Machine Learning

Course : Machine Learning Lecture : Joseph Credit Hours : 3 Credits Semester Sessions : 3 Semester, 2021/2022 Session Prepared By : Kenzo LECTURE NOTES Which is friendly to humans. You don't have to have a physical body, you don't have to have artificial intelligence are actually programs that have mathematical form (instructions); we call it an agent. Different from ordinary programs that produce actions based on instructions, the goal of artificial intelligence is to create program capable of programming (the program output is a programs). In theory, a program is an automaton3 that executes something instructions. As with programs in general, intelligence agents cannot atan also executes an instruction. What makes it different from pro-Ordinary gram is the ability to learn. "Learning" is what is meant is not the same as the human learning process. Machines are able to learn if it is able to update parameters (explained in more detail later). In the scientific field of artificial intelligence, we want to create agents capable of doing work that requires human intelligence. Note, here it is called human intelligence; Even animals are intelligent, but human and animal intelligence is different; what we want to approximate is human intelligence. However, human intelligence is difficult to define because it has many aspects, for example reasoning (logic), the ability to communicate hasa, art, etc. Because human intelligence has many dimensions, we can try to solve the problem in smaller sub-fields (divide and conquer ). Even now, researchers don't know yet definitely what makes humans intelligent, what is truly intelligent, and how humans can become intelligent. Thus, intelligent science Artificial foundations are interdisciplinary, containing: psychology, linguistics, computer science puter, biology, etc. If you ask whether a deterministic program can called artificial intelligence, the answer is "yes", to some extent (up to certain level) because it meets the dimensions of acting rationally (explained in subchapter 1.2). The main problems in the field of artificial intelligence consist of (from classic to more modern) 5, that is:  Planning. Given a start state and a goal state, the agent must plan sequence of actions to change the start state to the goal state. Example per- The planning problem is planning the travel route from city A to city B. It could be that, when planning action sequences, there are obstacles (con-strains) that must be optimized.  Knowledge representation, namely representing internal knowledge formal form. With this formal representation, we can do inference with symbolic logical operations, for example prepositional logic, first-order logic, fuzzy theory, abductive reasoning,  Machine learning, namely techniques for inference data with a mathematical approach. The point of machine learning is to create (mathematical) models that reflect data patterns (along If you read this book, you will understand better). This is a discussion main topic of this book.  Multi-agent system, namely a system that has many active agents interact with each other to solve problems. Agent one do a certain thing, then work together to achievesolve bigger problems (can't be solved alone).  And so on, please refer to the Association conference topicfor the Advancement of Artificial Intelligence (AAAI). Note, the sub-sciences of knowledge representation and machine learning- ing both perform inference, but on different representations. Inference in the scientific field of knowledge representation includes how to (steps and process) get a decision san, given the premise. In machine learning, what is meant by inference is more focuses on the realm of variable relationships. For example, whether sales will be increases if we increase marketing costs. If you remember high school mathematics course (prepositional logic), you realize that making creating intelligent systems using symbolic knowledge representation is difficult. We have to define terms, logical rules, etc. Not to mention we have to define rules manually. Representation of knowledge Traditional methods are considered relatively less scalable, especially if we use work with big data. Meanwhile, machine learning is on area of data/science/knowledge representation in mathematical form because The science of machine learning is derived from mathematics and statistics. Nowadays, we are blessed with a lot of data (even unlimited), machine learning techniques become intuitive to perform inference on big data. This is what causes machine learning-ing has become popular because the construction of inference models can be done as automatic way.