NeuroSymbolic AI (CS 785)

Course Information

Course Description

Modern AI largely relies on deep learning, yet fundamental challenges—such as limited adaptability, poor generalization, reduced robustness, lack of explainability, insufficient abstraction, absence of common-sense reasoning, and weak causal reasoning—remain unresolved. Moreover, these advances often rely on massive datasets for training or pre-training. Symbolic methods offer a complementary path by enabling declarative specifications of expected behavior, embedding prior knowledge, and enriching AI systems with reasoning capabilities.

Neurosymbolic AI integrates data-driven neural models—effective at perception—with symbolic representations grounded in logic and probabilistic reasoning. This synthesis bridges intuitive, fast learning with deliberate, structured reasoning, yielding systems that are more adaptive, explainable, and robust.

This seminar investigates the theoretical and practical foundations of Neurosymbolic AI. Main topics include causal inference, hybrid neural-symbolic architectures, attention-based reasoning, and strategies for merging perceptual models with structured world representations.

Learning Outcomes

By the end of this course, students will be able to:

• Analyze the theoretical and practical foundations of Neurosymbolic AI, including causal inference, hybrid architectures, and reasoning.
• Critically analyze and present recent research papers in Neurosymbolic AI, demonstrating clear understanding of contributions, limitations, and open problems.
• Compare and evaluate strategies for integrating perceptual models with structured world representations.
• Design and conduct a research-oriented project that applies or extends Neurosymbolic AI methods.

Prerequisites

None.

Course Materials

No textbook is required. Readings are research papers from leading publishers and conferences, accessible through the NJIT library; students review them before each seminar discussion.

Grading

Assignments & Projects

Course Project

Students work in pairs on a semester-long project addressing a concrete problem using neurosymbolic algorithms.

• Project proposal: At most two pages covering the problem statement, data, proposed approach, and evaluation plan.
• Mid-semester milestones: A 30-minute project pitch, a written progress report, and feedback consultations with the instructor.
• Final report: A conference-style write-up due on Monday of the last week of classes.
• Code submission: One ZIP file on Canvas with everything needed to reproduce the main results.
• Final presentation: About one hour per team during the final weeks of the semester.

Paper Presentations and Discussion

Presentations are individual: about 60 minutes followed by 30 minutes of questions, with two papers presented each week. Students who are not presenting read both papers in advance and come prepared with questions.

Course Policies

• Late work: 10% penalty per day; work more than seven days late is not graded. Request extensions in advance.
• Attendance: Mandatory; paper discussion is a graded component, so attend and participate in every class meeting.
• Regrades: Request within two weeks of the work being returned.
• AI tools: Allowed as a supplementary aid to learning; using AI to generate or complete graded work is prohibited.
• Collaboration: Discuss ideas freely, but write and implement all submitted work independently, and list discussion partners on your submission.
• Communication: Begin email subjects with “CS 785:”. Expect replies within two business days.

The full syllabus includes the complete course and university policies.

Additional Information