Introduction to Probabilistic Knowledge Graphs: AI-Driven Inference and Real-World Applications

Course details

Probabilistic knowledge graphs (PKGs) are transforming the way data is structured and analyzed in AI and machine learning. In this course, AI engineer and author Vaibhava Lakshmi Ravideshik introduces the fundamentals of knowledge graphs and shows how probability theory helps manage uncertainty in data. Through hands-on examples, learn how to construct PKGs, integrate probabilistic reasoning, and apply inference techniques like Markov Chain Monte Carlo (MCMC). Explore real-world applications in decision-making, risk assessment, and predictive modeling, and gain insights into scaling challenges and emerging trends shaping the future of PKGs.

Learning objectives
Explain the fundamentals of knowledge graphs, including their structure, components, and real-world applications.
Apply probability theory concepts, including random variables and Bayesian networks, to handle uncertainty in data.
Construct and analyze PKGs by integrating uncertainty into traditional knowledge graphs.
Use probabilistic inference techniques, such as Markov Chain Monte Carlo (MCMC), to make predictions and decisions in PKGs.
Identify challenges and future trends in the field of PKGs, such as scalability, data quality, and AI advancements.

Concepts

Introduction

Diving into the world of probabilistic knowledge graphs

What Are Knowledge Graphs

What are knowledge graphs (KGs)
Core components of a knowledge graph
Why knowledge graphs for AI
Real-world KG I - Google's knowledge graph
Real-world KG II - DBpedia
Simple KG implementation, part I
Simple KG implementation, part 2
LLM-based KG implementation, part 1
LLM-based KG implementation, part 2
Summary - Basics of knowledge graphs

Overview of Probability Concepts

Foundations of probability theory
Random variables and distributions
Expectation and variance
Conditional probability and Bayes' Theorem
Bayesian networks
Directed edges and influence flow
Probabilistic modeling basics for PKGs
The urge for MCMC
MCMC and intuition
The Metropolis-Hastings algorithm
From sampling to confidence
Hamiltonian Monte Carlo (HMC)
No-U-Turn Sampler (NUTS)
Basics of Pyro programming, part 1
Basics of Pyro programming, part 2
A wrap-up on probabilistic concepts for knowledge graphs

Introducing Uncertainty in Knowledge Graphs

Uncertainty - The reason we build PKGs
Sources of uncertainty
Example - Uncertainty in a medical KG
Risks of ignoring uncertainty
Confidence and provenance
Building probabilistic KGs, part 1
Building probabilistic KGs, part 2
A wrap-up on uncertainty

Data Sources and Collection Methods

Graph schema design
Example ontology template
Structured data for PKGs
Typical structured sources
Limitations of structured data
Real-world example - Structured data
Unstructured data for PKGs
Limitations of unstructured data
Example - Biomedical literature mining
Wrapping up data types for PKGs

Inferencing with PKGs

Preview of KGEs
Bayesian PKG using KG embeddings
What is inference in a PKG
Inference approaches in PKGs
PSL
Markov logic networks
Probabilistic graphical models
Path-based probabilistic reasoning
Bayesian inference using Pyro, part 1
Bayesian inference using Pyro, part 2
Bayesian inference using Pyro, part 3

PKGs in Modern AI Systems

PKGs as a technical and philosophical approach
Bayesian GNNs, part 1
Bayesian GNNs, part 2
Neurosymbolic systems, part 1
Neurosymbolic systems, part 2
Autonomous agents
Justified beliefs rather than absolute answers