplace mechanism exhibits two limitations. Firstly, the privacy leakage of data can be exacerbated when the general differential private mechanism is accessed repeatedly with the same input owing to the sequential property of differential privacy. Secondly, the Laplace mechanism may not be suitable for some applications that solely involve positive samples as it can yield unwanted negative samples from the Laplace distribution.We address these issues by utilizing the Gamma distribution to handle database entries that must be consist of positive values ranging from 0 to infinity. In our approach, the epsilon parameter of our mechanism is determined by the value with noise according to the definition of differential privacy. Notably, the range of the noise is unbounded on the right thereby epsilon to approach infinity as the value with noise increases. To mitigate this, we impose constraints on the range of the noise in order to reasonably restrict the epsilon value of the mechanism. However, it should be noted that these constraints may impact the probability of ensuring epsilon-differential privacy and necessitate the imposition of a minimum boundary on the values of dataset. Additionally, we propose a new noise parameter that can be used to adjust the probability of ensuring differential privacy for a fixed epsilon.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128334207","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 : 2023-07-02DOI: 10.1109/SSP53291.2023.10207974
Tien Quyet Do, Thanh Ta Minh
Blockchain technology has revolutionized the way transactions are conducted and verified in a decentralized manner. The performance analysis of Ethereum smart contract is crucial in understanding its limitations and potential for various applications. This study aimed to evaluate the gas cost of different sort algorithms and the impact of block size on the throughput of Ethereum network. The results showed that the gas cost of search algorithms such as quick sort and bubble sort varied significantly, with quick sort having a lower cost. Additionally, increasing the block size had a positive impact on the throughput of the Ethereum network, with a higher number of transactions processed per second. These findings provide valuable insights into the performance of Ethereum smart contracts and highlight the importance of considering gas cost and block size in the design and implementation of blockchain-based systems.
{"title":"Performance analysis of Ethereum smart contracts: A Study on Gas cost and block size impact","authors":"Tien Quyet Do, Thanh Ta Minh","doi":"10.1109/SSP53291.2023.10207974","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10207974","url":null,"abstract":"Blockchain technology has revolutionized the way transactions are conducted and verified in a decentralized manner. The performance analysis of Ethereum smart contract is crucial in understanding its limitations and potential for various applications. This study aimed to evaluate the gas cost of different sort algorithms and the impact of block size on the throughput of Ethereum network. The results showed that the gas cost of search algorithms such as quick sort and bubble sort varied significantly, with quick sort having a lower cost. Additionally, increasing the block size had a positive impact on the throughput of the Ethereum network, with a higher number of transactions processed per second. These findings provide valuable insights into the performance of Ethereum smart contracts and highlight the importance of considering gas cost and block size in the design and implementation of blockchain-based systems.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126303585","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 : 2023-07-02DOI: 10.1109/SSP53291.2023.10207931
Manh-Hung Ha, O. Chen
Recurrent Neural Networks (RNNs) have been successful in figuring out applications on time series data. Particularly, effectively capturing local features can ameliorate the performance of RNN. Accordingly, we propose a Rectified Attention Gate Unit (RAGU) which amends Gated Recurrent Unit (GRU) with two special attention mechanisms for RNNs. These two attention mechanisms are a Convolutional Attention (ConvAtt) module performing the convolutional operations on the current input and the previous hidden state to fairly establish the spatiotemporal relationship, and an Attention Module (AM) taking outputs from ConvAtt to fulfill the integrated attention computations for discovering the contextual dependency. Experimental results reveal that RNN using the proposed RAGUs has superior accuracies than RNNs using the other cell units on the HMDB51 and MNIST datasets. Therefore, RAGU proposed herein is an effective model which can bring out outstanding performance for various time series applications.
{"title":"Rectified Attention Gate Unit in Recurrent Neural Networks for Effective Attention Computation","authors":"Manh-Hung Ha, O. Chen","doi":"10.1109/SSP53291.2023.10207931","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10207931","url":null,"abstract":"Recurrent Neural Networks (RNNs) have been successful in figuring out applications on time series data. Particularly, effectively capturing local features can ameliorate the performance of RNN. Accordingly, we propose a Rectified Attention Gate Unit (RAGU) which amends Gated Recurrent Unit (GRU) with two special attention mechanisms for RNNs. These two attention mechanisms are a Convolutional Attention (ConvAtt) module performing the convolutional operations on the current input and the previous hidden state to fairly establish the spatiotemporal relationship, and an Attention Module (AM) taking outputs from ConvAtt to fulfill the integrated attention computations for discovering the contextual dependency. Experimental results reveal that RNN using the proposed RAGUs has superior accuracies than RNNs using the other cell units on the HMDB51 and MNIST datasets. Therefore, RAGU proposed herein is an effective model which can bring out outstanding performance for various time series applications.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126679227","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 : 2023-07-02DOI: 10.1109/SSP53291.2023.10208072
Pham Tuan, N. Trung, M. Adel, E. Guedj
Alzheimer’s Disease (AD) is a most common type of neurodegenerative brain disease in elderly people. Early diagnosis of AD is crucial for providing suitable care. Positron Emission Tomography (PET) images and machine learning can be used to support this purpose. In this paper, we present a method for ranking the effectiveness of brain regions of interest (ROI) to distinguish between stable mild cognitive impairment (sMCI) from progressive mild cognitive impairment (pMCI) in brain PET images based on AutoEncoder (AE). Experiments on the ADNI dataset show that our proposed method significantly improves classifier performance when compared to other popular feature ranking methods such as Fisher score, T-score, and Lasso. Our results suggest that instead of focusing on designing a complex AE structure, we can also use simple-but-multiple AEs for feature ranking. The proposed method could be easily applied to any image dataset where a feature selection is needed.
{"title":"AutoEncoder-based Feature Ranking for Predicting Mild Cognitive Impairment Conversion using FDG-PET Images","authors":"Pham Tuan, N. Trung, M. Adel, E. Guedj","doi":"10.1109/SSP53291.2023.10208072","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10208072","url":null,"abstract":"Alzheimer’s Disease (AD) is a most common type of neurodegenerative brain disease in elderly people. Early diagnosis of AD is crucial for providing suitable care. Positron Emission Tomography (PET) images and machine learning can be used to support this purpose. In this paper, we present a method for ranking the effectiveness of brain regions of interest (ROI) to distinguish between stable mild cognitive impairment (sMCI) from progressive mild cognitive impairment (pMCI) in brain PET images based on AutoEncoder (AE). Experiments on the ADNI dataset show that our proposed method significantly improves classifier performance when compared to other popular feature ranking methods such as Fisher score, T-score, and Lasso. Our results suggest that instead of focusing on designing a complex AE structure, we can also use simple-but-multiple AEs for feature ranking. The proposed method could be easily applied to any image dataset where a feature selection is needed.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"80 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122615685","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 : 2023-07-02DOI: 10.1109/SSP53291.2023.10207937
Zuokun Ouyang, M. Jabloun, P. Ravier
Long-term forecasting problem for time series has been actively studied during the last several years, and preceding Transformer-based models have exploited various self-attention mechanisms to discover the long-range dependencies. However, the hidden dependencies required by the forecasting task are not always appropriately extracted, especially the nonlinear serial dependencies in some datasets. In this paper, we propose a novel Transformer-based model, namely Rankformer, leveraging the rank correlation function and decomposition architecture for long-term time series forecasting tasks. Rankformer outperforms four state-of-the-art Transformer-based models and two RNN-based models for different forecasting horizons on different datasets on which extensive experiments were conducted.
{"title":"Rankformer: Leveraging Rank Correlation for Transformer-based Time Series Forecasting","authors":"Zuokun Ouyang, M. Jabloun, P. Ravier","doi":"10.1109/SSP53291.2023.10207937","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10207937","url":null,"abstract":"Long-term forecasting problem for time series has been actively studied during the last several years, and preceding Transformer-based models have exploited various self-attention mechanisms to discover the long-range dependencies. However, the hidden dependencies required by the forecasting task are not always appropriately extracted, especially the nonlinear serial dependencies in some datasets. In this paper, we propose a novel Transformer-based model, namely Rankformer, leveraging the rank correlation function and decomposition architecture for long-term time series forecasting tasks. Rankformer outperforms four state-of-the-art Transformer-based models and two RNN-based models for different forecasting horizons on different datasets on which extensive experiments were conducted.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"49 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131387643","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 : 2023-07-02DOI: 10.1109/SSP53291.2023.10207976
A. Trapp, P. Wolfsteiner
Higher-order spectra (HOS) provide the frequency-domain decomposition of higher-order moments by cross-frequency correlation. They establish the frequency-domain equivalent to correlation functions and form powerful representations for assessing nonlinear, non-Gaussian, or non-stationary systems and processes. HOS are subdivided into moment and cumulant spectra. While the latter provide a clear assessment of statistical dependence and favorable mathematical properties, cumulant spectra cannot be estimated directly. Their concept requires the identification and removal of spectra of lower order, analogously to their scalar-valued counterparts. So far, HOS applications have been based on third and fourth order and so has the derivation of cumulant spectra. Computational power, advanced methods, and new estimators put forward the interest in expanding HOS analysis to orders above four. This paper presents the combinatorial framework to define nth-order cumulant spectra in the frequency domain. On this basis, sixth-order spectral estimates are employed to differentiate two processes of same PSD and trispectrum.
{"title":"Derivation of N-TH Order Cumulant Spectra","authors":"A. Trapp, P. Wolfsteiner","doi":"10.1109/SSP53291.2023.10207976","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10207976","url":null,"abstract":"Higher-order spectra (HOS) provide the frequency-domain decomposition of higher-order moments by cross-frequency correlation. They establish the frequency-domain equivalent to correlation functions and form powerful representations for assessing nonlinear, non-Gaussian, or non-stationary systems and processes. HOS are subdivided into moment and cumulant spectra. While the latter provide a clear assessment of statistical dependence and favorable mathematical properties, cumulant spectra cannot be estimated directly. Their concept requires the identification and removal of spectra of lower order, analogously to their scalar-valued counterparts. So far, HOS applications have been based on third and fourth order and so has the derivation of cumulant spectra. Computational power, advanced methods, and new estimators put forward the interest in expanding HOS analysis to orders above four. This paper presents the combinatorial framework to define nth-order cumulant spectra in the frequency domain. On this basis, sixth-order spectral estimates are employed to differentiate two processes of same PSD and trispectrum.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128104890","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 : 2023-07-02DOI: 10.1109/SSP53291.2023.10208014
Jitendra Tugnait
We consider the problem of estimating differences in two Gaussian graphical models (GGMs) which are known to have similar structure. The GGM structure is encoded in its precision (inverse covariance) matrix. In many applications one is interested in estimating the difference in two precision matrices to characterize underlying changes in conditional dependencies of two sets of data. Most existing methods for differential graph estimation are based on a lasso penalized loss function. In this paper, we analyze a log-sum penalized D-trace loss function approach for differential graph learning. An alternating direction method of multipliers (ADMM) algorithm is presented to optimize the objective function. Theoretical analysis establishing consistency in estimation in high-dimensional settings is provided. We illustrate our approach using a numerical example where log-sum penalized D-trace loss significantly outperforms lasso-penalized D-trace loss as well as smoothly clipped absolute deviation (SCAD) penalized D-trace loss.
{"title":"Estimation of Differential Graphs via Log-Sum Penalized D-Trace Loss","authors":"Jitendra Tugnait","doi":"10.1109/SSP53291.2023.10208014","DOIUrl":"https://doi.org/10.1109/SSP53291.2023.10208014","url":null,"abstract":"We consider the problem of estimating differences in two Gaussian graphical models (GGMs) which are known to have similar structure. The GGM structure is encoded in its precision (inverse covariance) matrix. In many applications one is interested in estimating the difference in two precision matrices to characterize underlying changes in conditional dependencies of two sets of data. Most existing methods for differential graph estimation are based on a lasso penalized loss function. In this paper, we analyze a log-sum penalized D-trace loss function approach for differential graph learning. An alternating direction method of multipliers (ADMM) algorithm is presented to optimize the objective function. Theoretical analysis establishing consistency in estimation in high-dimensional settings is provided. We illustrate our approach using a numerical example where log-sum penalized D-trace loss significantly outperforms lasso-penalized D-trace loss as well as smoothly clipped absolute deviation (SCAD) penalized D-trace loss.","PeriodicalId":296346,"journal":{"name":"2023 IEEE Statistical Signal Processing Workshop (SSP)","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131798828","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：