Fall 2021 / CSCE 590: Trusted AI - Special Topic
Quick Info - When and Where
Tuesday/Thursday 2:50 pm – 4:05 pm
In person at 300 Main St. B111. Recordings to be available on Blackboard.
Instructor Information
Instructor: Biplav Srivastava
E-mail: biplav.s AT sc.edu
Office Hours: 1130-1230 MW; other times by appointment
GitHub for slides, sample code.
Attendance Policy
Students are expected to attend lectures and also go through class videos after the lecture. They are expected to participate in quizzes, do their project and complete paper reading.
Suggested Reading
Material from my earlier teachings: CSCE 590-1: Data to Decisions (Spring 2021), and CSCE 771: Computer Processing of Natural Languages (Fall 2020).
Articles and papers as announced in class.
Research Interests
See ongoing research activities in this area at "Trusted AI" page.
Details
Full details will be available in syllabus here.
Bulletin Description
This will be an advanced Artificial Intelligence (AI) course focused on understanding reasons why Artificial Intelligence (AI) systems can be problematic and what can be done to make them trust-worthy. We will briefly cover AI as a decision support technology involving data processing, generating analysis and communicating insights to users. Then we will study how output of AI can be sensitive to issues in data, algorithmic steps (pre-, during and post-) and interaction with people of diverse background. We will conclude with techniques to remove or mitigate issues. The students will be exposed to latest tools and will do a project using AI technique of their choosing. The course will cover AI sub-fields of learning, reasoning, representation, preferences and uncertainty.
Prerequisites
Experience with a first course in data structures (CSCE 350), programming (CSCE 330) and introductory AI course (CSCE 580, 590 or equivalent) is advised.
Learning outcomes
As a result of successful participation in this course, undergraduate students will be able to:
L1: Explain, execute and create AI-based analytical methods to process data: (a) unstructured data, (b) semi-structured data, (c) structured data
L2: Explain AI methods in data analysis: (a) Learning methods, (b) Reasoning, (c) Representation and standardization – knowledge graphs/ ontology, (d) Preferences, (e) Handling Uncertainty
L3: Identify trust issues in AI methods: (a) fairness and bias, (b) harmful language, (c) data privacy
L4: Methods and tools to promote trust: (a) Data sampling and synthetic data, (b) Testing and rating for communication, (c) Algorithmic innovations like differential privacy and explanations
As a result of successful participation in this course, graduate students will be able to do all of the above, and:
L5: Evaluate gaps in Trusted AI tools and create new datasets to handle them
L6: Explain emerging standards, frameworks and laws.
L7: Explain research findings in open areas and critique their contributions
Evaluate gaps in analytical methods and create new ones to process data