Session1 03.hsian-an wang

0 likes99 views

This document discusses using text models to improve the accuracy of optical character recognition (OCR) on Chinese rare books. It conducted experiments using n-gram, backward/forward n-gram, and LSTM models on OCR data from ancient medicine books. The backward and forward 4-gram model achieved the highest correction rate at 97.57%. Mixing the LSTM 6-gram model with the OCR's top 5 candidates and probability of the top candidate further improved accuracy to 97.71%, demonstrating that combining text models with OCR probabilities can better correct OCR errors than text models alone. In conclusion, text models are effective for increasing OCR accuracy on rare books, with backward/forward 4-gram and LSTM 6-gram

Technology

Towards a Higher Accuracy of Optical
Character Recognition of Chinese Rare
Books in Making Use of Text Model
Hsiang-An Wang
Academia Sinica
Center for Digital Cultures

Limitation (Missing and Extra Word)
OCR Original OCR Original
3

Experiment: Data Collection
• Training dataset: 187 ancient medicine books
from the Scripta Sinica Database (about 40
million words)
• Testing dataset: 1 relevant ancient medicine
book named “ ” with a total of
185,000 words
• The OCR results are about 180,000 words
correct and about 5000 incorrect words,
which means the correct rate is about 97.3 %
4

Experiment: Building a N-gram Model
• Relied on the sequence of words in the
training dataset, and thus we picked the
highest frequency of output.
• " "
– 2-gram: input to predict " "
– 3-gram: input predict " "
– 4-gram: input predict " "
– ...
5

Experiment: Building a
Backward and Forward N-gram Model
• Relied on the sequence of backward and forward
words in the training dataset, and thus we picked the
highest frequency of output.
• Since the backward and forward N-gram are divided
into two different sets of N-gram, therefore, the
model can be used when the same word is found
afterwards.
• " "
– Backward 4-gram: input to predict " "
– Forward 4-gram: input to predict " "
6

Experiment: Building a LSTM Model
• Used the Word2vec to project text into the vector
space with 200 dimension
• Used LSTM with three layers of neural network
• Picked the highest score of softmax layer to
predict the word
• " "
– LSTM 2-gram: input to predict " "
– LSTM 3-gram: input to predict " "
– LSTM 4-gram: input to predict " "
7

The Modification of Correctness Rate
in N-gram Model
• 7-gram can achieve the best correction rate
8

The Modification of Correctness Rate in
Backward and Forward N-gram Model
• Backward and Forward 4-gram can achieve
the best correction rate
9

The Modification of Correctness Rate
in LSTM Model
• LSTM 6-gram can achieve the best correction
rate
•
10

Model The ratio of the
correct result of OCR
changes to the
wrong one
The ratio of making
the incorrect result
of OCR changes to
the right one
The ratio of
accuracy of OCR
and the text model
OCR X X 97.30%
7-gram 0.35% 13.06% 97.49%
LSTM 6-gram 0.1% 7.33% 97.5%
BF 4-gram 0.08% 9.54% 97.57%
Comparison of 7-gram, LSTM 6-gram
and BF 4-gram Text Models
• Backward and Forward 4-gram has the best
performance, with the lowest modification error
result and the highest correct results
11

Three Text models with
OCR Top 5 Candidate Words
• The OCR software we use is a Convolution Neural
Network model and to calculate the probability of
classification through softmax function
• When the probability of OCR Top 1 is lower than 95%,
it determines the word might be wrong and will use
mixed model
• Pick the word that has the highest score of the text
model also appeared in OCR Top 5 candidate words
12

Conclusion: Using Text Model
• N-gram, backward and forward N-gram or LSTM N-
gram text model can increase the ratio of accuracy of
OCR
• Backward and Forward 4-gram model has the lowest
modification error result and the highest correct
result
14

Conclusion: Mixing Text Models with
the Probability of OCR
• By mixing rules of OCR Top 5 candidate words
and probability of Top 1 with text model, it can
archive better result than using text model only
• Mixing the LSTM 6-gram with the probability of
OCR model has the highest correct results
15

1. The datasets for this project on question pair similarity come from Kaggle and include over 4 million question pairs split between train and test sets. 2. Several neural network architectures were implemented including CNNs, LSTMs, and bidirectional LSTMs with different word embeddings. 3. The best performing model was an LSTM with Glove word embeddings, which achieved a validation loss of 0.434 after 25 minutes of training. 4. Additional text preprocessing techniques like handling class imbalance and question symmetry were explored and improved the model performance further.

Learning from similarity and information extraction from structured documents...Infrrd

��

The document discusses challenges in extracting information from similar-looking structured documents, such as invoices, and explores deep learning techniques to improve extraction results. It evaluates various models, including siamese networks and one-shot learning, on a dataset of over 25,000 documents while testing for performance across multiple setups and architectures. The conclusion emphasizes the necessity of combining various architectural components for optimal performance and examines the implications of dataset size and generalization capabilities.

BL Demo Day - July2011 - (7) OCR Profiler and Post-CorrectionIMPACT Centre of Competence

��

The document presents the TR5 post-correction system developed at Ludwig-Maximilians-Universität München, which facilitates the efficient correction of OCR'd historical documents. It features customizable, user-friendly interfaces, error profiling technologies, and methodologies designed to significantly improve correction speeds. The evaluation results indicate that the system demonstrates high precision and recall for OCR error detection, highlighting its effectiveness in addressing systematic errors and recognizing historical language variants.

Postcorrection and profiler_bne_demodayIMPACT Centre of Competence

��

The document describes the TR5 Profiler and Post-Correction System developed by Ludwig-Maximilians-Universität München. The system uses innovative language technology to identify and present optical character recognition (OCR) errors in historical documents to enable efficient post-correction. It presents the recognized text and image snippets word-by-word for comparison, and proposes correction candidates based on analyses of the full text and errors. The underlying language technology models OCR output as the result of two "noisy channels" and exploits historical variant and OCR error patterns to rank correction candidates.

Bne demoday postcorrection_and_profilerIMPACT Centre of Competence

��

The document describes the TR5 post-correction system developed at Ludwig-Maximilians-Universität München, which provides a user-friendly interface for correcting OCR errors in historical documents. It incorporates innovative language technology to identify and propose corrections for various OCR errors, ultimately improving correction speed by analyzing text and error profiles. Evaluations demonstrate that the system performs significantly better than traditional methods, enabling faster and more precise post-correction of historical variants and systematic errors.

Off-line English Character Recognition: A Comparative Surveyidescitation

��

This document presents a comprehensive survey on offline English character recognition using neural networks, analyzing various methodologies and algorithms developed over the past few decades. It evaluates the effectiveness of models such as Support Vector Machines (SVM) and hybrids with Genetic Algorithms in improving recognition accuracy and computational efficiency. Additionally, the proposed modular multilayered neural network demonstrates high accuracy rates in recognizing both uppercase and lowercase handwritten characters.

Contribution of recurrent connectionist language models in improving lstm bas...anna8885

��

This paper proposes using recurrent connectionist language models to improve LSTM-based Arabic text recognition in videos. It trains RNN and RNNME language models on a large Arabic text corpus and integrates them into an LSTM-CTC optical character recognition system using a modified beam search decoding scheme. Experimental results show the connectionist language models outperform n-gram models, improving word recognition rate by over 16% compared to the baseline model without a language model. The full system also outperforms a commercial OCR engine by over 35% word recognition rate.

Automated Speech Recognition Pruthvij Thakar

��

This document describes the implementation of various neural network architectures for speech recognition using a dataset from VoxForge. It discusses preprocessing audio data into acoustic features, and implementing recurrent neural networks (RNNs), convolutional neural networks (CNNs), and combinations of CNNs and RNNs as acoustic models. Five models are implemented and evaluated: RNN with time-distributed dense layer; CNN plus RNN; deeper RNN; bidirectional RNN; and a custom architecture with CNN and deep RNN layers. The best performing model is selected for predicting speech from the test data.

T EXT M INING AND C LASSIFICATION OF P RODUCT R EVIEWS U SING S TRUCTURED S U...csandit

��

This document presents a study on text mining and multi-label classification of product reviews using a structured support vector machine (SSVM) approach. The authors discuss the challenges of text classification and detail the methodology, including preprocessing steps and the feature extraction process, which involves term frequency-inverse document frequency (tf-idf) and similarity matching. The system demonstrated an accuracy of 80.4% in classifying reviews from a dataset comprised of various electronic gadgets.

Telugu letters dataset and parallel deep convolutional neural network with a...International Journal of Reconfigurable and Embedded Systems

��

The document discusses the development of an automated Telugu character recognition (TCR) model utilizing a dataset of 645 Telugu letters processed by a parallel deep convolutional neural network optimized with stochastic gradient descent. Key processes include normalization, smoothing, and interpolation to enhance performance in recognizing Telugu characters, especially in real-time applications. It also addresses challenges like document quality and distortion that affect OCR systems' accuracy.

Handwritten Text Recognition and Translation with AudioIRJET Journal

��

This document presents research on handwritten text recognition and translation with audio. The researchers used convolutional neural networks to classify handwritten words and characters. For word classification, they directly classified whole words. For character classification, they first separated characters from words using an LSTM model and then classified each character. They trained models on the IAM Handwriting Dataset containing over 100k words. Pre-processing steps like noise removal, skew correction, and line segmentation were used to prepare images for classification. Both word-level and character-level classification models were explored. The character classification approach showed better results due to a smaller output size for the softmax layer. The recognized text could then be translated and output with audio.

International Journal on Natural Language Computing (IJNLC)basindavid68

��

The document reviews prompt-free few-shot text classification methods, emphasizing their performance and limitations using open-source pre-trained sentence transformers across various benchmark datasets. It outlines the potential of leveraging foundation models and notes that while traditional methods require extensive data, few-shot learning can be effective with minimal examples. The study evaluates multiple use cases including sentiment analysis and topic modeling, comparing their results with various language models to highlight their efficiency and applicability in real-world tasks.

A Review of Prompt-Free Few-Shot Text Classification Methodskevig

��

This document reviews prompt-free few-shot text classification methods, examining their performance and limitations while using open-source pre-trained sentence transformers. It discusses the challenges of filtering and categorizing feedback in various industries, highlights the advantages of leveraging foundation models, and describes empirical evaluations across multiple datasets for applications like sentiment analysis and emotion classification. The findings suggest that while traditional models require extensive data, few-shot approaches can achieve high accuracy even with limited examples, setting the stage for efficient model fine-tuning strategies.

A REVIEW OF PROMPT-FREE FEW-SHOT TEXT CLASSIFICATION METHODSkevig

��

This document reviews prompt-free few-shot text classification methods, emphasizing the use of open-source pre-trained sentence transformers to categorize text-based comments effectively. It highlights both the advantages and limitations of these methods through a comprehensive study across various benchmark datasets, including sentiment analysis and topic modeling, and suggests that prompting instruction-fine-tuned language models can yield better results for complex criteria. The paper also discusses the practical implications of leveraging foundation models to streamline text classification processes in various industries.

Session6 01.helmut schmidIMPACT Centre of Competence

��

The document presents the development of deep learning-based morphological taggers and lemmatizers specifically for annotating historical texts, with a focus on Middle High German. It discusses the challenges faced due to spelling and dialectal variations and showcases the architecture and performance of a novel neural network-based tagging system that outperforms traditional methods. The new tools aim to enhance the accuracy of part-of-speech tagging and lemmatization for historical corpora, with results indicating significant improvements in handling unseen words and morphological features.

Session7 03.katrien depuydtIMPACT Centre of Competence

��

The Nederlab project (2013-2018) aimed to create a diachronic corpus of Dutch language texts, spanning from 600 AD to the present, addressing metadata challenges associated with digitized material. It identifies key metadata requirements necessary for accurate research, such as provenance, author identification, genre classification, and version control. The document concludes by emphasizing the need for improved metadata models that cater to both library standards and diachronic research needs.

Session7 02.peter kiralyIMPACT Centre of Competence

��

The document discusses the validation and quality assessment of 126 million MARC records, detailing the history and structure of the MARC format which originated in the 1960s. It presents a quality assessment workflow including data ingestion, measurement, aggregation, and reporting, along with identifying issues in catalog records. Additionally, it highlights a project related to quality assessment using metrics, statistics of various library catalogs, and tools employed in the process.

Session6 04.giuseppe celanoIMPACT Centre of Competence

��

This document discusses the standoff annotation methodology for the Ancient Greek and Latin Dependency Treebank, outlining revisions and standardizations for improved annotation accuracy. It highlights the advantages and disadvantages of inline versus standoff annotation, as well as challenges in text extraction and tokenization. The document emphasizes the structure and components of the annotation system, including references to various texts and case studies in classical literature.

Session6 03.sandra youngIMPACT Centre of Competence

��

The document discusses the adaptation of lexicography techniques to evaluate nomenclature usage in biodiversity literature, focusing on the ambiguity between biological taxonomies and scientific nomenclature. It details methods used for corpus analysis, including the application of word sketches and graph transformations to identify hierarchical relations and common name disambiguation. The research aims to enhance understanding of naming conventions and proposes next steps for further evaluation against existing ontologies.

Session6 02.jeremi ochabIMPACT Centre of Competence

��

The document discusses the stylometry of literary papyri and addresses how to improve uncertain metadata, such as authorship and dating, using text extraction and data cleaning techniques. It describes methods for clustering texts through distance-based and community detection algorithms, along with their effectiveness and limitations. The conclusions emphasize the need for regularization in clustering and propose future directions including the use of n-grams and supervised machine learning to enhance textual analysis.

Session5 04.evangelos varthisIMPACT Centre of Competence

��

This document outlines the implementation of a databaseless web REST API for Migne's Patrologia Graeca, focusing on its ability to search unstructured texts. It compares RDBMS and NoSQL systems, describes the auto-transformation process of textual data into JSON files, and discusses the overall system architecture. Additionally, it highlights potential extensions for semantic enrichment and offers insight into future developments for the API.

Session5 03.george rehmIMPACT Centre of Competence

��

The document presents an overview of curation technologies aimed at enhancing cultural heritage archives through an interactive workbench, highlighting a project focused on the German reunification data set. It discusses the processing pipeline involving OCR, NER, and clustering for improved data access, and the goal is to create a user-friendly dashboard for intuitive analysis and exploration. Future work includes user studies, linking tools to original documents, and developing additional services.

Session5 02.tom derrickIMPACT Centre of Competence

��

The document discusses cross-disciplinary collaborations aimed at enhancing access to non-Western language materials, specifically focusing on South Asian printed books and Arabic manuscripts within the British Library's digital collections. It details challenges in optical character recognition (OCR) for Bengali and Arabic scripts, including unique characteristics of the scripts and results from various competitions and initiatives aimed at improving OCR accuracy. Future plans include further training of OCR software, as well as workshops in South Asia to continue enhancing the accessibility and usability of these historical texts.

Session5 01.rutger vankoertIMPACT Centre of Competence

��

The TRIADO project (2017-2019) aimed to enhance the accessibility and usability of archives by implementing digital methods for processing a sample of 13.8 meters of archival data. The findings indicated that OCR technology, particularly ABBYY, was effective with a word error rate of 15%, while auto-classification and date extraction showed promise despite a 20% error rate. Future steps include digitizing additional archives, linking with external data resources, and improving automatic transcription capabilities.

Session4 04.senka drobacIMPACT Centre of Competence

��

The document discusses efforts to improve Optical Character Recognition (OCR) of historical newspapers and journals in Finland, initially digitized by the National Library of Finland with an accuracy rate of about 90-91%. Using the open-source ocropy system, the authors aim to achieve a character accuracy rate above 98.5% by training models on Finnish and Swedish texts from 1771 to 1939. Future work includes enhancing the training data with additional Finnish antiqua samples and exploring deep neural networks to improve memory capacity.

Session3 04.arnau baroIMPACT Centre of Competence

��

The document presents an unsupervised method for automatically transcribing encoded manuscripts, aiming to enhance decryption of historical ciphers using interdisciplinary approaches from linguistics, computer science, and image processing. It outlines challenges, such as varying symbol alphabets and handwriting styles, and details a novel pipeline that incorporates preprocessing, clustering, and transcription of symbols. The results indicate the method's potential to reduce user intervention and improve transcription accuracy compared to traditional supervised methods.

Session3 03.christian clausnerIMPACT Centre of Competence

��

The report discusses the challenges and methods for extracting statistical information from 130 years of digitized Medical Officer of Health reports, which contains over 70,000 documents. Current practices reveal that standard OCR is inadequate for accurate data extraction, necessitating the development of automated solutions combined with advanced recognition techniques. Future work aims to enhance data accessibility and quality through improved table recognition algorithms and integrated data resources.

Session3 02.kimmo ketunnenIMPACT Centre of Competence

��

This document discusses the challenges and results of extracting articles from a large digitized Finnish historical newspaper collection (1771-1929) using the Pivaj software. The analysis revealed that while current algorithms achieve around 80-85% accuracy, the Pivaj system, particularly its offline application, showed promising results in learning to segment and label articles correctly. Despite some limitations, such as the handling of advertisement-heavy pages, Pivaj outperformed other systems, including Docworks, in identifying articles within the newspaper pages.

Session3 01.clemens neudeckerIMPACT Centre of Competence

��

The OCR-D project is an open-source framework aimed at enhancing OCR capabilities for historical printed documents, leveraging advancements in artificial intelligence to meet the growing demand for high-quality text corpora. The project, funded until 2020, involves a coordination project and eight specific modules that focus on various OCR-related tasks such as image optimization, layout analysis, and automated postcorrection. Documentation and specifications are openly available, supporting sustainable and reproducible workflows for the digital humanities community.

Session2 04.ashkan ashkpourIMPACT Centre of Competence

��

- The document describes a project to fill gaps in knowledge about diamond mining, trading, and polishing in Borneo by developing a workflow using various CLARIAH tools and resources. - The workflow involved digitizing a diamond encyclopedia, extracting concepts and place names, linking the data to external sources to create linked open data, and querying newspaper archives to build a corpus of relevant articles. - Promising results showed mining, trading, and polishing continued in Borneo for Southeast Asian customers, and described previously unknown diamond fields and polishing locations in Borneo. The project aims to apply the workflow to other commodities like sugar.

More Related Content

Similar to Session1 03.hsian-an wang (6)

T EXT M INING AND C LASSIFICATION OF P RODUCT R EVIEWS U SING S TRUCTURED S U...csandit

��

Telugu letters dataset and parallel deep convolutional neural network with a...International Journal of Reconfigurable and Embedded Systems

��

Handwritten Text Recognition and Translation with AudioIRJET Journal

��

International Journal on Natural Language Computing (IJNLC)basindavid68

��

A Review of Prompt-Free Few-Shot Text Classification Methodskevig

��

A REVIEW OF PROMPT-FREE FEW-SHOT TEXT CLASSIFICATION METHODSkevig

��

T EXT M INING AND C LASSIFICATION OF P RODUCT R EVIEWS U SING S TRUCTURED S U...csandit

��

Telugu letters dataset and parallel deep convolutional neural network with a...International Journal of Reconfigurable and Embedded Systems

��

Handwritten Text Recognition and Translation with AudioIRJET Journal

��

International Journal on Natural Language Computing (IJNLC)basindavid68

��

A Review of Prompt-Free Few-Shot Text Classification Methodskevig

��

A REVIEW OF PROMPT-FREE FEW-SHOT TEXT CLASSIFICATION METHODSkevig

��

More from IMPACT Centre of Competence (20)

Session6 01.helmut schmidIMPACT Centre of Competence

��

Session7 03.katrien depuydtIMPACT Centre of Competence

��

Session7 02.peter kiralyIMPACT Centre of Competence

��

Session6 04.giuseppe celanoIMPACT Centre of Competence

��

Session6 03.sandra youngIMPACT Centre of Competence

��

Session6 02.jeremi ochabIMPACT Centre of Competence

��

Session5 04.evangelos varthisIMPACT Centre of Competence

��

Session5 03.george rehmIMPACT Centre of Competence

��

Session5 02.tom derrickIMPACT Centre of Competence

��

Session5 01.rutger vankoertIMPACT Centre of Competence

��

Session4 04.senka drobacIMPACT Centre of Competence

��

Session3 04.arnau baroIMPACT Centre of Competence

��

Session3 03.christian clausnerIMPACT Centre of Competence

��

Session3 02.kimmo ketunnenIMPACT Centre of Competence

��

Session3 01.clemens neudeckerIMPACT Centre of Competence

��

Session2 04.ashkan ashkpourIMPACT Centre of Competence

��

Session2 03.juri opitzIMPACT Centre of Competence

��

The document discusses the automatic reconstruction of emperor itineraries from the Regesta Imperii, a collection of over 150,000 historical records. It addresses challenges in accurately mapping place names and coordinates from nearly a millennium of history, detailings methods used for place name and coordinate prediction, including logistic regression and itinerary modeling. Future work aims to improve these predictions through enhanced models and historical data sources.

Session2 02.christian reulIMPACT Centre of Competence

��

The document discusses a case study on the automatic semantic text tagging of Daniel Sanders' historical lexicon using optical character recognition (OCR) and typography classification. It highlights the methods for labeling typography classes, ground truth production, and the effectiveness of combining OCR outputs to improve recognition accuracy. The study aims to create a workflow for digitizing historical lexica and suggests further experimentation with different typographical attributes and OCR models.

Session2 01.emad mohamedIMPACT Centre of Competence

��

This document describes the SOS system for segmenting, stemming, and standardizing Arabic text. It presents the challenges of processing Arabic cultural heritage texts which contain orthographic variations. The system uses gradient boosting machines and achieves state-of-the-art performance on segmentation and derives stemming as a byproduct. It also standardizes orthography with high accuracy, which further improves segmentation. The system addresses issues like hamza forms and letter confusions that previous systems did not handle well.

Session1 04.florian finkIMPACT Centre of Competence

��

The document presents the a-i-pocoto system, which integrates automated and interactive OCR post-correction methods for improving the accuracy of OCR results on historical documents. It details the process of automatic post-correction using supervised machine learning, involving multiple OCRs and profiling for error detection, and describes an interactive tool known as Pocoto for manual corrections. Evaluation results indicate improvements in OCR word accuracy through various experimental setups, though certain steps like lexicon extension showed limited benefits and suggested training adjustments for improving decision-making in corrections.

Session6 01.helmut schmidIMPACT Centre of Competence

��

Session7 03.katrien depuydtIMPACT Centre of Competence

��

Session7 02.peter kiralyIMPACT Centre of Competence

��

Session6 04.giuseppe celanoIMPACT Centre of Competence

��

Session6 03.sandra youngIMPACT Centre of Competence

��

Session6 02.jeremi ochabIMPACT Centre of Competence

��

Session5 04.evangelos varthisIMPACT Centre of Competence

��

Session5 03.george rehmIMPACT Centre of Competence

��

Session5 02.tom derrickIMPACT Centre of Competence

��

Session5 01.rutger vankoertIMPACT Centre of Competence

��

Session4 04.senka drobacIMPACT Centre of Competence

��

Session3 04.arnau baroIMPACT Centre of Competence

��

Session3 03.christian clausnerIMPACT Centre of Competence

��

Session3 02.kimmo ketunnenIMPACT Centre of Competence

��

Session3 01.clemens neudeckerIMPACT Centre of Competence

��

Session2 04.ashkan ashkpourIMPACT Centre of Competence

��

Session2 03.juri opitzIMPACT Centre of Competence

��

Session2 02.christian reulIMPACT Centre of Competence

��

Session2 01.emad mohamedIMPACT Centre of Competence

��

Session1 04.florian finkIMPACT Centre of Competence

��

Recently uploaded (20)

FIDO Seminar: Authentication for a Billion Consumers - Amazon.pptxFIDO Alliance

��

UserCon Belgium: Honey, VMware increased my billstijn40

��

VMware’s pricing changes have forced organizations to rethink their datacenter cost management strategies. While FinOps is commonly associated with cloud environments, the FinOps Foundation has recently expanded its framework to include Scopes—and Datacenter is now officially part of the equation. In this session, we’ll map the FinOps Framework to a VMware-based datacenter, focusing on cost visibility, optimization, and automation. You’ll learn how to track costs more effectively, rightsize workloads, optimize licensing, and drive efficiency—all without migrating to the cloud. We’ll also explore how to align IT teams, finance, and leadership around cost-aware decision-making for on-prem environments. If your VMware bill keeps increasing and you need a new approach to cost management, this session is for you!

9-1-1 Addressing: End-to-End Automation Using FMESafe Software

��

This session will cover a common use case for local and state/provincial governments who create and/or maintain their 9-1-1 addressing data, particularly address points and road centerlines. In this session, you'll learn how FME has helped Shelby County 9-1-1 (TN) automate the 9-1-1 addressing process; including automatically assigning attributes from disparate sources, on-the-fly QAQC of said data, and reporting. The FME logic that this presentation will cover includes: Table joins using attributes and geometry, Looping in custom transformers, Working with lists and Change detection.

The Future of AI Agent Development Trends to Watch.pptxLisa ward

��

cnc-processing-centers-centateq-p-110-en.pdfAmirStern2

��

מרכז עיבודים תעשייתי בעל 3/4/5 צירים, עד 22 החלפות כלים עם כל אפשרויות העיבוד הדרושות.��בעל שטח עבודה גדול ומחשב נוח וקל להפעלה בשפה העברית/רוסית/אנגלית/ספרדית/ערבית ועוד.. מסוגל לבצע פעולות עיבוד שונות המתאימות לענפים שונים: קידוח אנכי, אופקי, ניסור, וכרסום אנכי.

War_And_Cyber_3_Years_Of_Struggle_And_Lessons_For_Global_Security.pdfbiswajitbanerjee38

��

Russia is one of the most aggressive nations when it comes to state coordinated cyberattacks — and Ukraine has been at the center of their crosshairs for 3 years. This report, provided the State Service of Special Communications and Information Protection of Ukraine contains an incredible amount of cybersecurity insights, showcasing the coordinated aggressive cyberwarfare campaigns of Russia against Ukraine. It brings to the forefront that understanding your adversary, especially an aggressive nation state, is important for cyber defense. Knowing their motivations, capabilities, and tactics becomes an advantage when allocating resources for maximum impact. Intelligence shows Russia is on a cyber rampage, leveraging FSB, SVR, and GRU resources to professionally target Ukraine’s critical infrastructures, military, and international diplomacy support efforts. The number of total incidents against Ukraine, originating from Russia, has steadily increased from 1350 in 2021 to 4315 in 2024, but the number of actual critical incidents has been managed down from a high of 1048 in 2022 to a mere 59 in 2024 — showcasing how the rapid detection and response to cyberattacks has been impacted by Ukraine’s improved cyber resilience. Even against a much larger adversary, Ukraine is showcasing outstanding cybersecurity, enabled by strong strategies and sound tactics. There are lessons to learn for any enterprise that could potentially be targeted by aggressive nation states. Definitely worth the read!

CapCut Pro Crack For PC Latest Version {Fully Unlocked} 2025pcprocore

��

👉𝗡𝗼𝘁𝗲:𝗖𝗼𝗽𝘆 𝗹𝗶𝗻𝗸 & 𝗽𝗮𝘀𝘁𝗲 𝗶𝗻𝘁𝗼 𝗚𝗼𝗼𝗴𝗹𝗲 𝗻𝗲𝘄 𝘁𝗮𝗯> https://pcprocore.com/ 👈◀ CapCut Pro Crack is a powerful tool that has taken the digital world by storm, offering users a fully unlocked experience that unleashes their creativity. With its user-friendly interface and advanced features, it’s no wonder why aspiring videographers are turning to this software for their projects.

GenAI Opportunities and Challenges - Where 370 Enterprises Are Focusing Now.pdfPriyanka Aash

��

Smarter Aviation Data Management: Lessons from Swedavia Airports and SwecoSafe Software

��

Managing airport and airspace data is no small task, especially when you’re expected to deliver it in AIXM format without spending a fortune on specialized tools. But what if there was a smarter, more affordable way? Join us for a behind-the-scenes look at how Sweco partnered with Swedavia, the Swedish airport operator, to solve this challenge using FME and Esri. Learn how they built automated workflows to manage periodic updates, merge airspace data, and support data extracts – all while meeting strict government reporting requirements to the Civil Aviation Administration of Sweden. Even better? Swedavia built custom services and applications that use the FME Flow REST API to trigger jobs and retrieve results – streamlining tasks like securing the quality of new surveyor data, creating permdelta and baseline representations in the AIS schema, and generating AIXM extracts from their AIS data. To conclude, FME expert Dean Hintz will walk through a GeoBorders reading workflow and highlight recent enhancements to FME’s AIXM (Aeronautical Information Exchange Model) processing and interpretation capabilities. Discover how airports like Swedavia are harnessing the power of FME to simplify aviation data management, and how you can too.

FIDO Seminar: Evolving Landscape of Post-Quantum Cryptography.pptxFIDO Alliance

��

"Database isolation: how we deal with hundreds of direct connections to the d...Fwdays

��

What can go wrong if you allow each service to access the database directly? In a startup, this seems like a quick and easy solution, but as the system scales, problems appear that no one could have guessed. In my talk, I'll share Solidgate's experience in transforming its architecture: from the chaos of direct connections to a service-based data access model. I will talk about the transition stages, bottlenecks, and how isolation affected infrastructure support. I will honestly show what worked and what didn't. In short, we will analyze the controversy of this talk.

FIDO Seminar: Targeting Trust: The Future of Identity in the Workforce.pptxFIDO Alliance

��

Creating Inclusive Digital Learning with AI: A Smarter, Fairer FutureImpelsys Inc.

��

Have you ever struggled to read a tiny label on a medicine box or tried to navigate a confusing website? Now imagine if every learning experience felt that way—every single day. For millions of people living with disabilities, poorly designed content isn’t just frustrating. It’s a barrier to growth. Inclusive learning is about fixing that. And today, AI is helping us build digital learning that’s smarter, kinder, and accessible to everyone. Accessible learning increases engagement, retention, performance, and inclusivity for everyone. Inclusive design is simply better design.

Improving Data Integrity: Synchronization between EAM and ArcGIS Utility Netw...Safe Software

��

Utilities and water companies play a key role in the creation of clean drinking water. The creation and maintenance of clean drinking water is becoming a critical problem due to pollution and pressure on the environment. A lot of data is necessary to create clean drinking water. For fieldworkers, two types of data are key: Asset data in an asset management system (EAM for example) and Geographic data in a GIS (ArcGIS Utility Network ). Keeping this type of data up to date and in sync is a challenge for many organizations, leading to duplicating data and creating a bulk of extra attributes and data to keep everything in sync. Using FME, it is possible to synchronize Enterprise Asset Management (EAM) data with the ArcGIS Utility Network in real time. Changes (creation, modification, deletion) in ArcGIS Pro are relayed to EAM via FME, and vice versa. This ensures continuous synchronization of both systems without daily bulk updates, minimizes risks, and seamlessly integrates with ArcGIS Utility Network services. This presentation focuses on the use of FME at a Dutch water company, to create a sync between the asset management and GIS.

ReSTIR [DI]: Spatiotemporal reservoir resampling for real-time ray tracing ...revolcs10

��

2025_06_18 - OpenMetadata Community Meeting.pdfOpenMetadata

��

The community meetup was held Wednesday June 18, 2025 @ 9:00 AM PST. Catch the next OpenMetadata Community Meetup @ https://www.meetup.com/openmetadata-meetup-group/ In this month's OpenMetadata Community Meetup, "Enforcing Quality & SLAs with OpenMetadata Data Contracts," we covered data contracts, why they matter, and how to implement them in OpenMetadata to increase the quality of your data assets! Agenda Highlights: 👋 Introducing Data Contracts: An agreement between data producers and consumers 📝 Data Contracts key components: Understanding a contract and its purpose 🧑‍🎨 Writing your first contract: How to create your own contracts in OpenMetadata 🦾 An OpenMetadata MCP Server update! ➕ And More!

Curietech AI in action - Accelerate MuleSoft developmentshyamraj55

��

CurieTech AI in Action – Accelerate MuleSoft Development Overview: This presentation demonstrates how CurieTech AI’s purpose-built agents empower MuleSoft developers to create integration workflows faster, more accurately, and with less manual effort linkedin.com +12 curietech.ai +12 meetups.mulesoft.com +12 . Key Highlights: Dedicated AI agents for every stage: Coding, Testing (MUnit), Documentation, Code Review, and Migration curietech.ai +7 curietech.ai +7 medium.com +7 DataWeave automation: Generate mappings from tables or samples—95%+ complete within minutes linkedin.com +7 curietech.ai +7 medium.com +7 Integration flow generation: Auto-create Mule flows based on specifications—speeds up boilerplate development curietech.ai +1 medium.com +1 Efficient code reviews: Gain intelligent feedback on flows, patterns, and error handling youtube.com +8 curietech.ai +8 curietech.ai +8 Test & documentation automation: Auto-generate MUnit test cases, sample data, and detailed docs from code curietech.ai +5 curietech.ai +5 medium.com +5 Why Now? Achieve 10× productivity gains, slashing development time from hours to minutes curietech.ai +3 curietech.ai +3 medium.com +3 Maintain high accuracy with code quality matching or exceeding manual efforts curietech.ai +2 curietech.ai +2 curietech.ai +2 Ideal for developers, architects, and teams wanting to scale MuleSoft projects with AI efficiency Conclusion: CurieTech AI transforms MuleSoft development into an AI-accelerated workflow—letting you focus on innovation, not repetition.

PyCon SG 25 - Firecracker Made Easy with Python.pdfMuhammad Yuga Nugraha

��

Explore the ease of managing Firecracker microVM with the firecracker-python. In this session, I will introduce the basics of Firecracker microVM and demonstrate how this custom SDK facilitates microVM operations easily. We will delve into the design and development process behind the SDK, providing a behind-the-scenes look at its creation and features. While traditional Firecracker SDKs were primarily available in Go, this module brings a simplicity of Python to the table.

10 Key Challenges for AI within the EU Data Protection Framework.pdfPriyanka Aash

��

Techniques for Automatic Device Identification and Network Assignment.pdfPriyanka Aash

��

FIDO Seminar: Authentication for a Billion Consumers - Amazon.pptxFIDO Alliance

��

UserCon Belgium: Honey, VMware increased my billstijn40

��

9-1-1 Addressing: End-to-End Automation Using FMESafe Software

��

The Future of AI Agent Development Trends to Watch.pptxLisa ward

��

cnc-processing-centers-centateq-p-110-en.pdfAmirStern2

��

War_And_Cyber_3_Years_Of_Struggle_And_Lessons_For_Global_Security.pdfbiswajitbanerjee38

��

CapCut Pro Crack For PC Latest Version {Fully Unlocked} 2025pcprocore

��

GenAI Opportunities and Challenges - Where 370 Enterprises Are Focusing Now.pdfPriyanka Aash

��

Smarter Aviation Data Management: Lessons from Swedavia Airports and SwecoSafe Software

��

FIDO Seminar: Evolving Landscape of Post-Quantum Cryptography.pptxFIDO Alliance

��

"Database isolation: how we deal with hundreds of direct connections to the d...Fwdays

��

FIDO Seminar: Targeting Trust: The Future of Identity in the Workforce.pptxFIDO Alliance

��

Creating Inclusive Digital Learning with AI: A Smarter, Fairer FutureImpelsys Inc.

��

Improving Data Integrity: Synchronization between EAM and ArcGIS Utility Netw...Safe Software

��

ReSTIR [DI]: Spatiotemporal reservoir resampling for real-time ray tracing ...revolcs10

��

2025_06_18 - OpenMetadata Community Meeting.pdfOpenMetadata

��

Curietech AI in action - Accelerate MuleSoft developmentshyamraj55

��

PyCon SG 25 - Firecracker Made Easy with Python.pdfMuhammad Yuga Nugraha

��

10 Key Challenges for AI within the EU Data Protection Framework.pdfPriyanka Aash

��

Techniques for Automatic Device Identification and Network Assignment.pdfPriyanka Aash

��

Session1 03.hsian-an wang

1. Towards a Higher Accuracy of Optical Character Recognition of Chinese Rare Books in Making Use of Text Model Hsiang-An Wang Academia Sinica Center for Digital Cultures

2. Ink Bleed and Pool Quality 2

3. Limitation (Missing and Extra Word) OCR Original OCR Original 3

4. Experiment: Data Collection • Training dataset: 187 ancient medicine books from the Scripta Sinica Database (about 40 million words) • Testing dataset: 1 relevant ancient medicine book named “ ” with a total of 185,000 words • The OCR results are about 180,000 words correct and about 5000 incorrect words, which means the correct rate is about 97.3 % 4

5. Experiment: Building a N-gram Model • Relied on the sequence of words in the training dataset, and thus we picked the highest frequency of output. • " " – 2-gram: input to predict " " – 3-gram: input predict " " – 4-gram: input predict " " – ... 5

6. Experiment: Building a Backward and Forward N-gram Model • Relied on the sequence of backward and forward words in the training dataset, and thus we picked the highest frequency of output. • Since the backward and forward N-gram are divided into two different sets of N-gram, therefore, the model can be used when the same word is found afterwards. • " " – Backward 4-gram: input to predict " " – Forward 4-gram: input to predict " " 6

7. Experiment: Building a LSTM Model • Used the Word2vec to project text into the vector space with 200 dimension • Used LSTM with three layers of neural network • Picked the highest score of softmax layer to predict the word • " " – LSTM 2-gram: input to predict " " – LSTM 3-gram: input to predict " " – LSTM 4-gram: input to predict " " 7

8. The Modification of Correctness Rate in N-gram Model • 7-gram can achieve the best correction rate 8

9. The Modification of Correctness Rate in Backward and Forward N-gram Model • Backward and Forward 4-gram can achieve the best correction rate 9

10. The Modification of Correctness Rate in LSTM Model • LSTM 6-gram can achieve the best correction rate • 10

11. Model The ratio of the correct result of OCR changes to the wrong one The ratio of making the incorrect result of OCR changes to the right one The ratio of accuracy of OCR and the text model OCR X X 97.30% 7-gram 0.35% 13.06% 97.49% LSTM 6-gram 0.1% 7.33% 97.5% BF 4-gram 0.08% 9.54% 97.57% Comparison of 7-gram, LSTM 6-gram and BF 4-gram Text Models • Backward and Forward 4-gram has the best performance, with the lowest modification error result and the highest correct results 11

12. Three Text models with OCR Top 5 Candidate Words • The OCR software we use is a Convolution Neural Network model and to calculate the probability of classification through softmax function • When the probability of OCR Top 1 is lower than 95%, it determines the word might be wrong and will use mixed model • Pick the word that has the highest score of the text model also appeared in OCR Top 5 candidate words 12

13. Model The ratio of the correct result of OCR changes to the wrong one The ratio of making the incorrect result of OCR changes to the right one The ratio of accuracy of OCR and the text model OCR X X 97.30% 7-gram 0.012% 9% 97.63% LSTM 6-gram 0.13% 16% 97.71% BF 4-gram 0.009% 5.92% 97.55% Comparison of Three Text Models Mixed with the Probability of OCR • LSTM 6-gram mixed with the probability of OCR that has the best performance 13

14. Conclusion: Using Text Model • N-gram, backward and forward N-gram or LSTM N- gram text model can increase the ratio of accuracy of OCR • Backward and Forward 4-gram model has the lowest modification error result and the highest correct result 14

15. Conclusion: Mixing Text Models with the Probability of OCR • By mixing rules of OCR Top 5 candidate words and probability of Top 1 with text model, it can archive better result than using text model only • Mixing the LSTM 6-gram with the probability of OCR model has the highest correct results 15

16. Thank you for listening

�ݺ�ߣ

Session1 03.hsian-an wang

Recommended

More Related Content

Similar to Session1 03.hsian-an wang (6)

More from IMPACT Centre of Competence (20)

Recently uploaded (20)

Session1 03.hsian-an wang