The Khatri-Rao(KR) subspace method is a high resolution method for direction-of-arrival(DOA) estimation.Combined with 2q level nested array,the KR subspace method can detect O(N2q) sources with N sensors.However,the m...The Khatri-Rao(KR) subspace method is a high resolution method for direction-of-arrival(DOA) estimation.Combined with 2q level nested array,the KR subspace method can detect O(N2q) sources with N sensors.However,the method cannot be applicable to Gaussian sources when q is equal to or greater than 2 since it needs to use 2q-th order cumulants.In this work,a novel approach is presented to conduct DOA estimation by constructing a fourth order difference co-array.Unlike the existing DOA estimation method based on the KR product and 2q level nested array,the proposed method only uses second order statistics,so it can be employed to Gaussian sources as well as non-Gaussian sources.By exploiting a four-level nested array with N elements,our method can also identify O(N4) sources.In order to estimate the wideband signals,the proposed method is extended to the wideband scenarios.Simulation results demonstrate that,compared to the state of the art KR subspace based methods,the new method achieves higher resolution.展开更多
对分块实对称正定矩阵A，B，C和D，证明了一个矩阵等式( A ⊙ B ) # ( C ⊙ D ) = ( A # C ) ⊙ ( B # D )，这里A ⊙ B和A # B分别是A与B的Tracy-Singh乘积和几何平均，如果A和B是分块实对称矩阵，则有矩阵不等式 ≥ ，其中是矩阵和的K...对分块实对称正定矩阵A，B，C和D，证明了一个矩阵等式( A ⊙ B ) # ( C ⊙ D ) = ( A # C ) ⊙ ( B # D )，这里A ⊙ B和A # B分别是A与B的Tracy-Singh乘积和几何平均，如果A和B是分块实对称矩阵，则有矩阵不等式 ≥ ，其中是矩阵和的Khatri -Rao乘积。展开更多
Text classification is a pivotal task in natural language understanding,and its performance has seen remarkable advancements with the rise of Pre-trained Language Models(PLMs).Recently,the proliferation of PLMs has ma...Text classification is a pivotal task in natural language understanding,and its performance has seen remarkable advancements with the rise of Pre-trained Language Models(PLMs).Recently,the proliferation of PLMs has made it increasingly challenging to choose the most suitable model for a given dataset.Since fine-tuning the sheer number of models is impractical,Transferability Estimation(TE)has become a promising solution to efficient model selection.Unlike current TE methods that focus solely on fixed and hard class assignments to evaluate the quality of model-encoded features,our approach further takes into account the intersample and inter-model variations represented by soft class assignments.We achieve this by utilizing class embeddings to predict posterior class assignments,with the logarithm of the maximum posterior evidence serving as the transferability score.Moreover,we found that the informative sub-space of the dataset can lead to more accurate calculation of soft class assignments,where we achieve efficient annotation of informative samples by eliciting the powerful judging ability of large language model.The resulting posterior evidence over the informative sub-space,LogIPE,enables us to capture subtle differences between models,enhancing the accuracy of model selection and validated by extensive experiments conducted on a wide range of text classification datasets as well as candidate PLMs.展开更多
基金Project(2010ZX03006-004) supported by the National Science and Technology Major Program of ChinaProject(YYYJ-1113) supported by the Knowledge Innovation Program of the Chinese Academy of SciencesProject(2011CB302901) supported by the National Basic Research Program of China
文摘The Khatri-Rao(KR) subspace method is a high resolution method for direction-of-arrival(DOA) estimation.Combined with 2q level nested array,the KR subspace method can detect O(N2q) sources with N sensors.However,the method cannot be applicable to Gaussian sources when q is equal to or greater than 2 since it needs to use 2q-th order cumulants.In this work,a novel approach is presented to conduct DOA estimation by constructing a fourth order difference co-array.Unlike the existing DOA estimation method based on the KR product and 2q level nested array,the proposed method only uses second order statistics,so it can be employed to Gaussian sources as well as non-Gaussian sources.By exploiting a four-level nested array with N elements,our method can also identify O(N4) sources.In order to estimate the wideband signals,the proposed method is extended to the wideband scenarios.Simulation results demonstrate that,compared to the state of the art KR subspace based methods,the new method achieves higher resolution.
文摘对分块实对称正定矩阵A，B，C和D，证明了一个矩阵等式( A ⊙ B ) # ( C ⊙ D ) = ( A # C ) ⊙ ( B # D )，这里A ⊙ B和A # B分别是A与B的Tracy-Singh乘积和几何平均，如果A和B是分块实对称矩阵，则有矩阵不等式 ≥ ，其中是矩阵和的Khatri -Rao乘积。
基金supported in part by the National Natural Science Foundation of China(Grant No.62477001).
文摘Text classification is a pivotal task in natural language understanding,and its performance has seen remarkable advancements with the rise of Pre-trained Language Models(PLMs).Recently,the proliferation of PLMs has made it increasingly challenging to choose the most suitable model for a given dataset.Since fine-tuning the sheer number of models is impractical,Transferability Estimation(TE)has become a promising solution to efficient model selection.Unlike current TE methods that focus solely on fixed and hard class assignments to evaluate the quality of model-encoded features,our approach further takes into account the intersample and inter-model variations represented by soft class assignments.We achieve this by utilizing class embeddings to predict posterior class assignments,with the logarithm of the maximum posterior evidence serving as the transferability score.Moreover,we found that the informative sub-space of the dataset can lead to more accurate calculation of soft class assignments,where we achieve efficient annotation of informative samples by eliciting the powerful judging ability of large language model.The resulting posterior evidence over the informative sub-space,LogIPE,enables us to capture subtle differences between models,enhancing the accuracy of model selection and validated by extensive experiments conducted on a wide range of text classification datasets as well as candidate PLMs.
