003d Knowledge Representation - Uncertainty.pdf
Document Details
Uploaded by HardierNeptunium
Tags
Full Transcript
APP002-4-2-IAI& Introduction to Artificial Intelligence KNOWLEDGE REPRESENTATION - UNCERTAINTY Module Code & Module Title Slide Title SLIDE 1 TOPIC LEARNING OUTCOMES...
APP002-4-2-IAI& Introduction to Artificial Intelligence KNOWLEDGE REPRESENTATION - UNCERTAINTY Module Code & Module Title Slide Title SLIDE 1 TOPIC LEARNING OUTCOMES At the end of this topic, you should be able to: 1. Explain the concept of uncertainty in Artificial Intelligence? 2. Name at least two (2) sources of uncertainty in AI 3. Name at least two (2) types of uncertainty in AI 4. Choose the appropriate technique for addressing uncertainty in AI 5. Choose the appropriate method to solve problems with uncertain knowledge 6. Name at least two (2) Importance of understanding uncertainty in AI 7. Calculate the uncertainty in AI 8. Distinguish the differences between frequentist and Bayesian approaches Module Code & Module Title Slide Title SLIDE 2 CONTENTS & STRUCTURE 1. What is uncertainty in Artificial Intelligence? 2. Sources of uncertainty in AI 3. Types of uncertainty in AI 4. Techniques for addressing uncertainty in AI 5. Ways to solve problems with uncertain knowledge 6. Importance of understanding uncertainty in AI 7. Approaches for calculating uncertainty in AI 8. Comparison between frequentist and Bayesian approaches Module Code & Module Title Slide Title SLIDE 3 Recap From Last Lesson How Do Frames Handle Inheritance and Defaults in AI Knowledge Representation? Image credits: google.com Module Code & Module Title Slide Title SLIDE 4 Recap From Last Lesson Frames can inherit information from parent frames in a hierarchical structure. Child frames inherit slots and slot values from their parent frames by default. This allows some information to be specified at a general, abstract level, while more specific details are added in child frames. Slots can have default values that are used if no other value is specified. Default values allow some assumptions to be built into a frame representation. For example, a frame representing a bird might have a "can fly" slot defaulted to true, since most birds can fly. Inheritance can interact with defaults in a couple ways. Child frames can override parent slot values, allowing exceptions to defaults to be specified. And some representations distinguish between strictly-inherited and non-strictly inherited slots - the latter don't pass down default values to children. Frame systems also generally have mechanisms for handling conflicts that arise between multiple parents, or between parents and children. Things like recency-based overrides, Image credits: google.com specificity-based overrides, and explicit precedence rules come into play here. Module Code & Module Title Slide Title SLIDE 5 What is uncertainty in Artificial Intelligence? Uncertainty in artificial intelligence (AI) refers to the lack of complete and accurate information when making decisions or drawing conclusions in a computer or AI system. Image credits: google.com Module Code & Module Title Slide Title SLIDE 6 Sources of Uncertainty in AI Here are some common sources of uncertainty in AI: Data Uncertainty: AI models are trained on data, and the quality and accuracy of the data can affect the performance of the model. Model uncertainty: AI models are complex and may have various parameters and hyperparameters that need to be tuned. Image credits: google.com Module Code & Module Title Slide Title SLIDE 7 Sources of Uncertainty in AI…cont. Here are some common sources of uncertainty in AI: Algorithmic uncertainty: AI algorithms may be based on different mathematical formulations, leading to different results for the same problem. Environmental uncertainty: AI systems operate in dynamic environments, and changes in the environment can affect the performance of the system. Image credits: google.com Module Code & Module Title Slide Title SLIDE 8 Sources of Uncertainty in AI…cont. Here are some common sources of uncertainty in AI: Human uncertainty: Human behavior and preferences are difficult to predict, leading to uncertainty in the use and adoption of AI systems. Uncertainty in AI inference: AI systems use reasoning techniques to make decisions or predictions. However, these techniques can be uncertain due to the complexity of the problems they address, or the limitations of the data used to train the models. Image credits: google.com Module Code & Module Title Slide Title SLIDE 9 Sources of Uncertainty in AI…cont. Here are some common sources of uncertainty in AI: Uncertainty in AI perception: AI systems perceive their environment using sensors and cameras that may be subject to noise, occlusions, or other interference. This can lead to uncertainty in the accuracy of data used to train AI models or in the effectiveness of AI systems in real-world applications. Image credits: google.com Module Code & Module Title Slide Title SLIDE 10 Sources of Uncertainty in AI…cont. Here are some common sources of uncertainty in AI: Uncertainty in AI communication: AI systems communicate with humans through natural language processing or computer vision. However, language and visual cues can be ambiguous or misunderstood, leading to uncertainty in effective communication between humans and AI systems. Image credits: google.com Module Code & Module Title Slide Title SLIDE 11 Types of Uncertainty in AI Epistemic uncertainty: epistemic uncertainty refers to the lack of knowledge or information about a model. Parameter uncertainty: this type of uncertainty is specific to probabilistic models such as Bayesian neural networks. It reflects uncertainty about the values of the model parameters and is characterized by probability distributions over these parameters. Image credits: google.com Module Code & Module Title Slide Title SLIDE 12 Types of Uncertainty in AI…cont. Uncertainty in decision making: In reinforcement learning, for example, agents often have to make decisions in environments with uncertain outcomes, leading to uncertainty in decision making. Uncertainty in natural language understanding: In natural language processing (NLP), understanding and producing human language can be inherently uncertain due to ambiguity, polysemy (multiple meanings), and contextual interpretations. Image credits: google.com Module Code & Module Title Slide Title SLIDE 13 Types of Uncertainty in AI…cont. Uncertainty in Probabilistic Inference: Bayesian methods and probabilistic graphical models are widely used in AI to model uncertainty. Uncertainty can arise from the process of probabilistic inference itself and can affect the reliability of model predictions. Image credits: google.com Module Code & Module Title Slide Title SLIDE 14 Techniques for Addressing Uncertainty in AI Probabilistic logic programming (PLP) is useful for computer programmers when they are not completely sure of the facts and rules they are working with. PLP uses probabilities to help them make decisions and learn from data. Fuzzy logic programming deals with uncertainties in logic programming, there is a method called Fuzzy Logic Programming (FLP).It combines regular logic with something called “fuzzy” logic. Image credits: google.com Module Code & Module Title Slide Title SLIDE 15 Techniques for Addressing Uncertainty in AI…cont. Hybrid logic programming (HLP) deals with situations where things are unclear or don’t quite fit together in logic programming. It allows them to use various types of logic to express and make sense of rules and complex facts. Image credits: google.com Module Code & Module Title Slide Title SLIDE 16 Ways to Solve Problems with Uncertain Knowledge Probability plays a central role in AI because it provides a formal framework for dealing with uncertainty. AI systems use probabilistic models and reasoning to make informed decisions, assess risk, and quantify uncertainty so that they can operate effectively in complex and uncertain real-world scenarios. In probabilistic reasoning, there are two ways to solve problems when we are unsure of information: 1. Bayesian Rule 2. Bayesian statistics Image credits: google.com Module Code & Module Title Slide Title SLIDE 17 Ways to Solve Problems with Uncertain Knowledge…cont. Mathematically, Bayes’ theorem is expressed as follows: Here, The posterior probability, represented by P(A|B), is the chance of event A happening when event B has happened. P(B|A) shows how likely event B is when event A has already happened. The prior probability, P(A), is the initial chance of event A happening before any new information is considered. P(B) is the probability of event B happening, whether or not event A has happened. Image credits: google.com Module Code & Module Title Slide Title SLIDE 18 Ways to Solve Problems with Uncertain Knowledge…cont. Bayesian statistics is a type of statistics that uses probability to analyze data. The framework helps us draw conclusions and estimate probabilities based on data and prior knowledge. Bayesian statistics has been used in various fields to deal with uncertainty and make informed decisions. It has been used in environmental modeling, social sciences, and medical research. Module Code & Module Title Slide Title SLIDE 19 Importance of Understanding Uncertainty in AI Reliable decision-making: AI applications often involve critical decisions, such as medical diagnoses or autonomous vehicle navigation. Recognizing uncertainty ensures that AI systems make reliable, risk-aware decisions. Quantification of trust: Quantifying uncertainty allows AI models to express confidence in their predictions. This information is invaluable for users to assess the reliability of AI-based recommendations. Image credits: google.com Module Code & Module Title Slide Title SLIDE 20 Importance of Understanding Uncertainty in AI..cont. Ethical considerations: In AI ethics, transparency and accountability are crucial. Understanding uncertainty enables developers and users to better understand AI decisions, promoting trust and the responsible use of AI. Robustness: AI systems that are able to deal with uncertainty are more resilient to unforeseen circumstances and fluctuations in input data, contributing to their overall robustness. Image credits: google.com Module Code & Module Title Slide Title SLIDE 21 Quick Review Question Name at least two (2) Importance of understanding uncertainty in AI. Image credits: google.com Module Code & Module Title Slide Title SLIDE 22 Approaches to Calculate Uncertainty in AI Frequentist approach Bayesian approach Image credits: google.com Module Code & Module Title Slide Title SLIDE 23 Approaches to Calculate Uncertainty in AI..cont. The frequentist approach to uncertainty is based on the idea of long-term frequencies. The frequentist approach to uncertainty is often used in statistical inference, i.e. the use of data to make inferences about a population. Image credits: google.com Module Code & Module Title Slide Title SLIDE 24 Approaches to Calculate Uncertainty in AI..cont. The Bayesian approach to uncertainty is based on the idea of the degree of belief. The Bayesian approach to dealing with uncertainty is often used in the field of machine learning, the area of computer science that deals with the ability of machines to learn from data. Image credits: google.com Module Code & Module Title Slide Title SLIDE 25 Approaches to Calculate Uncertainty in AI..cont. Suppose we toss a coin and get heads. What is the probability that the next toss will also be heads? Image credits: google.com Module Code & Module Title Slide Title SLIDE 26 Comparison Frequentist and Bayesian Approaches The frequentist and Bayesian approaches differ in several important respects: Image credits: google.com Module Code & Module Title Slide Title SLIDE 27 Comparison Frequentist and Bayesian Approaches…cont. Question 1: A fair coin is tossed 100 times. heads comes out 60 times. What is the probability of getting heads on the next toss? Answer for Frequentist approach: ? Bayesian Approach: ? Image credits: google.com Module Code & Module Title Slide Title SLIDE 28 Comparison Frequentist and Bayesian Approaches…cont. Question 1: A fair coin is tossed 100 times. heads comes out 60 times. What is the probability of getting heads on the next toss? Answer for Frequentist approach: The probability of getting heads in the next throw is 0.6, because this is the proportion of cases in which heads appeared in the 100 throws. Image credits: google.com Module Code & Module Title Slide Title SLIDE 29 Comparison Frequentist and Bayesian Approaches…cont. Question 1: Answer for Bayesian approach: Image credits: google.com Module Code & Module Title Slide Title SLIDE 30 Comparison Frequentist and Bayesian Approaches…cont. Question 2: A survey of 1000 people found that 60% of them support a particular candidate. You believe that there is a 10% chance that the survey is biased. What is your posterior probability that the true proportion of people who support the candidate is between 50% and 70%? Answer for Frequentist approach: ? Bayesian Approach: ? Image credits: google.com Module Code & Module Title Slide Title SLIDE 31 Comparison Frequentist and Bayesian Approaches…cont. Question 2: Answer for Frequentist approach: where: p is the sample proportion (in this case, 0.6) n is the sample size (in this case, 1000) margin of error of approximately 0.03 This means that we are 95% confident that the true proportion of people who support the candidate is between 0.57 and 0.63. Module Code & Module Title Slide Title Image credits: google.com SLIDE 32 Comparison Frequentist and Bayesian Approaches…cont. Question 2: Answer for Bayesian approach: where: P(A) is the prior probability of A P(B) is the probability of B P(A | B) is the posterior probability of A given B P(B | A) is the likelihood of B given A Module Code & Module Title Slide Title Image credits: google.com SLIDE 33 Comparison Frequentist and Bayesian Approaches…cont. Question 2: Answer for Bayesian approach: posterior probability of approximately 0.95 This means that after considering the results of the survey, we are 95% confident that the true proportion of people who support the candidate is between 50% and 70%. Module Code & Module Title Slide Title Image credits: google.com SLIDE 34 Quick Review Question What are the main differences between the frequentist and Bayesian approaches? Image credits: google.com Module Code & Module Title Slide Title SLIDE 35 Quick Review Question What are the main differences between the frequentist and Bayesian approaches? Image credits: google.com Module Code & Module Title Slide Title SLIDE 36 Question and Answer Session Module Code & Module Title Slide Title SLIDE 37 Module Code & Module Title Slide Title SLIDE 38 Summary / Recap of Main Points Why is it so important to learn uncertainty in AI? 1. It reflects the complexity of the real world. 2. Improves decision making 3. Improves model reliability 4. Facilitates better training and evaluation 5. Supports ethical AI development Module Code & Module Title Slide Title SLIDE 39 What To Expect Next Week In Class Preparation for Class Knowledge representation - Knowledge Representation – Uncertainty Logical Systems Module Code & Module Title Slide Title SLIDE 40