Art and Science of Machine Learning em Português Brasileiro

Google Cloud via Coursera

Go to Course: https://www.coursera.org/learn/art-science-ml-br

Introduction

### Course Review: Art and Science of Machine Learning em Português Brasileiro #### Overview In today's data-driven world, machine learning (ML) has emerged as a pivotal technology, transforming the way we interact with information and make decisions. The course, **Art and Science of Machine Learning em Português Brasileiro**, offers a comprehensive introduction to this fascinating field, catering specifically to Portuguese speakers. With a blend of theoretical insights and practical applications, this course is designed to provide learners with the essential skills and knowledge needed to navigate the complexities of machine learning. #### Course Structure This extensive course consists of **six modules**, each focusing on various critical facets of machine learning: 1. **Introdução (Introduction)**: - The course begins with a fundamental overview of machine learning, establishing the essential skills required for successful model tuning and optimization. Key concepts such as intuition, common sense, and experimentation in ML are emphasized, laying a solid foundation for learners. 2. **A arte do ML (The Art of ML)**: - Here, learners delve into the practical aspects of machine learning, focusing on optimizing model performance by adjusting batch size and learning rate. This module emphasizes hands-on application, guiding learners to translate concepts into code using TensorFlow. 3. **Ajuste de hiperparâmetros (Hyperparameter Tuning)**: - This module distinguishes between parameters and hyperparameters, introducing traditional grid search methods and more sophisticated techniques. The course also highlights the utility of Cloud ML Engine in automating hyperparameter tuning, making it accessible even for those less familiar with coding. 4. **Uma pitada de ciência (A Pinch of Science)**: - Bridging the artistry of ML with scientific principles, this section covers regularization, logistic regression, and model performance assessment. This robust integration helps learners appreciate the balance between creative modeling and empirical evidence. 5. **A ciência das redes neurais (The Science of Neural Networks)**: - A deep dive into the workings of neural networks, this module details their structures and functionalities, providing learners with a solid understanding of one of the most influential ML methodologies. 6. **Embeddings**: - The final module introduces embeddings as a means to handle sparse data, accelerating training and minimizing memory consumption. This section also discusses dimensionality reduction and how it contributes to more straightforward, generalized models. #### Learning Outcomes By the end of the course, participants will: - Acquire vital intonation skills necessary for successful ML model optimization. - Understand how to adjust batch size and learning rates to enhance model performance. - Gain proficiency in differentiating and tuning hyperparameters with both traditional and innovative methods. - Develop insights into regularization techniques and the application of logistic regression. - Cultivate a thorough understanding of neural networks and their underlying science. - Implement embeddings effectively to manage data sparsity and improve training efficiency. #### Recommendation I highly recommend the **Art and Science of Machine Learning em Português Brasileiro** course for anyone looking to build a career in the exciting field of machine learning. The course is structured in a way that balances theoretical knowledge and practical skills, making it suitable for both beginners and those with some familiarity in the field. The use of real-world coding examples in TensorFlow will embolden learners to implement what they’ve learned directly, enhancing not just understanding but also retention. Moreover, the use of Portuguese as the medium of instruction makes this course particularly valuable for Portuguese-speaking professionals and students, ensuring they can fully grasp complex concepts without the language barrier. This course is not just another online learning experience; it’s a gateway to mastering machine learning with a well-rounded approach that embraces both the art and the science of this transformative technology. Join now and elevate your understanding of machine learning to new heights!

Syllabus

Introdução

Este é o curso Art and Science of Machine Learning. Falaremos sobre as habilidades essenciais de intuição, bom senso e experimentação em ML para ajustar e otimizar modelos e ter melhor desempenho. Você aprenderá a generalizar os modelos usando técnicas de regularização e conhecerá os efeitos dos hiperparâmetros, como tamanho de lote e taxa de aprendizado, sobre o desempenho do modelo. Também abordaremos alguns algoritmos mais comuns de otimização de modelo e mostraremos como especificar um método de otimização no código do TensorFlow.

A arte do ML

Neste módulo, você aprenderá a ajustar o tamanho do lote e a taxa de aprendizado para melhorar o desempenho do modelo, otimizá-lo e aplicar os conceitos ao código do TensorFlow.

Ajuste de hiperparâmetros

Neste módulo, você aprenderá a diferenciar parâmetros e hiperparâmetros. Em seguida, veremos a abordagem tradicional de pesquisa de grade e outras com algoritmos mais inteligentes. Por fim, você verá como o Cloud ML Engine facilita a automação do ajuste de hiperparâmetros.

Uma pitada de ciência

Neste módulo, falaremos da ciência junto com a arte do machine learning. Primeiro vamos falar sobre como fazer a regularização da esparsidade e criar modelos mais simples e concisos. Depois abordaremos a regressão logística e veremos como determinar o desempenho.

A ciência das redes neurais

Neste módulo, vamos nos aprofundar na ciência, especificamente as redes neurais.

Embeddings

Neste módulo, você aprenderá a usar embeddings para gerenciar dados esparsos, acelerando o treinamento e reduzindo o consumo de memória dos modelos de machine learning que usam esses dados. Os embeddings também são uma forma de reduzir a dimensionalidade e tornar os modelos mais simples e generalizáveis.

Resumo