Intelligence artificielle et données
488 documents à télécharger gratuitement
Cours, examens, TD, TP et exercices de intelligence artificielle et données. Thèmes couverts : machine learning, apprentissage automatique, big data, data science, fouille de données.
This document explores the development of a business intelligence solution for the Arab Tunisian Bank (ATB). The methodology integrates SCRUM, an agile framework for iterative project management, to design and implement a data warehouse and decision-support tools. Key tools include SQL Server Integration Services (SSIS) for ETL processes, alongside comparative evaluation of ROLAP, MOLAP, and HOLAP. The solution emphasizes risk classification and reporting through dynamic dashboards, enhancing ATB’s strategic decision-making capabilities.
This document outlines the conceptualization and development of the Magpie Business Intelligence platform. It begins with fundamental Business Intelligence concepts, comparing multiple open-source tools for Extract, Transform, Load (ETL), data warehouses, and reporting generators to select the most efficient components. The design phase focuses on user roles and the system's dashboard construction, emphasizing data indicators and information collection. The implementation leverages environments such as SQL Server, MySQL, Eclipse Indigo, and Tomcat to develop a robust and scalable platform,...
This document presents the development of a decision-support system for the Arab Tunisian Bank (ATB), as part of an undergraduate applied Informatics degree. The project utilizes the SCRUM methodology to structure the creation of a data warehouse, establish an ETL pipeline, and implement tools for data analysis and reporting. The project emphasizes integrating predictive and agile methodologies, comparing Business Intelligence solutions, and designing a risk classification DataMart to facilitate informed decision-making. Key outcomes include the successful deployment of a comprehensive BI p...
This document introduces SQL Server Integration Services (SSIS) for building ETL (Extract, Transform, Load) processes within the domain of Business Intelligence. It explains the role of ETLs in data movement and transformation between databases, stressing their importance in decision-making environments. The document also provides a comprehensive walkthrough of the SSIS workspace, including its components such as project creation, solution explorer, control flow, data flow, and event management features. Finally, it establishes a foundation for practical learning by detailing the tools and...
This document provides an introductory course on SQL Server Integration Services (SSIS) within the realm of Business Intelligence (BI). It explains ETL (Extract, Transform, Load) processes as a data-centric methodology, covering data extraction, transformation, and loading into decision-support storage structures. The course introduces SQL Server 2008 Integration Services tools and environments, including the Business Intelligence Development Studio, emphasizing features like control flow, data flow, and connection management. It aims to equip readers for practical SSIS use and certificatio...
This project focused on creating a Business Intelligence solution for the Tunisian insurance company STAR. The methodology involved employing the Scrum BI framework and a Bottom-Up approach to construct data marts by processing and cleaning data using ETL tools. The solution created a dynamic reporting infrastructure via Power BI and a custom web platform to address inefficiencies, such as data access issues, reporting complexity, and limited BI tools. Future work includes implementing additional insurance branches and advanced data mining algorithms.
This document outlines the structured process for validating and assessing practical internships for students enrolled in the Applied License in Decision Informatics at the Institut Supérieur de Gestion, University of Tunis. It specifies details regarding the student's profile, hosting organization, responsibilities, weekly progress, and professional performance observations. The document also includes sections for evaluating the student's adaptation, communication, initiative, and technical competencies, culminating in validation criteria and a final grade assigned by institutional and org...
This document introduces the concept of Business Intelligence (BI) or Decision Support Systems (DSS), focusing on their growing importance since the 1990s. It explains how BI systems enable companies to extract, transform, and store data in a coherent manner for effective decision-making. Methodologies such as dimensional modeling for data normalization, utilization of ETL tools, and OLAP for multidimensional analysis are emphasized. The tutorial also covers system architecture components, including data warehouses, data marts, and visualization tools essential for presenting actionable ins...
This document discusses the conception of a data warehouse within the context of business intelligence. It introduces fundamental data warehousing concepts such as dimensions, facts, and ETL processes, alongside advanced modeling techniques like star and snowflake schemas. Through case studies and examples, it explains how to design efficient data marts and full-fledged data warehouses, concluding with a critique of existing methodologies and indicators for successful design. The aim is to ensure strategic decision-making through robust data analytics infrastructure.
This article, authored by Yazid Grim, explores foundational concepts of OLAP (Online Analytical Processing), a critical technology in Business Intelligence. It provides an in-depth explanation of OLAP's purpose, differences from OLTP systems, strengths, and limitations. An example illustrates how OLAP facilitates complex, multi-dimensional data analysis for strategic business decisions. The article further discusses the evolution of OLAP definitions, its technical requirements, and the FASMI (Fast, Analysis, Server, Multidimensional, Information) criteria, emphasizing its role in transformi...
This document by Mohamed Taslimanka Sylla and Fleur-Anne Blain guides readers through best practices for designing and managing Business Intelligence (BI) projects. It emphasizes a methodological approach inspired by Kimball's dimensional lifecycle, covering theoretical principles like agile practices, business requirement definition, and dimensional modeling. The practical section demonstrates a case study involving a university's challenges with process inefficiencies, applying SQL analysis and interviews to design an optimized BI solution. The document highlights the importance of unders...
This project involves the analysis and visualization of sales data using Power BI and Power Query. The objective is to create various hierarchies and measures to evaluate sales performance across different dimensions and time periods. The final output will include pivot tables and calculated measures demonstrating sales trends and comparisons.
The paper introduces a novel three-polar surface representation constructed by summing geodesic potentials generated from three reference points on a 3D surface. This method enhances stability under errors in reference point positioning compared to conventional unipolar schemes. Through theoretical validation and experiments, the three-polar representation demonstrates superior accuracy and stability for 3D human face descriptions using the Hausdorff shape distance metric. A comparative study using the Stanford Bunny model and the Bosphorus 3D face dataset confirms its robustness and perfor...
This document is a collaboration by first-year Master’s students at the National School of Computer Sciences, supervised by Professor Faouzi Ghorbel. The work summarizes courses on computer vision, 3D techniques, stereoscopic calibration, geometric moments for shape recognition, the expectation-maximization algorithm, and dimensionality reduction using Fisher's method. The methodologies cover camera calibration, mathematical principles in image geometric transformations, algorithms for pattern recognition, and statistical approaches to data dimensionality reduction. The document provides th...
The document introduces the fundamental concepts of 2D and 3D image representation and processing. It distinguishes between low-level image processing (e.g., noise reduction and contrast enhancement) and high-level image analysis (e.g., information extraction). Key methodologies for 3D image representation, including explicit, implicit, parametric, and polygonal models, are discussed, as well as their applications in fields like archeology, medicine, and biometrics. Finally, the limitations of 2D imaging, such as pose and illumination issues, are addressed, with solutions provided by 3D ima...
This book serves as an introduction to data science concepts and applications. It lays out the core principles of the data science process and offers insights into data engineering and the big data landscape. The reader will also find practical advice for leveraging data science skills in the job market.
This document outlines the syllabus for the TERI (Traitement et Reconnaissance d'Images) course, aimed at providing a foundational understanding of image processing and pattern recognition. Core topics include discrete image representation, filtering, segmentation, and recognition, while integrating advanced topics such as artificial intelligence and visual perception mechanisms. Practical sessions include machine labs and hands-on exercises to reinforce theoretical concepts. The syllabus also covers industrial applications and advanced mathematical models like convolution, Fourier transfor...
This thesis explores human segmentation in images and videos, focusing on two novel methods. The first method builds upon Histogram of Oriented Gradients (HOG) descriptors combined with Support Vector Machine (SVM) classifiers to detect human contours and reconstruct silhouettes through optimal graph-based pathways. The second method involves interactive graph cutting guided by silhouette templates, which are then adapted through part-based templates tailored to human posture. Both methods extend to video segmentation by leveraging additional temporal data, improving automatic trimap genera...
This paper presents a novel projection-based algorithm to segment Arabic text into lines, words, and characters. The proposed method utilizes profile amplitude filtering and edge detection techniques to handle the unique challenges posed by the cursive nature and overlapping characters in Arabic script. The system processes input images through line, word, and character segmentation stages, effectively addressing skew correction and character shape complexities. The algorithm demonstrates promising results on various printed documents with different Arabic fonts, achieving robust line, word...
This paper presents a hierarchical texture segmentation method based on wavelet decomposition. Initially, a coarse segmentation at the highest resolution is performed using texture prototypes identified via a fuzzy classifier. Ambiguous pixels are identified through neighborhood analysis and refined progressively by deferring their classification to lower resolution levels. The wavelet-based hierarchical approach ensures accurate texture differentiation, leveraging shape factors derived from multiple resolution levels for improved classification.
The document analyzes monthly consumption patterns using statistical methods such as trend calculations and cyclic coefficients. It computes seasonal adjustments and predicts current year consumption based on prior data. Key results include the average monthly and annual consumption, forecasted values for individual months, and percentage cyclic variation. These findings support strategic decision-making around resource allocation and consumption management.
The document outlines an estimation process for the current year's consumption of an unspecified article based on the previous year’s monthly data and related percentage adjustments (Rn values). The data provided includes explicit monthly consumption figures and corresponding variation percentages. The methodology involves predicting future values by applying these percentages to the prior year's consumption data, with a specific rounding requirement for the final predicted values. This serves as a mathematical and statistical exercise in forecasting based on historical data trends.
This document describes advancements in the Tesseract OCR engine through the integration of both legacy and neural network-based LSTM models, enabling support for over 100 languages. It details the methodologies for training using synthetic datasets, including procedures such as Unicode normalization, grapheme encoding, and realistic text rendering using thousands of fonts. Performance testing reveals significant accuracy improvements, particularly when combining neural networks with language models, although challenges remain with specific languages. The overall conclusion highlights the r...
Tesseract OCR is an open-source engine developed by HP between 1984 and 1994. The architecture leverages novel features like adaptive classification, two-pass word recognition, and efficient handling of white-on-black text. It includes unique line-finding and baseline-fitting techniques, enabling it to handle skewed and curved text with high accuracy. Despite its limitations in linguistic analysis, Tesseract's efficient workflows and compact training models established its superior accuracy over commercial OCR systems of its time.