인프런 영문 브랜드 로고
인프런 영문 브랜드 로고
AI

/

Deep Learning & Machine Learning

Deep Learning Next Generation Innovation Technology - Introduction to Physical Information Neural Networks and Pytorch Practice

This is a lecture that studies the physical information neural network, one of the next-generation innovative technologies of deep learning, and implements it directly using Pytorch. Let's learn the next-generation innovative technology of artificial intelligence with me!

(5.0) 7 reviews

118 students

PyTorch
Deep Learning(DL)
Machine Learning(ML)
Artificial Neural Network
Thumbnail

This course is prepared for Basic Learners.

What you will learn!

  • The concept of physical information neural network, the next-generation innovative technology in artificial intelligence

  • Building a Physical Information Neural Network Using PyTorch

The emerging next-generation deep learning model
Physical Information Neural Network (PINN)

Recently, deep learning models that integrate the laws of physics are becoming a new key to solving problems using artificial intelligence. Jensen Huang, CEO of Nvidia, emphasized the possibility, saying that the next wave of AI will be AI that learns the physical world . Among them, the most notable model is the Physical Information Neural Network (PINN) .



[Google Trends] Soaring interest in physics-informed neural networks


Physics-informed neural networks are artificial neural networks that are created by learning physical information. They are a technology that can accurately build complex systems with limited data by combining the performance of neural networks with physical information, and are also a next-generation technology in the industrial field. We see it as an innovative technology.

NVIDIA, a leader in artificial intelligence (AI) computing, has also introduced physics-based machine learning models as an innovative technology and released its AI framework Modulus. Amazon, Philips, ExxonMobil, SpaceX, BMW, Siemens, etc., including NVIDIA, Many companies are investing and developing physics-based neural networks, and physics-based machine learning models are expected to drive next-generation innovation across industries.

Why Physical Information Neural Networks?

(1) Solving various difficult problems

Physical information neural networks, which integrate physical laws into deep learning, solve various difficult problems that existing deep learning could not solve, and are expanding the industrial group that applies deep learning . In particular, it has recently been introduced to medicine (new drug development), environment (climate prediction), and architecture (structural design), and is attracting attention as an attractive technology.

Nvidia's Modulus

(2) Less data usage

Supervised learning, the basic learning method of artificial neural networks, generally requires a large amount of data. On the other hand, physics-based learning can build accurate models without data or with only a small amount of data because it predicts based on physics laws.

Physical information neural network

(3) Establishing a system that combines transparency and efficiency

Physical information neural networks can be integrated with various technologies to improve accuracy in various fields, and can greatly improve computational speed compared to existing methods . In addition, since the prediction and decision process of the model is based on physical laws, it can help solve the "black box" problem of deep learning relatively.

FEM vs PINN

From theory to implementation
Contains the basics of physical information neural networks

In this lecture, after learning the concepts, we will implement models for various problems.

To ensure that math does not become a barrier, we cover the concept of differentiation first.
Introduces the concept of physical information neural networks and the learning principles of neural networks.
You can implement a physical information neural network yourself with 6 practical exercises.


Things to note before taking the class

Practice environment

  • The training will be conducted in Google Colaboratory, which does not require separate installation . A Google account (free) is required, and if Colab is not available, the training may be disrupted.

Learning Materials

  • All slides and code used in the class are provided.

  • Implement the model using Pytorch.

Player Knowledge and Notes

  • Some basic knowledge of deep learning, such as supervised learning and gradient descent, is required.

  • This lecture contains a lot of mathematical content such as differential equations and numerical analysis. However, the lecture is structured so that even those without knowledge of differential calculus can listen to it.

  • This is a question-based course. Questions about the class are always welcome :)

Recommended for
these people!

Who is this course right for?

  • Those preparing for universities/graduate schools related to artificial intelligence

  • Anyone who wants to follow the future of artificial intelligence

  • Anyone who wants to experience the next-generation deep learning innovation technology

Need to know before starting?

  • Passion to do

  • Deep Learning Basics

  • Basic knowledge of Python

Hello
This is

4,542

Students

298

Reviews

250

Answers

4.7

Rating

7

Courses

안녕하세요.

딥러닝/머신러닝 관련 유튜브를 운영하는 딥러닝 호형입니다.

수학/데이터 분석을 전공하고 다수의 딥러닝 프로젝트를 완료하고 수행하고 있습니다.

 

머신러닝, 고급 머신러닝, 딥러닝, 최적화 이론, 강화 학습 등의 인공지능 내용과 선형 대수학, 미적분, 확률과 통계, 해석학, 수치해석 등의 수학 내용까지 여러분들과 공유할 수 있는 지식을 가지고 있습니다. 

 

모두 만나서 반갑습니다!

 

* 관련 이력

현) SCI(E) 논문, 국제 학회 발표 다수

현) 인공지능 관련 대학교 자문 다수

전) K기업 전임 연구원 - 데이터 분석 및 시뮬레이션: 신제품 개발, 성능 향상, 신기술 적용

"딥러닝을 위한 파이토치 입문" 저서 (세종도서 학술부문 2022 우수도서로 선정)

 

 

 

 

Curriculum

All

19 lectures ∙ (2hr 22min)

Lecture resources

are provided.

Published: 
Last updated: 

Reviews

Not enough reviews.
Become the author of a review that helps everyone!