Big Data for Reliability and Security

big-data.jpg

Instructor

Saurabh Bagchi, Professor Electrical & Computer Engineering, Computer Science, CRISP Center Director

Course Outcome

This course builds an understanding of techniques for meeting reliability and security requirements in connected computing systems, covering not only traditional threats but new threats posed by big data and large-scale systems. Students will learn big data analytic and machine learning techniques for improving reliability and security and develop software to apply to real-world datasets under realistic conditions.

Prerequisites: Knowledge of Python and Introductory Statistics

Topic Include: 

Foundational material on reliability and security

  • Introduction: Motivation, System view of reliable and secure design, Terminology 
  • Security landscape for connected systems: Traditional threats, new threats due to large-scale systems, new threats due to big data 
  • Reliability landscape for connected systems: Traditional concerns, new concerns due to large-scale systems, new concerns due to big data

Data analytic techniques for dependability

  • Supervised and unsupervised learning technique
  • Neural Networks building blocks
  • Techniques for dealing with large-scale data; regularization, feature engineering, dimensionality reduction, etc.
  • What is our tool chest of data analytic techniques: what to use and when
  • Data analytic techniques used for reliability and security: strengths, weaknesses opportunities

Big data security and insecurity

  • Attacks against big data algorithms: evasion and poisoning attacks 
  • White box and black box attacks 
  • Defenses: Adversarial training, defensive distillation, adversarial example detection 
  • Machine learning at scale: Federated learning 
  • Federated learning for security and privacy 

Case Studies and Challenge Problems

  • Case studies on adversarial Machine Learning: Image and video manipulation 
  • Systems for big data processing: Spark, TensorFlow on the cluster, TensorFlow Light. Benchmarks for big data processing
  • Challenge problems
    • Challenge problem 1: Predicting computer system failures
    • Challenge problem 2: Proximity detection through Bluetooth signals

CEUs: 1.5


For further information and to register, please email noncredit@purdue.edu.