machine learning: machine learning: introduction introduction

Click here to load reader

Download Machine Learning: Machine Learning: Introduction Introduction

Post on 26-Jan-2015




7 download

Embed Size (px)




  • 1. Machine Learning:IntroductionDr Valentina Plekhanova University of Sunderland, UKThe Field of Machine LearningThe goal of machine learning is to develop methods, techniques and tools for building intelligent learning machines, that can solve the problem in combination with machines, examples. an available data set of training examples.Note: A learning machine or an algorithm does not solve the problem directly. Valentina PlekhanovaMachine Learning: Introduction 2 The Meaning of LearningWhen a learning machine improves its performance at a given task over time, without reprogramming, it can be said to have learned something. Learning: improvement of performance with experience at a given taskimprove over task Twith respect to performance measure Pbased on experience E Example from Machine/Computer Vision field:learn to recognise objects from a visual scene or an image T: identify all objectsP: accuracy (e.g. a number of objects correctly recognised)E: a database of objects recordedValentina PlekhanovaMachine Learning: Introduction 31

2. Components of a Learning ProblemTask: the behaviour or task thats being improved, e.g.classification, object recognition, acting in an environment. Data: the experiences that are being used to improveperformance in the task. Measure of improvements: How can the improvementbe measured? Examples:Provide more accurate solutions (e.g. increasing the accuracy inprediction)Cover a wider range of problemsObtain answers more economically (e.g. improved speed)Simplify codified knowledgeNew skills that were not presented initially Valentina Plekhanova Machine Learning: Introduction 4 Notes onLearning Problem & Learning Tasks Learning Problem: Precise model that describes:Problemwhat is to be learnedhow it is donewhat measures are to be used in analysing and comparing theperformance of different solutions.Learning tasks are formal definitions of what we want towhat learn. learn.Valentina Plekhanova Machine Learning: Introduction 5 Learning Algorithm In Machine learning: Learning algorithm - Learning Agent. Note: A learning agent is not concerned with constructing an exact answer (e.g. an exact concept definition). Valentina Plekhanova Machine Learning: Introduction 6 2 3. Learning Feedback Learning feedback can be provided by the systemenvironment or the agents themselves. Supervised learning: specifies the desired activities/objectives oflearning feedback from a teacher Unsupervised learning: no explicit feedback is provided and theobjective is to find out useful and desired activities on the basis of trial-and-errorand self-organisation processes a passive observerReinforcement learning: specifies the utility of the actual activity ofthe learner and the objectives is to maximise this utility feedback from a critic[ Sen & Weiss, 1999 ] Valentina Plekhanova Machine Learning: Introduction 7Ways of LearningRote learning, i.e. learning from memory; in a mechanical way Learning from examples and by practice Learning from instructions/advice/explanations Learning by analogy Learning by discovery Valentina Plekhanova Machine Learning: Introduction 8Inductive & Deductive LearningInductive Learning: Reasoning from a set of examplesLearning to produce a general rules. The rules should be applicable to new examples, but there is no guarantee that the result will be correct. Deductive Learning: Reasoning from a set of knownLearning facts and rules to produce additional rules that are guaranteed to be true. Valentina Plekhanova Machine Learning: Introduction 9 3 4. Assessment of Learning AlgorithmsThe most common criteria for learning algorithms assessments are:Accuracy (e.g. percentages of correctly classified +s and s) Efficiency (e.g. examples needed, computational tractability) Robustness (e.g. against noise, against incompleteness) Special requirements (e.g. incrementality, concept drift)incrementality, Concept complexity (e.g. representational issues examples & BK) Transparency (e.g. comprehensibility for the human user) Valentina Plekhanova Machine Learning: Introduction 10 Some Theoretical Settings Inductive Logic Programming (ILP)Probably Approximately Correct (PAC) LearningLearning as Optimisation (Tasks: Reinforcement Learning) Bayesian Learning Valentina Plekhanova Machine Learning: Introduction 11Each task can be related to one or more methods Concept Learning: an Example Methods: Lazy Learning; Incremental Decision Tree Methods: Learning; Boosting; Support Vector Machines (SVM).Theories: e.g. Inductive Logic Programming (ILP); Theories: Probably Approximately Correct (PAC) Learning; Bayesian Learning; Statistical Learning. Valentina Plekhanova Machine Learning: Introduction 124 5. SummaryA Model of Learning: Key AspectsLearner: who or what is doing the learning, e.g. an algorithm, acomputer program.Domain: what is being learned, e.g. a function, a concept.Goal: why the learning is done.Representation: the way the objects to be learned are represented. Algorithmic Technology: the algorithmic framework to be used, e.g.decision trees, lazy learning, artificial neural networks, support vectormachines. [Sen & Weiss, 1999 ]Valentina Plekhanova Machine Learning: Introduction13SummaryA Model of Learning: Key Aspects Information Source: the information/training data the program uses forlearning, e.g. positive/negative examples, feedback from actions. Training/Learning Scenario: the description of the learning process, e.g.interactive, supervised / unsupervised, etc.Prior Knowledge: what is known in advance about the domain, e.g.about specific properties of the concepts to be learned. Success Criteria: the criteria for successful learning.Performance: e.g. the amount of time, space and computational powerneeded in order to learn a certain task. [Sen & Weiss, 1999 ]Valentina Plekhanova Machine Learning: Introduction14 5

View more