1 / 12

Applications of Inductive Logic Programming(ILP)

Applications of Inductive Logic Programming(ILP). 류정우. Table of Contents. Introduction ILP Compared to Other Approaches to ML Predicting the Mutagenicity of Chemical Compounds Conclusion. Introduction. ILP(Inductive Logic Programming) Inductive machine learning + Logic programming

andie
Download Presentation

Applications of Inductive Logic Programming(ILP)

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Applications of Inductive Logic Programming(ILP) 류정우

  2. Table of Contents • Introduction • ILP Compared to Other Approaches to ML • Predicting the Mutagenicity of Chemical Compounds • Conclusion 1999-06-19

  3. Introduction • ILP(Inductive Logic Programming) • Inductive machine learning + Logic programming • Components • B : Background knowledge • E+ : Positive Example • E- : Negative Example • H : Predicate logic formula, The Result • All Example in E+ can be derived from B  H • No Negative Example in E- can be derived from B  H 1999-06-19

  4. QuickSort Example(1) • Examples(E+,E-) example(qsort( [], [a], [a] ), true). example(qsort( [a], [a], [a] ), false). example(qsort( [d,f,b,e,c,g,a], [a,b,c,d,e,f,g], [] ), true). example(qsort( [f,e,g], [e,f,g], [] ), true). example(qsort( [b,c,a], [a,b,c,d,e,f,g], [d,e,f,g] ), true). • Background Knowledge(B) partition( X, [], [], []). partition( X, [Y | Rest], [Y | Smalls], Bigs) :- gt(X,Y), !, partition(X, Rest, Smalls, Bigs). partition(X, [Y | Rest], Smalls, [Y | Bigs]) :- partition( X, Rest, Smalls, Bigs). 1999-06-19

  5. QuickSort Example(2) • Induced Definition(H) • Using a ILP System, Markus qsort( [], L, L). qsort( [X | L], SL1A, SL2B) :- partition(L, X, L1, L2), qsort( L2, SL2A, SL2B), qsort( L1, SL1A, [X | SL2A] ). 1999-06-19

  6. ILP Compared to Other Approaches to ML(1) • Attribute-based Learning • Decision tree,Neural network,Nearest neighbor • Advantages • Simple, Efficient, Effective for handling noise • Limits • Only for non-relational descriptions of objects • Limited form of Background knowledge • Induction almost has to start from scratch 1999-06-19

  7. ILP Compared to Other Approaches to ML(2) • ILP • Advantage • General representation of Background Knowledge • EX) Learning about properties of chemical compounds • use molecular structure for background knowledge • Disadvantage • computational complexity 1999-06-19

  8. Predicting the Mutagenicity of Chemical Compounds(1) • SAR(Structure/Activity) Relationship of chemical compounds • Traditional Way - Attribute-based • using global attributes of a molecule • EX) LUMO(Energy of the Lowest Molecular Orbital) • can’t use the patterns in the molecule structure 188 compounds successful Classical Regression 230 heteroaromatic nitro compounds These need ILP! 42 compounds failed 1999-06-19

  9. Predicting the Mutagenicity of Chemical Compounds(2) • Using Progol - A ILP System Relationship atoms bonds charge Class Compound Compound High/Low mutagenicity • Background Knowledge • Classified Results of 230 Compounds • 18300 Prolog facts • LUMO(Energy of Lowest Unoccupied Molecular Orbital) • only for the 188 compounds amenable to regression 1999-06-19

  10. Predicting the Mutagenicity of Chemical Compounds(3) • Result from Progol • For 188 Compounds • EX) Mutagenic if it has a RUMO value < -1.937 • 89% Accuracy(matching accuracy of regression) • but easy to comrehend and automatically generated • The Remaining 42 Compounds • 1 Rule with accuracy 88% • regression : 62% 1999-06-19

  11. Predicting the Mutagenicity of Chemical Compounds(4) • Good Prediction Accuracy • New Chemical Insight Discovery of New Structural Feature! Five-membered aromatic ring with a nitrogen atom U Y = Z Mutagenicity V Double bond W X 1999-06-19

  12. Conclusion • ILP의 적용 분야 Difficult, industrially or scientifically relevant NOT satisfactorily solved Relational Background > Attribute-value form Major Advantage Major Obstacle Background Knowledge in the form of General Program like Prolog Inefficiency - handling numerical data 1999-06-19

More Related