Agent-Based HOL Reasoning 1

Similar documents
The World s Most Widely Applicable Modal Logic Theorem Prover and its Associated Infrastructure

Is it Reasonable to Employ Agents in Automated Theorem Proving?

Gödel s Proof of God s Existence

LCF + Logical Frameworks = Isabelle (25 Years Later)

Formal verification of IA-64 division algorithms

Combining and Automating Classical and Non-Classical Logics in Classical Higher-Order Logics

Systematic Verification of the Modal Logic Cube in Isabelle/HOL

Church and Curry: Combining Intrinsic and Extrinsic Typing

Tutorial on Reasoning in Expressive Non-Classical Logics with Isabelle/HOL

Classical First-Order Logic

Computational Logic and the Quest for Greater Automation

Designing and Evaluating Generic Ontologies

Computer Proof Assistants and Univalent Foundations of Mathematics

Developing Modal Tableaux and Resolution Methods via First-Order Resolution

Requirements Validation. Content. What the standards say (*) ?? Validation, Verification, Accreditation!! Correctness and completeness

An Introduction to Proof Assistants

26. Januar Introduction to Computational Semantics

Utilizing Church s Type Theory as a Universal Logic 1

Modular Architecture for Proof Advice AITP Components

Interactive Theorem Provers

The Legacy of Hilbert, Gödel, Gentzen and Turing

02 Propositional Logic

Computer-Checked Meta-Logic

Automated theorem proving in algebra

Validating QBF Invalidity in HOL4

NICTA Advanced Course. Theorem Proving Principles, Techniques, Applications

Introduction to Turing Machines

Matching Logic: Syntax and Semantics

NICTA Advanced Course. Theorem Proving Principles, Techniques, Applications. Gerwin Klein Formal Methods

System Description: ara An Automatic Theorem Prover for Relation Algebras

Classical First-Order Logic

via Topos Theory Olivia Caramello University of Cambridge The unification of Mathematics via Topos Theory Olivia Caramello

First-Order Theorem Proving and Vampire. Laura Kovács (Chalmers University of Technology) Andrei Voronkov (The University of Manchester)

Automated and Human Proofs in General Mathematics: An Initial Comparison

The Curry-Howard Isomorphism

Interactive Theorem Proving in Industry

Automata Theory. Definition. Computational Complexity Theory. Computability Theory

Formalization of Mathematics: why Algebraic Topology?

Modal Logics. Most applications of modal logic require a refined version of basic modal logic.

Theorem Proving for Verification

The Formal Proof Susan Gillmor and Samantha Rabinowicz Project for MA341: Appreciation of Number Theory Boston University Summer Term

Mathematics, Proofs and Computation

Computational Logic for Computer Science

Pei Wang( 王培 ) Temple University, Philadelphia, USA

Vinter: A Vampire-Based Tool for Interpolation

Translating a Satallax Refutation to a Tableau Refutation Encoded in Coq

The roots of computability theory. September 5, 2016

SAT/SMT/AR Introduction and Applications

Formal Logic and Deduction Systems

Nonclassical logics (Nichtklassische Logiken)

Learning Goals of CS245 Logic and Computation

Recursion Theory. Joost J. Joosten

Undecidability in Epistemic Planning

λ Slide 1 Content Exercises from last time λ-calculus COMP 4161 NICTA Advanced Course Advanced Topics in Software Verification

Temporal Logic - Soundness and Completeness of L

Towards a Mechanised Denotational Semantics for Modelica

Lecture 11: Gödel s Second Incompleteness Theorem, and Tarski s Theorem

Seamless Model Driven Development and Tool Support for Embedded Software-Intensive Systems

Handbook of Logic and Proof Techniques for Computer Science

Automating Access Control Logics in Simple Type Theory with LEO-II

Type Theory and Constructive Mathematics. Type Theory and Constructive Mathematics Thierry Coquand. University of Gothenburg

Combining Deduction and Algebraic Constraints for Hybrid System Analysis

Dynamic Semantics. Dynamic Semantics. Operational Semantics Axiomatic Semantics Denotational Semantic. Operational Semantics

A Crisp Representation for Fuzzy SHOIN with Fuzzy Nominals and General Concept Inclusions

Outline. Overview. Syntax Semantics. Introduction Hilbert Calculus Natural Deduction. 1 Introduction. 2 Language: Syntax and Semantics

Higher-Order Automated Theorem Provers

From Constructibility and Absoluteness to Computability and Domain Independence

Lecture Notes: Axiomatic Semantics and Hoare-style Verification

CS187 - Science Gateway Seminar for CS and Math

Mappings For Cognitive Semantic Interoperability

Research Statement Christopher Hardin

Evidential Paradigm and Intelligent Mathematical Text Processing

Proof Theory and Subsystems of Second-Order Arithmetic

Analogical Reasoning with SMT and HDTP

Automated Reasoning Lecture 5: First-Order Logic

Uniform Schemata for Proof Rules

6. Logical Inference

Higher-Order Aspects and Context in SUMO

Higher-Order Aspects and Context in SUMO

Last Time. Inference Rules

Intel s Successes with Formal Methods

Mathematical Logic. Introduction to Reasoning and Automated Reasoning. Hilbert-style Propositional Reasoning. Chiara Ghidini. FBK-IRST, Trento, Italy

Abstractions and Decision Procedures for Effective Software Model Checking

Linguistics and logic of common mathematical language I. Peter Koepke and Merlin Carl, Mathematical Institute Universität Bonn

Logic in Computer Science. Frank Wolter

Connection method - Scholarpedia

Combining Induction Axioms By Machine

Beyond First-Order Logic

Semantic Equivalences and the. Verification of Infinite-State Systems 1 c 2004 Richard Mayr

Propositions as Types

1. A motivation for algebraic approaches to logics

Inductive Theorem Proving

Propositional Logic: Logical Agents (Part I)

Using Gentzen system techniques with existing automated theorem provers and rewriting tools

COMP 2600: Formal Methods for Software Engineeing

Valentin Goranko Stockholm University. ESSLLI 2018 August 6-10, of 29

Can we measure the difficulty of an optimization problem?

First-Order Theorem Proving and Vampire

Formal Verification of Mathematical Algorithms

Going Polymorphic TH1 Reasoning for Leo-III

Transcription:

Agent-Based HOL Reasoning 1 Alexander Steen Max Wisniewski Christoph Benzmüller Freie Universität Berlin 5th International Congress on Mathematical Software (ICMS 2016) 1 This work has been supported by the DFG under grant BE 2501/11-1 (Leo-III).

Talk outline 1. Automated Theorem Proving 2. Higher-Order Logic 3. Agents 4. Leo-III Prover, Agent-Based HOL Reasoning, ICMS 2016 2

Automated Theorem Proving (ATP) ATP as computer-assisted reasoning No human interaction needed Autonomous proof search Output of result/proof Translate problem to formal representation... underlying theory as axioms... claim/goal as conjecture System result: Theorem, counter satisfiable,... and a proof? Further approaches: Interactive Theorem Proving, Semi-Interactive... Agent-Based HOL Reasoning, ICMS 2016 3

Automated Theorem Proving (ATP) ATP as computer-assisted reasoning No human interaction needed Autonomous proof search Output of result/proof Translate problem to formal representation... underlying theory as axioms... claim/goal as conjecture System result: Theorem, counter satisfiable,... and a proof? Further approaches: Interactive Theorem Proving, Semi-Interactive... Agent-Based HOL Reasoning, ICMS 2016 3

Automated Theorem Proving (ATP) ATP as computer-assisted reasoning No human interaction needed Autonomous proof search Output of result/proof Translate problem to formal representation... underlying theory as axioms... claim/goal as conjecture System result: Theorem, counter satisfiable,... and a proof? Further approaches: Interactive Theorem Proving, Semi-Interactive... Agent-Based HOL Reasoning, ICMS 2016 3,

Recent successes in Automated Deduction Proof of Four color theorem [AppelHaken, 1976] [Gonthier, 2005] Proof of Kepler s conjecture (Flyspeck project) [Hales et al., 2014] Formal inspection (verification) of Gödel s Ontological Argument and various versions of it [BenzmüllerWoltzenlogel Paleo, IJCAI,2016] Agent-Based HOL Reasoning, ICMS 2016 4

Formalization: Hands on More concrete application example Cantor s surjective theorem: There exists no surjective function f from a set to its power set What is an appropriate formalization? Can it be proved automatically? Agent-Based HOL Reasoning, ICMS 2016 5

Formalization: Hands on More concrete application example Cantor s surjective theorem: There exists no surjective function f from a set to its power set What is an appropriate formalization? Can it be proved automatically? Agent-Based HOL Reasoning, ICMS 2016 5

Higher Order Logic (HOL) Due to Alonzo Church s Simple type theory (a typed λ-calculus) [Church, J.Symb.L., 1940] Simple types T generated by base types and Typically, base types are o and i Agent-Based HOL Reasoning, ICMS 2016 6

Higher Order Logic (HOL) Due to Alonzo Church s Simple type theory (a typed λ-calculus) [Church, J.Symb.L., 1940] Simple types T generated by base types and Typically, base types are o and i Type of truth-values Agent-Based HOL Reasoning, ICMS 2016 6

Higher Order Logic (HOL) Due to Alonzo Church s Simple type theory (a typed λ-calculus) [Church, J.Symb.L., 1940] Simple types T generated by base types and Typically, base types are o and i Type of individuals Agent-Based HOL Reasoning, ICMS 2016 6

Higher Order Logic (HOL) Due to Alonzo Church s Simple type theory (a typed λ-calculus) [Church, J.Symb.L., 1940] Simple types T generated by base types and Typically, base types are o and i Terms defined by (α T) s, t ::=p α X α Formulae of HOL are those terms with type o Semantics well-understood (not mentioned here) Agent-Based HOL Reasoning, ICMS 2016 6

Higher Order Logic (HOL) Due to Alonzo Church s Simple type theory (a typed λ-calculus) [Church, J.Symb.L., 1940] Simple types T generated by base types and Typically, base types are o and i Terms defined by (α T) s, t ::=p α X α (λx α.s β ) α β (s α β t α ) β Formulae of HOL are those terms with type o Semantics well-understood (not mentioned here) Agent-Based HOL Reasoning, ICMS 2016 6

Higher Order Logic (HOL) Due to Alonzo Church s Simple type theory (a typed λ-calculus) [Church, J.Symb.L., 1940] Simple types T generated by base types and Typically, base types are o and i Terms defined by (α T) s, t ::=p α X α (λx α.s β ) α β (s α β t α ) β ( o o s o ) o (s o o o o t o ) o ( α X α.s o ) o Formulae of HOL are those terms with type o Semantics well-understood (not mentioned here) Agent-Based HOL Reasoning, ICMS 2016 6

Higher Order Logic (HOL) Due to Alonzo Church s Simple type theory (a typed λ-calculus) [Church, J.Symb.L., 1940] Simple types T generated by base types and Typically, base types are o and i Terms defined by (α T) s, t ::=p α X α (λx α.s β ) α β (s α β t α ) β ( o o s o ) o (s o o o o t o ) o Π α (α o) o (λx α. s o ) α o Formulae of HOL are those terms with type o Semantics well-understood (not mentioned here) Agent-Based HOL Reasoning, ICMS 2016 6

Higher Order Logic (HOL) Due to Alonzo Church s Simple type theory (a typed λ-calculus) [Church, J.Symb.L., 1940] Simple types T generated by base types and Typically, base types are o and i Terms defined by (α T) s, t ::=p α X α (λx α.s β ) α β (s α β t α ) β ( o o s o ) o (s o o o o t o ) o ( α X α.s o ) o Formulae of HOL are those terms with type o Semantics well-understood (not mentioned here) Agent-Based HOL Reasoning, ICMS 2016 6

Why HOL? Why HOL? (Pragmatically) expressive language Concise expressions Higher-order quantification Anonymous functions ( built-in comprehension) Allows easy representation of common notions (e.g. induction principle) Allows encoding of various non-classical logics (e.g. modal logics, free logic,...) Previous example There exists no surjective function from a set to its power set Agent-Based HOL Reasoning, ICMS 2016 7

Why HOL? Why HOL? (Pragmatically) expressive language Concise expressions Higher-order quantification Anonymous functions ( built-in comprehension) Allows easy representation of common notions (e.g. induction principle) Allows encoding of various non-classical logics (e.g. modal logics, free logic,...) Previous example There exists no surjective function from a set to its power set Agent-Based HOL Reasoning, ICMS 2016 7

Why HOL? Why HOL? (Pragmatically) expressive language Concise expressions Higher-order quantification Anonymous functions ( built-in comprehension) Allows easy representation of common notions (e.g. induction principle) Allows encoding of various non-classical logics (e.g. modal logics, free logic,...) Previous example There exists no surjective function from a set to its power set Agent-Based HOL Reasoning, ICMS 2016 7

Why HOL? Why HOL? (Pragmatically) expressive language Concise expressions Higher-order quantification Anonymous functions ( built-in comprehension) Allows easy representation of common notions (e.g. induction principle) Allows encoding of various non-classical logics (e.g. modal logics, free logic,...) Previous example There exists no surjective function from a set to its power set F ι (ι o). Y ι o. X ι. F X = Y Agent-Based HOL Reasoning, ICMS 2016 7

Why HOL? Why HOL? (Pragmatically) expressive language Concise expressions Higher-order quantification Anonymous functions ( built-in comprehension) Allows easy representation of common notions (e.g. induction principle) Allows encoding of various non-classical logics (e.g. modal logics, free logic,...) Previous example There exists no surjective function from a set to its power set F ι (ι o). Y ι o. X ι. F X = Y Agent-Based HOL Reasoning, ICMS 2016 7

Why HOL? Why HOL? (Pragmatically) expressive language Concise expressions Higher-order quantification Anonymous functions ( built-in comprehension) Allows easy representation of common notions (e.g. induction principle) Allows encoding of various non-classical logics (e.g. modal logics, free logic,...) Previous example There exists no surjective function from a set to its power set F ι (ι o). Y ι o. X ι. F X = Y Agent-Based HOL Reasoning, ICMS 2016 7

Call the Agents Reasoning in HOL is involved Complicated routines (some of those undecidable) Different (orthogonal) reasoning techniques Tableaux methods (Satallax) Resolution methods (LEO-II) First-order encoding methods (tptp_isabelle) Paramodulation methods (?) Not (yet) clear, if there s a superior approach Agent-Based HOL Reasoning, ICMS 2016 8

Call the Agents Reasoning in HOL is involved Complicated routines (some of those undecidable) Different (orthogonal) reasoning techniques Tableaux methods (Satallax) Resolution methods (LEO-II) First-order encoding methods (tptp_isabelle) Paramodulation methods (?) Not (yet) clear, if there s a superior approach Agent-Based HOL Reasoning, ICMS 2016 8

Call the Agents Reasoning in HOL is involved Complicated routines (some of those undecidable) Different (orthogonal) reasoning techniques Tableaux methods (Satallax) Resolution methods (LEO-II) First-order encoding methods (tptp_isabelle) Paramodulation methods (?) Not (yet) clear, if there s a superior approach Why not combine them all? Have the best of all worlds! Motivation of Leo-III Employ specialists as independent agents Let them find a proof cooperatively Independence allows parallel execution. Agent-Based HOL Reasoning, ICMS 2016 8

Architecture of Leo-III Shared blackboard architecture Independent agents collaboratively acting on it Agents may be internal or external Coordination of agents by combinatorical auction game scheduling Agent-Based HOL Reasoning, ICMS 2016 9

Current state of Leo-III Built around a paramodulation-based HOL calculus... currently as sequential loop At any point the loop may request assistance from external agents Internal agents Normalization variants Relevance filtering (pre-processing) Sequential loops Goal state checker External agents Leo-II Satallax exemplary every TPTP-compatible prover can be used Agent-Based HOL Reasoning, ICMS 2016 10

Current state of Leo-III Built around a paramodulation-based HOL calculus... currently as sequential loop At any point the loop may request assistance from external agents Internal agents Normalization variants Relevance filtering (pre-processing) Sequential loops Goal state checker External agents Leo-II Satallax exemplary every TPTP-compatible prover can be used Agent-Based HOL Reasoning, ICMS 2016 10

Current state of Leo-III Built around a paramodulation-based HOL calculus... currently as sequential loop At any point the loop may request assistance from external agents Internal agents Normalization variants Relevance filtering (pre-processing) Sequential loops Goal state checker External agents Leo-II Satallax exemplary every TPTP-compatible prover can be used Agent-Based HOL Reasoning, ICMS 2016 10

First assessment First assessment is positive... Competed in this year s CADE ATP System Competition (CASC) Experimental paramodulation calculus seems feasible Cooperation with external provers beneficial Agent-Based HOL Reasoning, ICMS 2016 11

First assessment First assessment is positive... Competed in this year s CADE ATP System Competition (CASC) Experimental paramodulation calculus seems feasible Cooperation with external provers beneficial... but Poor parameter settings No first-order cooperation yet A lot of things to do until competitive Agent-Based HOL Reasoning, ICMS 2016 11

Live Demo Demo Agent-Based HOL Reasoning, ICMS 2016 12

Modifications Sequential loop agents only for the current version Split internal saturation into more fine-grained agent tasks Use machine learning techniques for coordination of agents, search space traversal Include more specialist systems/external agents: First-order provers SAT/SMT solvers for internal reasoning assistance Consistency checkers (e.g. nitpick) Unification specialists Agent-Based HOL Reasoning, ICMS 2016 13

Modifications Sequential loop agents only for the current version Split internal saturation into more fine-grained agent tasks Use machine learning techniques for coordination of agents, search space traversal Include more specialist systems/external agents: First-order provers SAT/SMT solvers for internal reasoning assistance Consistency checkers (e.g. nitpick) Unification specialists Agent-Based HOL Reasoning, ICMS 2016 13

Perspective Leo-III s underlying datastructure/architecture framework as a stand-alone package For the implementation of HO reasoners on top of it Easily extensible by agent-based design Already available! On the long run: Reasoning-as-a-service for domain specific tools Many non-classical logics through internal embedding into HOL Machine learning components could also be re-used by applications Natural language processing Algebra systems,... Agent-Based HOL Reasoning, ICMS 2016 14

Perspective Leo-III s underlying datastructure/architecture framework as a stand-alone package For the implementation of HO reasoners on top of it Easily extensible by agent-based design Already available! On the long run: Reasoning-as-a-service for domain specific tools Many non-classical logics through internal embedding into HOL Machine learning components could also be re-used by applications Natural language processing Algebra systems,... Agent-Based HOL Reasoning, ICMS 2016 14

Conclusion and Further work Conclusion Presented Leo-III as an agent-based blackboard proving system Based on (polymorphic) higher-order logic Flexible (ad-hoc) inclusion of specialist systems LeoPARD accessible for own developments Further work Split internal saturation into more fine-grained agent tasks Do experiments to find good parameters Lots of technical improvements Implementation of/connection to more external specialists Better proof output Agent-Based HOL Reasoning, ICMS 2016 15