Knowledge Representation Reading: Chapter 10.1-10.2 Classes offered in spring Vision/Robotics NLP/Speech 4706 Spoken Language Processing (Hirschberg) 6998-3 Natural Language Processing for the Web (McKeown) http://cs.columbia.edu/~kathy/NLPforWeb.htm Machine Learning 6732 Computational Imaging (Nayar) 6735 Visual Databases (Kender) 6994-4 Computational Photography (Belhumeur) Other 4771 Machine Learning (Jebara) 4172 3D User Interfaces (Feiner) 2 Homework What’s important (i.e., this will be used in determining your grade): Finding features that make a difference You should expect to do some digging in the data Find a feature that requires manipulation of data Reformatting of data to provide a more consistent feature (e.g., gender, profession) Turn in a sample of your data file in ARFF format with the features you ended up using (5 instances only) Turn in a Weka log documenting the series of steps you used to arrive at your model We want the experimentation that backs up your claims in the report 3 When you notice a cat in profound meditation, The reason, I tell you is always the same: His Mind is engaged in a rapt contemplation of the Thought, of the thought, of the thought of his name. T.S. Eliot Old Possum’s Book of Practical Cats 4 Five Roles that KR plays A surrogate for some part of the real world A set of ontological commitments A fragmentary theory of intelligent reasoning A medium for pragmatically efficient computation A medium of human expression 5 KR as a surrogate Agents “reason” about models of the world, rather than the world itself Deduce properties without having to directly gather information from the world Predict consequences of potential actions rather than performing the actions directly 6 We always have two universes of discourse – call them “physical” and “phenomenal”, or what you will – one dealing with questions of quantitative and formal structure, the other with those qualities that constitute a “world.” All of us have our own distinctive mental worlds, our own inner journeyings and landscapes, and these, for most of us, require no clear neurological “correlate.” 7 Example 8 9 10 11 12 Given a representation What are its semantics? What is the meaning of its structures? What does it mean/refer to? Fidelity – how accurate is it? 13 Areas of Activity Designing formats for expressing information Encoding knowledge (knowledge engineering) Mostly “general purpose” knowledge representations (e.g., first order logic) Mostly identifying and describing conceptual vocabulary (ontologies) Declarative representations are the focus of KR technology Knowledge that is domain specific but task independent 14 Example of representations 15 KRs are never a complete model When modeling the real world, KRs are always imperfect “Consequently, even with a sound reasoning, incorrect conclusions are inevitable” 16 Ontological commitments A KR is a set of ontological commitments An ontology is a theory of what exists in the world Classes, objects, relations, attributes, properties, constraints, special individuals, etc. Provides a vocabulary for expressing knowledge 17 Example of KR structures 18 A Vocabulary for the World A KR makes a commitment to a particular ontology To describing the world with particular terms Taxonomy of the world Promiscuity vs. perspicacity 19 Example of a Vocabulary for the World 20 OMEGA http://omega.isi.edu:8007/index http://omega.is.edu/doc/browsers.h tml 21 22 Ontological Commitments “The commitments are in effect a strong pair of glasses that determine what we can see, bringing some part of the world into sharp focus, at the expense of blurring other parts.” A KR is not just a data structure “Part of what makes a language representational is that it carries meaning, I.e., there is a correspondence between its constructs and things in the external world.” 23 KR as a theory of reasoning Many knowledge representations offer fragmentary theories of intelligent reasoning Humans employ multiple strategies for representing and reasoning about the world 24 25 Impact of reasoning theory The selected theory affects methods and possible inference Only certain facts can be inferred Some methods of inference are “sanctioned” or illegal A better method of reasoning than undirected search Theory provides “recommendations” for strategies of inference 26 Efficient Computation Some work has focused on knowledge content and what could, in principle, be derived from it without concern for efficiency Sound Complete Tradeoff between efficiency and expressiveness 27 Heuristic Adequacy Providing a representation that supports adequately efficient problem solving Early heuristic systems Any-time computations 28 KR as a medium for human expression An intelligent system must have KRs that can be understood by humans We need to be able to encode knowledge in the knowledge base without significant effort We need to be able to understand what the system knows and how it draws it conclusions 29 Open Issues for KR How can a reasoning mechanism generate implicit knowledge? How can a system use knowledge to influence its behavior? How is incomplete or noisy knowledge handled? How can practical results be obtained when reasoning is intractable? 30 Different Forms of Knowledge Representation Logical representation schemes Procedural representation schemes Network representation schemes Structured representation schemes 31