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 introduces clustering methods, focusing on hierarchical classification and partitioning. It discusses the importance of similarity metrics in clustering, particularly the choice between Euclidean and Manhattan distances. The document also outlines the steps involved in hierarchical clustering and the concept of nested partitions.
This document presents an example of hierarchical classification on principal components (HCPC) using climate data from various European capitals. It aims to categorize capitals based on their temperature similarities throughout the year, utilizing statistical metrics and graphical representations. The analysis employs principal component analysis and clustering to unveil insights about data structure.
This document discusses various clustering methodologies, including hierarchical clustering and partitioning methods. It also emphasizes the importance of similarity measures in clustering, presenting metrics such as Euclidean and Manhattan distances. The text provides a detailed overview of hierarchical agglomerative clustering, highlighting its process and applications.
This thesis focuses on developing a decision-support system within a banking context, specifically for the Banque Internationale Arabe de Tunisie (BIAT). It employs modern Business Intelligence practices to design dashboards for the aggregation, normalization, and visualization of data. Using Ralph Kimball's methodology, the data warehouse architecture was meticulously structured to meet functional and non-functional requirements. The implementation achieved real-time data insights critical for the bank's marketing and digital development divisions, enabling enhanced decision-making and ope...
This document outlines the analysis, design, and implementation phases of building a business intelligence solution for Société Tunisienne d’Assurances et de Réassurances (STAR). It employs Scrum methodology tailored to business intelligence projects, creating data marts for monitoring production and claims. Data was extracted, transformed, and loaded (ETL) into a warehouse, followed by creating insightful dashboards and reports using Power BI. The solution improves operational reporting, performance tracking, and enables faster decision-making processes.
This document highlights the development of a decision support system for the Arab Tunisian Bank (ATB). Agile methodologies, particularly SCRUM, were employed in conjunction with Business Intelligence tools to create an optimized DataWarehouse and Datamart for analyzing client risk and asset classifications. The solution involved designing conceptual models, exploring ROLAP, MOLAP, and HOLAP approaches, and implementing Extract-Transform-Load (ETL) processes. The outcomes included enhanced data management and reporting capabilities for better financial risk assessment and client monitoring.
This report documents the development of a project aimed at generating calculations of mobile data consumption using advanced tools and methodologies. The authors evaluate existing organizational practices in OOREDOO Tunisie, propose solutions based on SCRUM BI methodology, and employ a robust Data Warehouse using tools such as Talend and Oracle Database. Key outcomes include cleaning, transforming, and integrating large data into an optimized structure, culminating in the creation of interactive dashboards for visualizing subscriber data and mobile data consumption trends.
This document details the development of a solution for estimating the prices of used cars in Tunisia. It follows an agile methodology using SCRUM, incorporating machine learning for price prediction. Core tasks include web scraping from various Tunisian car advertisement websites, data cleaning, and the implementation of linear regression and gradient regression models. The project concludes with the centralization of data in a SQL database and the development of a web interface for user interaction.
This document presents a final year project focused on the study, design, and implementation of a business intelligence dashboard system using Big Data technologies. The project uses tools like Apache Hadoop, Spark, Hive, and Arcadia Data to conceptualize and create dashboards from empirical data collected via probes (Astellia and Otarie). Key methodologies like Agile, GIMSI, and Balanced Scorecard were adopted. The result demonstrates a scalable and functional data visualization solution, offering performance indicators and insights tailored to the needs of the host organization, Orange Tu...
This document details the implementation of a Business Intelligence (BI) tool within the Odoo ERP system. Using the SCRUM methodology, it involves the design and development of various data warehouses to address distinct use cases: sales, purchasing, and stock management. The report also covers the integration of PostgreSQL for data storage and manipulation, and Power BI for visualization. Key findings include the establishment of dynamic dashboards enabling efficient data-driven decision-making for the company WAKAHAW.
This document outlines lab exercises focused on optimizing SSIS packages to perform incremental data extraction by using datetime columns, Change Tracking, and Change Data Capture mechanisms. It explains how to reduce ETL processing overhead by extracting only modified records using SQL Server features like CDC and staging updates. The document provides a step-by-step guide for tasks such as configuring SSIS packages, setting variables, modifying data extraction queries, enabling CDC, and testing ETL packages. Practical examples with SQL queries and debugging methods ensure hands-on learnin...
The document introduces the foundational concepts and historical evolution of Artificial Intelligence (AI), focusing on its technical fields, societal impact, and future prospects. It discusses AI's applications in daily life and industry, with a specific emphasis on Huawei's AI development strategy. Additionally, it presents the philosophical and technical schools of thought, including symbolism, connectionism, and behaviorism, along with the delineation of weak and strong AI. Finally, it highlights the layered technology stack and key advancements propelling AI development.
Huawei HiAI Platform, a three-layer AI ecosystem based on 'chip-device-cloud' architecture, enables developers to build intelligent applications through seamless integration of cloud, devices, and chip capabilities. The platform supports advanced features like distributed AI for real-time applications and personalized services, addressing industry-wide challenges such as high development thresholds and fragmented user experiences. Major highlights include HiAI Foundation for mobile AI computing, HiAI Engine for smart app capabilities, and HiAI Service for cross-device functionality. By coll...
This document provides an in-depth analysis of AI chips and their hardware architectures, focusing on Huawei's Ascend AI processors and the Da Vinci architecture to enhance AI computing capabilities. It introduces chip classifications, current industry standards, and comparisons between GPUs, CPUs, FPGAs, and TPUs while detailing Huawei's full-stack platform solutions. The Da Vinci architecture is elaborated upon, including its core computing units like matrix, vector, and scalar operations, and how Ascend processors integrate these elements for high-performance AI computing.
The document introduces the Huawei-developed MindSpore framework for AI computing, aiming to address challenges like high deployment difficulty and computational inefficiencies. It explores development trends and highlights its unique features such as unified dynamic and static graph representations, optimization across hardware and software layers, and the integration of advanced programming paradigms. MindSpore focuses on expanding AI inclusivity via innovations in performance optimization, deployment in diverse scenarios, and security frameworks, making it a competitive alternative in th...
The document explores the concept and advantages of deep learning frameworks, with a focus on PyTorch and TensorFlow. It delves into the basics of TensorFlow 2.x, including common operations, modules, and advanced features such as eager execution and AutoGraph for building computational graphs. The comparative strengths and limitations of both PyTorch and TensorFlow are analyzed, such as PyTorch's dynamic computational graph and TensorFlow's multi-platform scalability. Practical application is demonstrated through a case study: training a deep learning model for MNIST handwritten digit reco...
This document provides a comprehensive overview of deep learning, including its history, core concepts, methodologies, and practical applications. Fundamental topics such as neural networks, training processes, activation functions, and optimization algorithms are detailed. It contrasts traditional machine learning with deep learning and describes advancements enabled by multilayer perceptrons and gradient-based optimization methods. The material also discusses key challenges like the vanishing gradient problem and explores modern activation strategies such as ReLU to address these challenges.
This document provides a foundational overview of machine learning, discussing concepts such as supervised, unsupervised, semi-supervised, and reinforcement learning. It describes the machine learning process from data collection and cleansing to model training and deployment. Key distinctions between machine learning and rule-based systems are highlighted, emphasizing scenarios where machine learning is preferred. Real-world cases and examples showcase the application of learning methods in different tasks with varying complexities.
The document serves as an extensive guide for topics commonly asked during machine learning and data science interviews, providing concise cheat sheets for fundamental concepts such as bias-variance tradeoff, regression analysis, and principal component analysis. It offers mathematical insights, step-by-step explanations, and practical techniques for mitigating challenges like overfitting or class imbalance through concepts like regularization and synthetic data generation. Additionally, the guide includes simple explanations of popular machine learning algorithms, evaluation metrics, and a...
The document outlines a discriminant analysis conducted on a dataset of 46 companies (23 healthy and 23 failing) using financial ratios. Fisher's linear discriminant functions are applied to predict company classification based on ratios, presenting both equations for healthy and failing companies. A predictive example identifies a company's classification based on ratios. A second exercise demonstrates decision tree classification for qualitative observation data using the Gini Index criterion, including tree construction, rule extraction, and method evaluation.
These lecture notes introduce the concept of linear separation and its applications in classification problems, with mathematical formalism using the decision function g(x|w,b). The input data is described as pairs of feature vectors and class labels, with separation determined by the sign of g(x|w,b). The document also briefly mentions generalizing the method to multi-class problems and includes an example application using Support Vector Machines (SVM). Key constraints for optimization are provided to ensure functional margins.
The document is an exam for a Master's program in Business Analytics and Data Science at the Université Virtuelle de Tunis, primarily covering Machine Learning. It consists of problems focusing on the differentiation of data mining methods, discriminant analysis, and predictive modeling for employee retention. Additionally, it examines regression analyses on business investment, interpreting parameters and improving models using explanatory variables. The exam applies analytical techniques and mathematical formulations such as ordinary least squares and decision trees.
This document is the exam for the Machine Learning course as part of the Master's program in Business Analytics and Data Science at the Université Virtuelle de Tunis. It comprises various exercises evaluating theoretical knowledge and practical applications related to data mining and classification methods. Key topics include the differences between descriptive and predictive methods, classification techniques from artificial intelligence, and regression analysis.
This document outlines the third tutorial (TD no. 3) for a Data Mining course. It focuses on the k-Nearest Neighbors (kPPV) method, applying it to examples such as 'Playing Tennis' and text classification. The tutorial includes measures of dissimilarity and classification techniques using kPPV.