The following post is from Oge Marques, Professor at FAU, who will discuss 5 topics for getting started in AI, with many useful MATLAB links and a new course to learn more.
There is great demand for courses, webinars, and other educational initiatives that bring Artificial Intelligence (AI) and related topics – data science (DS), machine learning (ML), neural networks (NNs), deep learning (DL) – to a wider audience. After all, we live in a world of "AI + X" where companies in every segment of human activity need professionals who have domain expertise, augmented with AI skills.
Since much of AI today is actually machine learning, and the early stages of ML solutions require basic understanding of data science concepts, as a starting point I would like to recommend getting acquainted with the main concepts and terminology associated with the classical DS/ML/DL workflow.
In this blog post, I want to highlight 5 topics to master prior to starting your first AI project.
Figure 1: Example of deep learning workflow. See this blog post for additional details.
I have structured these topics in the form of "Tasks" that you can perform using the latest low-code/no-code capabilities offered in MATLAB. The selected tasks should help you broaden your view of the field of AI and allow a better understanding of its foundations, risks, applications, and implications.
Topic 1: Introduction to AI
I suggest you start from the main definitions, concepts, and terminology associated with AI, as well as historical highlights of the field as it developed over the past 60 years or so.
Can you answer the following questions:
- What is AI?
- How does it work?
- What AI can (and cannot) do today?
Being able to talk about AI at a high level will give you the tools you need to engage with colleagues and understand where AI can be used for greatest impact.
MATLAB task 1
: Invest some time understanding key MATLAB apps for AI, ML, and DL, notably Classification Learner
, Regression Leaner
, and Deep Network Designer
. These apps provide a low barrier to enter the world of designing, building, and understanding realistic AI applications using a wide variety of datasets.
Topic 2: Fundamentals of regression and classification
A large number of ML/DL applications consist of designing supervised learning solutions for two types of tasks: regression
(essentially the prediction of a numerical value, e.g., tomorrow's temperature based on historical data) and classification
(essentially assigning a class or category to a data item, e.g., given a photo of an animal, identify whether it is a cat or a dog).
MATLAB task 2
: Return to the Classification Learner and Regression Leaner apps to explore additional problems, select different models for the same problem, tweak parameters of the selected models, and compare different solutions in a principled way.
Invest some time playing with the MATLAB Live Editor
and learning some of its most useful features, such as tasks and interactive controls.
Topic 3: Fundamentals of neural networks
To better grasp the world of deep learning, you should have a solid understanding of the fundamentals of NNs and how they have evolved, from the early days of the perceptron to the deep architectures used today. By doing so, you will be better able to understand complex contemporary NN architectures and judge whether your application may benefit from the latest models or a simpler approach may suffice.
MATLAB task 3
: Return to the Deep Network Designer app and explore additional functionality, both when you want to create a network from scratch as well as when the use of transfer learning
strategies is appropriate for the problem at hand.
Topic 4: Representative contemporary DL architectures
A very useful shortcut for getting into the vast world of deep learning is to focus on the most representative contemporary DL architectures for any given problem. These include: convolutional neural networks (CNNs) for image classification
, recurrent neural networks (RNNs
) – including the popular LSTM
variant – for tasks involving sequential data, U-nets (for semantic image segmentation
), generative adversarial networks (GANs) for image translation
, YOLO (for object detection
in images and videos), and transformers
for natural language processing (NLP).
MATLAB task 4
: Import 3 deep learning networks from MATLAB Deep Learning Model hub
and compare the accuracy of each for a specific example here
Topic 5: Best practices for experiment management
Working on AI/ML/DL projects requires a fair amount of experimentation and multiple iterations of adjustments, fine-tuning, and optimizations. You can be much more efficient at this practice by setting up good experiment practices, and keeping track of all models trained.
MATLAB task 5
: Understand best practices for managing DL experiments
and illustrate them using MATLAB's Experiment Manager
Want a deeper dive?
All 5 topics and associated tasks will be covered in depth in the AI Bootcamp course
that is being offered by Florida Atlantic University this month. This bootcamp, now in its fourth edition, has been thoroughly revised, updated, and expanded to 15 hours to include hands-on activities using MATLAB Online. I will be your instructor for the course, which will be offered live, fully online, with lectures recorded for later (re)viewing.
The goals of this bootcamp are to provide an overview of the field of AI with emphasis on contemporary techniques, such as machine learning and deep learning, and their applications in many areas, including computer vision, natural language processing, and medical diagnosis.
I am confident that our AI Bootcamp
will provide its participants a technically rich, yet accessible, introduction to AI, deep learning, and related topics. Their experience will be further enriched by a broad variety of hands-on examples using low-code apps within the MATLAB Online environment.
I hope to see you there!
댓글을 남기려면 링크 를 클릭하여 MathWorks 계정에 로그인하거나 계정을 새로 만드십시오.