December 6, 2023

photo of Spot the robot picking up trash
The Spot Clean-Up Crew in the Introduction to Machine Learning course proposed to merge a supervised learning model with robots for autonomous trash pick-up. This image shows Spot using ML to pick up a piece of trash.

The field of Machine Learning (ML), a subset of artificial intelligence (AI), typically refers to computational and statistical methods for automated detection of meaningful patterns in data. ML/AI have become ubiquitous in all aspects of our lives including speech recognition (e.g. question answering from Alexa or Apple Siri), pattern recognition (face ID, detection of tumor/disease from medical images, etc.), spam/fraud email detection and self-driving cars, to name a few. While machine learning approaches have proved to be state-of-the-art methods in the fields of computer vision, speech recognition, natural language processing and more, it is not yet widespread in the engineering and science community.

Realizing the vital role of ML/AI for research and job competitiveness of our students, UT Austin’s Department of Aerospace Engineering and Engineering Mechanics (ASE/EM) began offering the “Introduction to Machine Learning” course (COE 379L/EM 397) in 2018. The course is offered each fall semester to ASE/EM undergraduate and graduate students and is also open students across the University.

Contrary to some beliefs that ML/AI will take over traditional mathematics and sciences, our main objective in this class is to teach the students three important skills:

  1. Use knowledge in engineering, science and mathematics to derive and gain insights into the current state-of-the-art machine learnings methods/algorithms.
  2. Implement the algorithms in an educational ML/AI SciKit-Learn environment.
  3. Extend/modify ML/AI methods to solve practical problems with real data.

As a byproduct, the students also learn how to use SciKit-learn library and state-of-the-art implementations of AI/ML algorithms. 

The class includes 6-7 bi-weekly homework assignments, each of which consist of theoretical questions (deriving or proving) and implementation of machine learning methods to solve real world problems with real data. There are three take-home exams and no final.

A key component of the class is the final project in which up to three students form a group to conduct new research or reproduce a research paper with modifications, building upon knowledge learned from class.

This year, 27 students were divided into 11 groups, with each group selecting a project of their interest. The diversity of the projects was impressive, yet not surprising, as students taking the course were from various disciplines, including a mix of graduates and undergraduates.

A total of 15 undergraduates across five majors included: 8 computational engineering, 3 aerospace engineering, 1 mechanical engineering, 1 physics and 2 electrical and computer engineering. A total of 12 graduate students  across 6 disciplines included: 1 chemical engineering, 1 aerospace engineering, 3 mechanical engineering , 2 civil engineering and 5 electrical and computer engineering. A handful were exchange students from Germany, Netherlands and Denmark.

Some project highlights of this year’s course include:

Supervised Learning Applied to Financial Market (Group 1) compared random forest method provided in class with long short-term memory neural network in predicting stock market conditions to know when to buy and sell to make profit. They planned to start a start-up company building upon this project. 


This figure shows the accuracy of the random forest, an approach provided in class, in predicting the stock market prices.

 

Prediction of Students’ Academic Performance Using Machine Learning Algorithms (Group 11) The key idea is to use current exam grades and information from faculty and departments to predict final grades for students. Their findings, following a published research paper, could facilitate decisions on helping struggling students by providing tutoring and improving teaching. This project highlighted the success of the class in teaching the students important ML/AI methods (e.g. naïve Bayes, random forests, k-nearest neighbors, support vector machines, and logistic regression) to attempt to solve such an important problem.

Detection of Malware Using Machine Learning based on Operation Code Frequency (Group 9) deployed their learning from class (i.e. K-Nearest Neighbors, Gaussian, Multinomial and Complement Naive Bayes classifier) to detect malwares without executing them, thus preventing the infection.

Feature-Extraction (Group 10) used the learned support vector machine method to classify text data for spam/fraud email detection.

Symplectic Neural Networks for Predicting Dynamical Systems (Group 5) went beyond the class materials with sophisticated mathematics to incorporate the structure of the dynamical system into deep learning to preserve the energy.

Physics-Informed Neural Networks for Solid Mechanics (Group 6) deployed a physics-informed constitutive neural network models to predict hyper-elastic material behavior using real data from their lab. They aim to publish their findings.

Spot Clean-Up Crew (Group 8) proposed to merge a supervised learning model with robots for autonomous trash pick-up. They plan to write a research paper on the project.

Students who took the class this fall say they benefited greatly from the knowledge gained about ML and feel it will benefit them moving forward in both their studies and future careers.

"Dr. Bui's Intro to Machine Learning class was one of the best classes I've taken at UT, due to both the wide applicability of the material covered and Dr. Bui's infectious passion for the subject. I especially enjoyed the large focus on the fundamental math behind each of the algorithms we were taught, which solidified my understanding and allowed me to apply these algorithms more confidently." – George Braun, COE undergraduate major

"This class is a great option for any student looking to start learning about ML— especially for undergraduate students looking to apply the concepts they’ve learned so far in a meaningful and practical way." - Anneris Rodriguez, COE undergraduate major

"The class provides a well-structured introduction to machine learning. The lectures have introduced me to concepts of unsupervised and supervised learning in a very engaging manner with detailed mathematical explanations and applicable real-world examples. The assignments, exam and project have tested my understanding the intricacies of these concepts, and it has been a rewarding experience." - Arjit Seth, ASE graduate student

Learn more about the Introduction to Machine Learning course.