CS60218: Foundations of AI Reasoning
Instructor: Dr. Somak Aditya
Instructor: Dr. Somak Aditya
For all announcements, consult the website first. For further clarifications, please ask the Teacher/TAs
Prerequisite:
You must have taken Machine Learning (CS60050) or equivalent. Having taken the Discrete Structures course (or Foundations of AI) course which provides background in Logic may help (but not necessary)
For some Assignments, programming language skills such as in Python, understanding of libraries such as pytorch will help. (But students often pick it up during the course of time).
This course provides a comprehensive introduction to the foundations and modern advances in AI reasoning, bridging classical symbolic approaches with contemporary neural and large language model–based methods. It covers core paradigms of reasoning including logical, probabilistic, and planning-based reasoning, along with graphical models and non-monotonic logics. The course further explores neuro-symbolic frameworks that integrate learning with rules, constraints, and probabilistic logic, enabling principled reasoning under uncertainty. In the latter part, the course focuses on reasoning with large language models, including chain-of-thought style reasoning, logical decoding, and LLM-modulo planning and verification frameworks. Overall, the course equips students with both theoretical foundations and practical insights into building AI systems capable of structured, interpretable, and reliable reasoning.
Lectures: WED 12-1, THU 11-12, FRI 9-10 Slot E3,
Venue: CSE 120 Credits: 3. LTP: 3-0-0
Instructor: Prof. Somak Aditya
Email: saditya@cse.iitkgp.ac.in
Office hours: TO BE DECIDED
Weightage (Tentative)
Midsem: 30%
Endsem: 30%
Quiz/Attendance/Class-Participation: 20%
Assignment/Guided-Project (Max 1-2 people): 20%
Inductive, Deductive, Abductive reasoning
Modalities of reasoning:
Logical
Spatial
Temporal
Causal
Reasoning as planning and optimization
Propositional Logic
First Order Logic
Syntax
Semantics
Non-monotonic Logic
Prolog
Answer Set Programming (ASP)
Stable Model Semantics
Action-fluents
Modal Logics (light coverage):
Temporal Logic
Allen’s Interval Algebra
RCC8 (Spatial reasoning)
Reasoning as Planning using Prolog / ASP syntax
Undirected Graphical Models (Markov Random Fields)
Exact learning
Inference
Hammersley–Clifford Theorem
Syntax, Semantics, Learning, and Inference
Markov Logic Networks (MLN)
ProbLog
Neural extensions of probabilistic logic:
Logic Tensor Networks
DeepProbLog
NeurASP
Sum-Product Networks
Integer Linear Programming (ILP)
Soft logic (t-norms) for weighted rules
Deep neural networks with logic rules as soft constraints
TableILP: Integrating ILP with neural networks
Primal–dual formulation for deep learning with constraints
Reasoning techniques using LLMs:
Chain-of-Thought
Tree-of-Thought
Program-of-Thought
Logical decoding (Neuro-Logic decoding)
Large reasoning models:
ORPO
GRPO
Neuro-symbolic extensions of LLMs:
LLMs as logical translators
Test-time scaling and verification of chain-of-thought
Monte Carlo Tree Search (MCTS) on chain-of-thought
LLM planning in LLM-modulo frameworks
Verification of LLM-generated plans
Sl. No. Topic Hours
1 Definition and Taxonomy of Reasoning 3
2 Fundamentals of Logic 2
3 Advanced Paradigms of Logic 4
4 Fundamentals of Graphical Models 2
5 Probabilistic Logic – Syntax, Semantics, Learning and Inference 10
6 Learning with Rules in Neural Networks 6
7 Reasoning and Large Language Models 8
8 LLM-Modulo Frameworks 2
Total 37 Hours
Neuro-Symbolic Artificial Intelligence: The State of the Art
Edited by Pascal Hitzler and Md Kamruzzaman Sarker (2022)
Advances in Neuro-Symbolic Reasoning and Learning
https://neurosymbolic.asu.edu/advances-in-neuro-symbolic-reasoning-and-learning/
Neuro Symbolic Reasoning and Learning (in press, 2023)
By Paulo Shakarian, Gerardo I. Simari, Chitta Baral, Bowen Xi, and Lahari Pokala
Aritra Dutta: aritradutta18@gmail.com
Sachin Vasistha: sachinvashistha6916@gmail.com (Additional TA)