Automated Machine Learning (AutoML - MT 102-010)

Spring 2022-2023 / Signature Work Mini-Term (4 days, 12 + 12 hours)

Course Period: March 13 - 16, 2023

  • Lectures: Monday / Tuesday / Wednesday / Thursday @ 09:00-12:00 (Classroom: IB 2028 + Zoom)
  • Labs: Monday / Tuesday / Wednesday / Thursday @ 14:00-17:00 (Classroom: IB 2028 + Zoom)
Instructor: Mustafa MISIR (Office: CC 3019), mustafa.misir [at] dukekunshan.edu.cn / mm940 [at] duke.edu


Machine Learning (ML) is a popular field with interdisciplinary characteristics relating to various fields including Computer Science, Mathematics and Statistics. ML aims at learning without being explicitly programmed, through data and experience. The target applications are the complex tasks which are challenging, impractical or unrealistic to program. ML can be used to address those sophisticated activities that humans or animals can routinely do such as speech recognition, image understanding and driving. The other functions to learn that ML concentrates on, are concerned with the ones requiring capabilities beyond human capacities, in terms of speed and memory.

Despite the successes of ML, its accomplishments rely on the decisions given for the different stages of the ML application pipeline. Automated ML (AutoML) approaches ML as a computational search problem with the goal of automatically exploring the (near)-best ML settings. This course will provide the basic AutoML concepts and ideas while introducing the existing algorithms and tools to effectively perform AutoML.

The course requires basic programming skills (Python) and introductory level knowledge on ML.

By the end of this course, you will be able to:
  1. analyze the ML pipeline for ML problems
  2. specify the design components of ML algorithms
  3. identify critical the hyper-parameters of ML algorithms
  4. determine the appropriate AutoML methods for for enhancing ML algorithms
  5. manipulate the given data concerned with a learning task so that the preferred ML algorithm can be effectively applied
  6. construct generalizable ML models that can address a given ML problem of unseen data
  7. build complete ML workflows in Python together with the relevant libraries / frameworks / tools besides effectively communicating your methods and results using Jupyter notebooks
+++ Follow Sakai for announcements and discussions

Pre-requisites

  • STATS 102: Introduction to Data Science
  • STATS 201: Introduction to Machine Learning for Social Science
  • ENVIR 208: Environmental Data Analytics
  • / Consent of the Instructor


There is no official textbook for this course. Still, the following books can be used as references.

Reference Books


Lecture Notes / Slides


Reference Courses


Other Books

Artificial Intelligence / Machine Learning - Quick / Easy Reads:
Machine Learning:
Python Programming:
Python Programming for Data Science / Analytics:
Data Visualization:

Other Materials / Resources