Reachability query is a fundamental problem on graphs, which has been extensively studied in acad... more Reachability query is a fundamental problem on graphs, which has been extensively studied in academia and industry. Since graphs are subject to frequent updates in many applications, it is essential to support efficient graph updates while offering good performance in reachability queries. Existing solutions compress the original graph with the Directed Acyclic Graph (DAG) and propose efficient query processing and index update techniques. However, they focus on optimizing the scenarios where the Strong Connected Components (SCCs) remain unchanged and have overlooked the prohibitively high cost of the DAG maintenance when SCCs are updated. In this paper, we propose DBL, an efficient DAG-free index to support the reachability query on dynamic graphs with insertion-only updates. DBL builds on two complementary indexes: Dynamic Landmark (DL) label and Bidirectional Leaf (BL) label. The former leverages landmark nodes to quickly determine reachable pairs whereas the latter prunes unreachable pairs by indexing the leaf nodes in the graph. We evaluate DBL against the state-of-the-art approaches on dynamic reachability index with extensive experiments on real-world datasets. The results have demonstrated that DBL achieves orders of magnitude speedup in terms of index update, while still producing competitive query efficiency.
Incorporating syntax into neural approaches in NLP has a multitude of practical and scientific be... more Incorporating syntax into neural approaches in NLP has a multitude of practical and scientific benefits. For instance, a language model that is syntax-aware is likely to be able to produce better samples; even a discriminative model like BERT with a syntax module could be used for core NLP tasks like unsupervised syntactic parsing. Rapid progress in recent years was arguably spurred on by the empirical success of the Parsing-Reading-Predict architecture of (Shen et al., 2018a), later simplified by the Order Neuron LSTM of (Shen et al., 2019). Most notably, this is the first time neural approaches were able to successfully perform unsupervised syntactic parsing (evaluated by various metrics like F-1 score). However, even heuristic (much less fully mathematical) understanding of why and when these architectures work is lagging severely behind. In this work, we answer representational questions raised by the architectures in (Shen et al., 2018a, 2019), as well as some transition-based ...
Massive volumes of data continuously generated on social platforms have become an important infor... more Massive volumes of data continuously generated on social platforms have become an important information source for users. A primary method to obtain fresh and valuable information from social streams is \emph{social search}. Although there have been extensive studies on social search, existing methods only focus on the \emph{relevance} of query results but ignore the \emph{representativeness}. In this paper, we propose a novel Semantic and Influence aware $k$-Representative ($k$-SIR) query for social streams based on topic modeling. Specifically, we consider that both user queries and elements are represented as vectors in the topic space. A $k$-SIR query retrieves a set of $k$ elements with the maximum \emph{representativeness} over the sliding window at query time w.r.t. the query vector. The representativeness of an element set comprises both semantic and influence scores computed by the topic model. Subsequently, we design two approximation algorithms, namely \textsc{Multi-Topic...
Proceedings of the 2017 ACM International Conference on Management of Data, 2017
Social influence has attracted significant attention owing to the prevalence of social networks (... more Social influence has attracted significant attention owing to the prevalence of social networks (SNs). In this paper, we study a new social influence problem, called personalized social influential tags exploration (PITEX), to help any user in the SN explore how she influences the network. Given a target user, it finds a size-k tag set that maximizes this user's social influence. We prove the problem is NP-hard to be approximated within any constant ratio. To solve it, we introduce a sampling-based framework, which has an approximation ratio of 1− 1+ with high probabilistic guarantee. To speedup the computation, we devise more efficient sampling techniques and propose best-effort exploration to quickly prune tag sets with small influence. To further enable instant exploration, we devise a novel index structure and develop effective pruning and materialization techniques. Experimental results on real large-scale datasets validate our theoretical findings and show high performances of our proposed methods.
Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data, 2020
Subgraph enumeration is important for many applications such as network motif discovery and commu... more Subgraph enumeration is important for many applications such as network motif discovery and community detection. Recent works utilize graphics processing units (GPUs) to parallelize subgraph enumeration, but they can only handle graphs that fit into the GPU memory. In this paper, we propose a new approach for GPU-accelerated subgraph enumeration that can efficiently scale to large graphs beyond the GPU memory. Our approach divides the graph into partitions, each of which fits into the GPU memory. The GPU processes one partition at a time and searches the matched subgraphs of a given pattern (i.e., instances) within the partition as in the small graph. The key challenge is on enumerating the instances across different partitions, because this search would enumerate considerably redundant subgraphs and cause the expensive data transfer cost via the PCI-e bus. Therefore, we propose a novel shared execution approach to eliminate the redundant subgraph searches and correctly generate all the instances across different partitions. The experimental evaluation shows that our approach can scale to large graphs and achieve significantly better performance than the existing single-machine solutions.
IEEE Transactions on Knowledge and Data Engineering, 2018
Representative subset selection (RSS) is an important tool for users to draw insights from massiv... more Representative subset selection (RSS) is an important tool for users to draw insights from massive datasets. Existing literature models RSS as the submodular maximization problem to capture the "diminishing returns" property of the representativeness of selected subsets, but often only has a single constraint (e.g., cardinality), which limits its applications in many real-world problems. To capture the data recency issue and support different types of constraints, we formulate dynamic RSS in data streams as maximizing submodular functions subject to general d-knapsack constraints (SMDK) over sliding windows. We propose a KNAPWINDOW framework (KW) for SMDK. KW utilizes the KNAPSTREAM algorithm (KS) for SMDK in append-only streams as a subroutine. It maintains a sequence of checkpoints and KS instances over the sliding window. Theoretically, KW is 1−ε 1+d-approximate for SMDK. Furthermore, we propose a KNAPWINDOWPLUS framework (KW +) to improve upon KW. KW + builds an index SUBKNAPCHK to manage the checkpoints and KS instances. SUBKNAPCHK deletes a checkpoint whenever it can be approximated by its successors. By keeping much fewer checkpoints, KW + achieves higher efficiency than KW while still guaranteeing a 1−ε ′ 2+2d-approximate solution for SMDK. Finally, we evaluate the efficiency and solution quality of KW and KW + in real-world datasets. The experimental results demonstrate that KW achieves more than two orders of magnitude speedups over the batch baseline and preserves high-quality solutions for SMDK over sliding windows. KW + further runs 5-10 times faster than KW while providing solutions with equivalent or even better utilities.
Many real applications in real-time news stream advertising call for efficient processing of long... more Many real applications in real-time news stream advertising call for efficient processing of long queries against short text. In such applications, dynamic news feeds are regarded as queries to match against an advertisement (ad) database for retrieving the k most relevant ads. The existing approaches to keyword retrieval cannot work well in this search scenario when queries are triggered at a very high frequency. To address the problem, we introduce new techniques to significantly improve search performance. First, we devise a two-level partitioning for tight upper bound estimation and a lazy evaluation scheme to delay full evaluation of unpromising candidates, which can bring three to four times performance boosting in a database with 7 million ads. Second, we propose a novel rank-aware block-oriented inverted index to further improve performance. In this index scheme, each entry in an inverted list is assigned a rank according to its importance in the ad. Then, we introduce a blo...
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021
Deception is an effective means of jamming against synthetic aperture radar (SAR). The performanc... more Deception is an effective means of jamming against synthetic aperture radar (SAR). The performance of deceptive jamming is affected by the accuracy of parameter measurement and the applied antijamming methods of SAR. In this article, we analyze the accuracy of current deceptive jamming evaluation indicators, propose a new method for evaluating the performance of deceptive jamming based on the combination of typical dominant deceptive jamming evaluation indicators, and recessive indicators extracted by convolutional neural networks, and obtain a level of deceptive jamming using a softmax activation function in a fully connected network. Deceptive jamming images affected by the SAR motion parameter measurement error are taken as training and test sets. Finally, the moving and stationary target acquisition and recognition database is used as a deceptive jamming template in order to verify the effectiveness of the proposed method. Index Terms-Convolutional neural network (CNN), deceptive jamming, electronic countermeasure (ECM), synthetic aperture radar (SAR). I. INTRODUCTION S YNTHETIC aperture radar (SAR) is widely used in military and civilian applications, such as target classification, identification, and detection due to its high resolution at all times of day and in all types of weather [1]. To protect targets or regions of interest from the malicious reconnaissance of SAR, research into jamming has been of great interest in electronic countermeasure
This study examines multi-label image classification challenges within a Kagle competition contex... more This study examines multi-label image classification challenges within a Kagle competition context, employing two distinct neural network architectures: the CLIP model, utilizing a transformer-based approach for integrating visual and textual data, and a hybrid model that combines GoogLeNet with LSTM to process image features alongside sequential text. The dataset, characterized by significant class imbalance, is addressed through sophisticated preprocessing techniques, the implementation of Focal Loss, and strategic data resampling to enhance model training. Experimental results demonstrate the superiority of the CLIP model in terms of precision and validation accuracy. This model's success can be attributed to its ability to generalize effectively from diverse and extensive pre-training on internet-sourced data, which is essential given the multimodal nature and imbalance present in the dataset. The CLIP model's dual encoding capability efficiently handles both visual and textual inputs, providing a robust framework for understanding complex data interactions. Additionally, the research highlights the critical role of precise model selection, advanced data preprocessing, and the optimization of loss functions and optimizers in improving performance in multi-label classification tasks. By leveraging textual information to refine visual data classification, the study advances computational tools and strategies, ensuring higher predictive accuracy and robustness in handling real-world data complexities.
This study presents an extensive comparative evaluation of Optical Character Recognition (OCR) pe... more This study presents an extensive comparative evaluation of Optical Character Recognition (OCR) performance across various image pre- processing techniques such as Edge Detection and Image Thresh- olding, and various datasets including the IAM dataset and FUNSD dataset, to investigate the strengths and weaknesses of each OCR engine under different circumstances such as recognizing different hand-writings and extracting key information from complex layout forms under poor image resolution.
The OCR engines presented in the study include Tesseract OCR, Keras OCR, Paddle OCR, and Microsoft Azure Computer Vision. An innovative design and implementation of an integrated OCR Scoring System which, intricately assesses each engine’s output, integrating confidence scores and Character Error Rates (CER) to determine the most accurate text extraction, is developed and evaluated.
The selection of evaluation metrics—accuracy, precision, recall, F1 score, Character Error Rate (CER), Word Error Rate (WER), and running time per image (RT)—was strategic to comprehensively dis- tinguish the performance of individual OCR engines. Precision and recall offer insights into an OCR engine’s inclusivity versus its fo- cus on detail, which bears significance in handling complex data sets. Running time was tracked to evaluate processing speed, providing a pragmatic lens for understanding the trade-offs between speed and accuracy in real-world applications. demonstrate that while individ- ual OCR engines have their strengths, the integrated
Our findings reveal that the Integrated Method, which amalgamates the strengths of individual OCR engines, consistently outperforms its constituent systems across all tested conditions. It demonstrates remarkable adaptability and resilience, achieving superior accuracy, precision, and recall rates, particularly in the face of the image qual- ity fluctuations inherent in real-world tasks. Azure OCR emerged as a significant contributor to this integrated success, showing exceptional processing speed and precision, with notable proficiency in decipher- ing complex handwriting—a capability that suggests a high degree of immunity to image quality variations.
This report details the development of classifiers with enhanced noise robustness, an essential f... more This report details the development of classifiers with enhanced noise robustness, an essential feature for improving the performance of machine learning models when faced with label noise in training datasets. The robustness of two neural network architectures, ResNet and GoogLeNet, is investigated over three datasets: Fashion MNIST 0.5, Fashion MNIST 0.6, and CIFAR. Our study utilizes forward and backward loss-correction techniques along with an important reweighting approach, employing transition matrices to train unbiased estimators for the clean data distribution from noisy datasets. For CIFAR, we estimate the transition matrix using anchor points due to the absence of a predefined matrix, while for Fashion MNIST, we use the provided matrices. Our result indicates the superiority of forward correction in improving accuracy and F1 scores. The report concludes with insights into the comparative performance of the correction techniques and their implications for future noise-robust model development.
Non-negative Matrix Factorization (NMF) is a potent technique for deriving low-rank approximation... more Non-negative Matrix Factorization (NMF) is a potent technique for deriving low-rank approximations of data matrices. However, its conventional formulation, anchored by the Least Squares Error loss function, is vulnerable to the detrimental impacts of noise and outliers. In this study, we investigate three advanced NMF variants: the L 2 Norm-based NMF, the L 2,1 Norm-based NMF, and notably, the Truncated-Cauchy (TC) NMF, which emerged as our model of choice due to its superior resilience against corrupted data. Our evaluation encompasses image clustering tasks using two prominent datasets: ORL and Extended Yale. By systematically introducing noise perturbations of diverse types and magnitudes, we rigorously test the robustness of each NMF variant. The matrix factorization quality is assessed via the Root Mean Squared Error, while clustering performance is evaluated through Accuracy and Normalised Mutual Information (NMI) metrics. The Truncated-Cauchy NMF, with its inherent robustness to outliers, consistently demonstrated superior performance, underscoring its potential as a leading choice for handling corrupted data scenarios.
Reachability query is a fundamental problem on graphs, which has been extensively studied in acad... more Reachability query is a fundamental problem on graphs, which has been extensively studied in academia and industry. Since graphs are subject to frequent updates in many applications, it is essential to support efficient graph updates while offering good performance in reachability queries. Existing solutions compress the original graph with the Directed Acyclic Graph (DAG) and propose efficient query processing and index update techniques. However, they focus on optimizing the scenarios where the Strong Connected Components (SCCs) remain unchanged and have overlooked the prohibitively high cost of the DAG maintenance when SCCs are updated. In this paper, we propose DBL, an efficient DAG-free index to support the reachability query on dynamic graphs with insertion-only updates. DBL builds on two complementary indexes: Dynamic Landmark (DL) label and Bidirectional Leaf (BL) label. The former leverages landmark nodes to quickly determine reachable pairs whereas the latter prunes unreachable pairs by indexing the leaf nodes in the graph. We evaluate DBL against the state-of-the-art approaches on dynamic reachability index with extensive experiments on real-world datasets. The results have demonstrated that DBL achieves orders of magnitude speedup in terms of index update, while still producing competitive query efficiency.
Incorporating syntax into neural approaches in NLP has a multitude of practical and scientific be... more Incorporating syntax into neural approaches in NLP has a multitude of practical and scientific benefits. For instance, a language model that is syntax-aware is likely to be able to produce better samples; even a discriminative model like BERT with a syntax module could be used for core NLP tasks like unsupervised syntactic parsing. Rapid progress in recent years was arguably spurred on by the empirical success of the Parsing-Reading-Predict architecture of (Shen et al., 2018a), later simplified by the Order Neuron LSTM of (Shen et al., 2019). Most notably, this is the first time neural approaches were able to successfully perform unsupervised syntactic parsing (evaluated by various metrics like F-1 score). However, even heuristic (much less fully mathematical) understanding of why and when these architectures work is lagging severely behind. In this work, we answer representational questions raised by the architectures in (Shen et al., 2018a, 2019), as well as some transition-based ...
Massive volumes of data continuously generated on social platforms have become an important infor... more Massive volumes of data continuously generated on social platforms have become an important information source for users. A primary method to obtain fresh and valuable information from social streams is \emph{social search}. Although there have been extensive studies on social search, existing methods only focus on the \emph{relevance} of query results but ignore the \emph{representativeness}. In this paper, we propose a novel Semantic and Influence aware $k$-Representative ($k$-SIR) query for social streams based on topic modeling. Specifically, we consider that both user queries and elements are represented as vectors in the topic space. A $k$-SIR query retrieves a set of $k$ elements with the maximum \emph{representativeness} over the sliding window at query time w.r.t. the query vector. The representativeness of an element set comprises both semantic and influence scores computed by the topic model. Subsequently, we design two approximation algorithms, namely \textsc{Multi-Topic...
Proceedings of the 2017 ACM International Conference on Management of Data, 2017
Social influence has attracted significant attention owing to the prevalence of social networks (... more Social influence has attracted significant attention owing to the prevalence of social networks (SNs). In this paper, we study a new social influence problem, called personalized social influential tags exploration (PITEX), to help any user in the SN explore how she influences the network. Given a target user, it finds a size-k tag set that maximizes this user's social influence. We prove the problem is NP-hard to be approximated within any constant ratio. To solve it, we introduce a sampling-based framework, which has an approximation ratio of 1− 1+ with high probabilistic guarantee. To speedup the computation, we devise more efficient sampling techniques and propose best-effort exploration to quickly prune tag sets with small influence. To further enable instant exploration, we devise a novel index structure and develop effective pruning and materialization techniques. Experimental results on real large-scale datasets validate our theoretical findings and show high performances of our proposed methods.
Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data, 2020
Subgraph enumeration is important for many applications such as network motif discovery and commu... more Subgraph enumeration is important for many applications such as network motif discovery and community detection. Recent works utilize graphics processing units (GPUs) to parallelize subgraph enumeration, but they can only handle graphs that fit into the GPU memory. In this paper, we propose a new approach for GPU-accelerated subgraph enumeration that can efficiently scale to large graphs beyond the GPU memory. Our approach divides the graph into partitions, each of which fits into the GPU memory. The GPU processes one partition at a time and searches the matched subgraphs of a given pattern (i.e., instances) within the partition as in the small graph. The key challenge is on enumerating the instances across different partitions, because this search would enumerate considerably redundant subgraphs and cause the expensive data transfer cost via the PCI-e bus. Therefore, we propose a novel shared execution approach to eliminate the redundant subgraph searches and correctly generate all the instances across different partitions. The experimental evaluation shows that our approach can scale to large graphs and achieve significantly better performance than the existing single-machine solutions.
IEEE Transactions on Knowledge and Data Engineering, 2018
Representative subset selection (RSS) is an important tool for users to draw insights from massiv... more Representative subset selection (RSS) is an important tool for users to draw insights from massive datasets. Existing literature models RSS as the submodular maximization problem to capture the "diminishing returns" property of the representativeness of selected subsets, but often only has a single constraint (e.g., cardinality), which limits its applications in many real-world problems. To capture the data recency issue and support different types of constraints, we formulate dynamic RSS in data streams as maximizing submodular functions subject to general d-knapsack constraints (SMDK) over sliding windows. We propose a KNAPWINDOW framework (KW) for SMDK. KW utilizes the KNAPSTREAM algorithm (KS) for SMDK in append-only streams as a subroutine. It maintains a sequence of checkpoints and KS instances over the sliding window. Theoretically, KW is 1−ε 1+d-approximate for SMDK. Furthermore, we propose a KNAPWINDOWPLUS framework (KW +) to improve upon KW. KW + builds an index SUBKNAPCHK to manage the checkpoints and KS instances. SUBKNAPCHK deletes a checkpoint whenever it can be approximated by its successors. By keeping much fewer checkpoints, KW + achieves higher efficiency than KW while still guaranteeing a 1−ε ′ 2+2d-approximate solution for SMDK. Finally, we evaluate the efficiency and solution quality of KW and KW + in real-world datasets. The experimental results demonstrate that KW achieves more than two orders of magnitude speedups over the batch baseline and preserves high-quality solutions for SMDK over sliding windows. KW + further runs 5-10 times faster than KW while providing solutions with equivalent or even better utilities.
Many real applications in real-time news stream advertising call for efficient processing of long... more Many real applications in real-time news stream advertising call for efficient processing of long queries against short text. In such applications, dynamic news feeds are regarded as queries to match against an advertisement (ad) database for retrieving the k most relevant ads. The existing approaches to keyword retrieval cannot work well in this search scenario when queries are triggered at a very high frequency. To address the problem, we introduce new techniques to significantly improve search performance. First, we devise a two-level partitioning for tight upper bound estimation and a lazy evaluation scheme to delay full evaluation of unpromising candidates, which can bring three to four times performance boosting in a database with 7 million ads. Second, we propose a novel rank-aware block-oriented inverted index to further improve performance. In this index scheme, each entry in an inverted list is assigned a rank according to its importance in the ad. Then, we introduce a blo...
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021
Deception is an effective means of jamming against synthetic aperture radar (SAR). The performanc... more Deception is an effective means of jamming against synthetic aperture radar (SAR). The performance of deceptive jamming is affected by the accuracy of parameter measurement and the applied antijamming methods of SAR. In this article, we analyze the accuracy of current deceptive jamming evaluation indicators, propose a new method for evaluating the performance of deceptive jamming based on the combination of typical dominant deceptive jamming evaluation indicators, and recessive indicators extracted by convolutional neural networks, and obtain a level of deceptive jamming using a softmax activation function in a fully connected network. Deceptive jamming images affected by the SAR motion parameter measurement error are taken as training and test sets. Finally, the moving and stationary target acquisition and recognition database is used as a deceptive jamming template in order to verify the effectiveness of the proposed method. Index Terms-Convolutional neural network (CNN), deceptive jamming, electronic countermeasure (ECM), synthetic aperture radar (SAR). I. INTRODUCTION S YNTHETIC aperture radar (SAR) is widely used in military and civilian applications, such as target classification, identification, and detection due to its high resolution at all times of day and in all types of weather [1]. To protect targets or regions of interest from the malicious reconnaissance of SAR, research into jamming has been of great interest in electronic countermeasure
This study examines multi-label image classification challenges within a Kagle competition contex... more This study examines multi-label image classification challenges within a Kagle competition context, employing two distinct neural network architectures: the CLIP model, utilizing a transformer-based approach for integrating visual and textual data, and a hybrid model that combines GoogLeNet with LSTM to process image features alongside sequential text. The dataset, characterized by significant class imbalance, is addressed through sophisticated preprocessing techniques, the implementation of Focal Loss, and strategic data resampling to enhance model training. Experimental results demonstrate the superiority of the CLIP model in terms of precision and validation accuracy. This model's success can be attributed to its ability to generalize effectively from diverse and extensive pre-training on internet-sourced data, which is essential given the multimodal nature and imbalance present in the dataset. The CLIP model's dual encoding capability efficiently handles both visual and textual inputs, providing a robust framework for understanding complex data interactions. Additionally, the research highlights the critical role of precise model selection, advanced data preprocessing, and the optimization of loss functions and optimizers in improving performance in multi-label classification tasks. By leveraging textual information to refine visual data classification, the study advances computational tools and strategies, ensuring higher predictive accuracy and robustness in handling real-world data complexities.
This study presents an extensive comparative evaluation of Optical Character Recognition (OCR) pe... more This study presents an extensive comparative evaluation of Optical Character Recognition (OCR) performance across various image pre- processing techniques such as Edge Detection and Image Thresh- olding, and various datasets including the IAM dataset and FUNSD dataset, to investigate the strengths and weaknesses of each OCR engine under different circumstances such as recognizing different hand-writings and extracting key information from complex layout forms under poor image resolution.
The OCR engines presented in the study include Tesseract OCR, Keras OCR, Paddle OCR, and Microsoft Azure Computer Vision. An innovative design and implementation of an integrated OCR Scoring System which, intricately assesses each engine’s output, integrating confidence scores and Character Error Rates (CER) to determine the most accurate text extraction, is developed and evaluated.
The selection of evaluation metrics—accuracy, precision, recall, F1 score, Character Error Rate (CER), Word Error Rate (WER), and running time per image (RT)—was strategic to comprehensively dis- tinguish the performance of individual OCR engines. Precision and recall offer insights into an OCR engine’s inclusivity versus its fo- cus on detail, which bears significance in handling complex data sets. Running time was tracked to evaluate processing speed, providing a pragmatic lens for understanding the trade-offs between speed and accuracy in real-world applications. demonstrate that while individ- ual OCR engines have their strengths, the integrated
Our findings reveal that the Integrated Method, which amalgamates the strengths of individual OCR engines, consistently outperforms its constituent systems across all tested conditions. It demonstrates remarkable adaptability and resilience, achieving superior accuracy, precision, and recall rates, particularly in the face of the image qual- ity fluctuations inherent in real-world tasks. Azure OCR emerged as a significant contributor to this integrated success, showing exceptional processing speed and precision, with notable proficiency in decipher- ing complex handwriting—a capability that suggests a high degree of immunity to image quality variations.
This report details the development of classifiers with enhanced noise robustness, an essential f... more This report details the development of classifiers with enhanced noise robustness, an essential feature for improving the performance of machine learning models when faced with label noise in training datasets. The robustness of two neural network architectures, ResNet and GoogLeNet, is investigated over three datasets: Fashion MNIST 0.5, Fashion MNIST 0.6, and CIFAR. Our study utilizes forward and backward loss-correction techniques along with an important reweighting approach, employing transition matrices to train unbiased estimators for the clean data distribution from noisy datasets. For CIFAR, we estimate the transition matrix using anchor points due to the absence of a predefined matrix, while for Fashion MNIST, we use the provided matrices. Our result indicates the superiority of forward correction in improving accuracy and F1 scores. The report concludes with insights into the comparative performance of the correction techniques and their implications for future noise-robust model development.
Non-negative Matrix Factorization (NMF) is a potent technique for deriving low-rank approximation... more Non-negative Matrix Factorization (NMF) is a potent technique for deriving low-rank approximations of data matrices. However, its conventional formulation, anchored by the Least Squares Error loss function, is vulnerable to the detrimental impacts of noise and outliers. In this study, we investigate three advanced NMF variants: the L 2 Norm-based NMF, the L 2,1 Norm-based NMF, and notably, the Truncated-Cauchy (TC) NMF, which emerged as our model of choice due to its superior resilience against corrupted data. Our evaluation encompasses image clustering tasks using two prominent datasets: ORL and Extended Yale. By systematically introducing noise perturbations of diverse types and magnitudes, we rigorously test the robustness of each NMF variant. The matrix factorization quality is assessed via the Root Mean Squared Error, while clustering performance is evaluated through Accuracy and Normalised Mutual Information (NMI) metrics. The Truncated-Cauchy NMF, with its inherent robustness to outliers, consistently demonstrated superior performance, underscoring its potential as a leading choice for handling corrupted data scenarios.
Uploads
Papers by Yuchen Li
The OCR engines presented in the study include Tesseract OCR, Keras OCR, Paddle OCR, and Microsoft Azure Computer Vision. An innovative design and implementation of an integrated OCR Scoring System which, intricately assesses each engine’s output, integrating confidence scores and Character Error Rates (CER) to determine the most accurate text extraction, is developed and evaluated.
The selection of evaluation metrics—accuracy, precision, recall, F1 score, Character Error Rate (CER), Word Error Rate (WER), and running time per image (RT)—was strategic to comprehensively dis- tinguish the performance of individual OCR engines. Precision and recall offer insights into an OCR engine’s inclusivity versus its fo- cus on detail, which bears significance in handling complex data sets. Running time was tracked to evaluate processing speed, providing a pragmatic lens for understanding the trade-offs between speed and accuracy in real-world applications. demonstrate that while individ- ual OCR engines have their strengths, the integrated
Our findings reveal that the Integrated Method, which amalgamates the strengths of individual OCR engines, consistently outperforms its constituent systems across all tested conditions. It demonstrates remarkable adaptability and resilience, achieving superior accuracy, precision, and recall rates, particularly in the face of the image qual- ity fluctuations inherent in real-world tasks. Azure OCR emerged as a significant contributor to this integrated success, showing exceptional processing speed and precision, with notable proficiency in decipher- ing complex handwriting—a capability that suggests a high degree of immunity to image quality variations.
The OCR engines presented in the study include Tesseract OCR, Keras OCR, Paddle OCR, and Microsoft Azure Computer Vision. An innovative design and implementation of an integrated OCR Scoring System which, intricately assesses each engine’s output, integrating confidence scores and Character Error Rates (CER) to determine the most accurate text extraction, is developed and evaluated.
The selection of evaluation metrics—accuracy, precision, recall, F1 score, Character Error Rate (CER), Word Error Rate (WER), and running time per image (RT)—was strategic to comprehensively dis- tinguish the performance of individual OCR engines. Precision and recall offer insights into an OCR engine’s inclusivity versus its fo- cus on detail, which bears significance in handling complex data sets. Running time was tracked to evaluate processing speed, providing a pragmatic lens for understanding the trade-offs between speed and accuracy in real-world applications. demonstrate that while individ- ual OCR engines have their strengths, the integrated
Our findings reveal that the Integrated Method, which amalgamates the strengths of individual OCR engines, consistently outperforms its constituent systems across all tested conditions. It demonstrates remarkable adaptability and resilience, achieving superior accuracy, precision, and recall rates, particularly in the face of the image qual- ity fluctuations inherent in real-world tasks. Azure OCR emerged as a significant contributor to this integrated success, showing exceptional processing speed and precision, with notable proficiency in decipher- ing complex handwriting—a capability that suggests a high degree of immunity to image quality variations.