%PDF-1.2 7 0 obj [5 0 R/XYZ 72 720] endobj 12 0 obj << /Title(TITLE PAGE) /A<< /S/GoTo /D(prelim.0) >> /Parent 11 0 R /Next 13 0 R >> endobj 13 0 obj << /Title(COMMITTEE MEMBERSHIP PAGE) /A<< /S/GoTo /D(prelim.1) >> /Parent 11 0 R /Prev 12 0 R /Next 14 0 R >> endobj 14 0 obj << /Title(ABSTRACT) /A<< /S/GoTo /D(prelim.2) >> /Parent 11 0 R /Prev 13 0 R /Next 15 0 R >> endobj 15 0 obj << /Title(TABLE OF CONTENTS) /A<< /S/GoTo /D(prelim.4) >> /Parent 11 0 R /Prev 14 0 R /Next 16 0 R >> endobj 17 0 obj << /Title(1. Pseudo-code implementation of value iteration.) /A<< /S/GoTo /D(figure.2.1) >> /Parent 16 0 R /Next 18 0 R >> endobj 18 0 obj << /Title(2. Pseudo-code implementation of policy iteration.) /A<< /S/GoTo /D(figure.2.2) >> /Parent 16 0 R /Prev 17 0 R /Next 19 0 R >> endobj 19 0 obj << /Title(3. Four computers in a ring topology, a graphical representation of factored transition and reward models in the 4-ring network, and a graphical representation of the linear value function approximation in the 4-ring network.) /A<< /S/GoTo /D(figure.2.3) >> /Parent 16 0 R /Prev 18 0 R /Next 20 0 R >> endobj 20 0 obj << /Title(4. Pseudo-code implementation of the least-squares value iteration.) /A<< /S/GoTo /D(figure.2.4) >> /Parent 16 0 R /Prev 19 0 R /Next 21 0 R >> endobj 21 0 obj << /Title(5. A graphical representation of combining factored transition and reward models with the linear value function approximation.) /A<< /S/GoTo /D(figure.2.5) >> /Parent 16 0 R /Prev 20 0 R /Next 22 0 R >> endobj 22 0 obj << /Title(6. Four transition functions from the 4-ring network administration problem.) /A<< /S/GoTo /D(figure.3.6) >> /Parent 16 0 R /Prev 21 0 R /Next 23 0 R >> endobj 23 0 obj << /Title(7. Expectations of three basis functions with respect to a transition function from the 4-ring network administration problem.) /A<< /S/GoTo /D(figure.3.7) >> /Parent 16 0 R /Prev 22 0 R /Next 24 0 R >> endobj 24 0 obj << /Title(8. Graphical relation between the optimal value function, and its complete and relaxed HALP approximations.) /A<< /S/GoTo /D(figure.3.8) >> /Parent 16 0 R /Prev 23 0 R /Next 25 0 R >> endobj 25 0 obj << /Title(9. Pseudo-code implementation of the MC-HALP solver.) /A<< /S/GoTo /D(figure.3.9) >> /Parent 16 0 R /Prev 24 0 R /Next 26 0 R >> endobj 26 0 obj << /Title(10. Pseudo-code implementation of the e-HALP solver.) /A<< /S/GoTo /D(figure.3.10) >> /Parent 16 0 R /Prev 25 0 R /Next 27 0 R >> endobj 27 0 obj << /Title(11. Pseudo-code implementation of a HALP solver with the cutting plane method.) /A<< /S/GoTo /D(figure.3.11) >> /Parent 16 0 R /Prev 26 0 R /Next 28 0 R >> endobj 28 0 obj << /Title(12. Pseudo-code implementation of the e-HALP separation oracle.) /A<< /S/GoTo /D(figure.3.12) >> /Parent 16 0 R /Prev 27 0 R /Next 29 0 R >> endobj 29 0 obj << /Title(13. Pseudo-code implementation of the MCMC-HALP separation oracle.) /A<< /S/GoTo /D(figure.3.13) >> /Parent 16 0 R /Prev 28 0 R /Next 30 0 R >> endobj 30 0 obj << /Title(14. Comparison of three methods for solving hybrid factored MDPs on the 4-ring network administration problem.) /A<< /S/GoTo /D(figure.3.14) >> /Parent 16 0 R /Prev 29 0 R /Next 31 0 R >> endobj 31 0 obj << /Title(15. Illustrations of three irrigation network topologies: 6-ring, 6-ring-of-rings, and 3 x 3 grid.) /A<< /S/GoTo /D(figure.3.15) >> /Parent 16 0 R /Prev 30 0 R /Next 32 0 R >> endobj 32 0 obj << /Title(16. Indexing in the description of the irrigation network transition model, and univariate reward and basis functions.) /A<< /S/GoTo /D(figure.3.16) >> /Parent 16 0 R /Prev 31 0 R /Next 33 0 R >> endobj 33 0 obj << /Title(17. Comparison of three HALP solvers on two irrigation network topologies.) /A<< /S/GoTo /D(figure.3.17) >> /Parent 16 0 R /Prev 32 0 R /Next 34 0 R >> endobj 34 0 obj << /Title(18. Scale-up potential of three HALP solvers on two irrigation network topologies.) /A<< /S/GoTo /D(figure.3.18) >> /Parent 16 0 R /Prev 33 0 R /Next 35 0 R >> endobj 35 0 obj << /Title(19. Univariate projections of three approximate value functions on the 6-ring irrigation network problem.) /A<< /S/GoTo /D(figure.3.19) >> /Parent 16 0 R /Prev 34 0 R /Next 36 0 R >> endobj 36 0 obj << /Title(20. Comparison of three HALP solvers on the 3 x 3 grid irrigation network problem.) /A<< /S/GoTo /D(figure.3.20) >> /Parent 16 0 R /Prev 35 0 R /Next 37 0 R >> endobj 37 0 obj << /Title(21. Pseudo-code implementation of the HAPI solver.) /A<< /S/GoTo /D(figure.3.21) >> /Parent 16 0 R /Prev 36 0 R /Next 38 0 R >> endobj 38 0 obj << /Title(22. Comparison of the e-HALP and e-HAPI methods on the 6-ring and 6-ring-of-rings irrigation network problems.) /A<< /S/GoTo /D(figure.3.22) >> /Parent 16 0 R /Prev 37 0 R /Next 39 0 R >> endobj 39 0 obj << /Title(23. Selecting transition functions based on the domains of state variables.) /A<< /S/GoTo /D(figure.4.23) >> /Parent 16 0 R /Prev 38 0 R /Next 40 0 R >> endobj 40 0 obj << /Title(24. Expectations of three basis functions.) /A<< /S/GoTo /D(figure.4.24) >> /Parent 16 0 R /Prev 39 0 R /Next 41 0 R >> endobj 41 0 obj << /Title(25. Comparison of two approaches to solving the rover problem.) /A<< /S/GoTo /D(figure.4.25) >> /Parent 16 0 R /Prev 40 0 R /Next 42 0 R >> endobj 42 0 obj << /Title(26. Value function approximations for the rover problem.) /A<< /S/GoTo /D(figure.4.26) >> /Parent 16 0 R /Prev 41 0 R /Next 43 0 R >> endobj 43 0 obj << /Title(27. Value function approximations for the rover problem.) /A<< /S/GoTo /D(figure.4.27) >> /Parent 16 0 R /Prev 42 0 R /Next 44 0 R >> endobj 44 0 obj << /Title(28. Comparison of greedy methods for learning basis functions on the 6-ring irrigation network and rover problems.) /A<< /S/GoTo /D(figure.5.28) >> /Parent 16 0 R /Prev 43 0 R /Next 45 0 R >> endobj 45 0 obj << /Title(29. Univariate projections of four approximate value functions on the 6-ring irrigation network problem.) /A<< /S/GoTo /D(figure.5.29) >> /Parent 16 0 R /Prev 44 0 R >> endobj 16 0 obj << /Title(LIST OF FIGURES) /A<< /S/GoTo /D(prelim.6) >> /Parent 11 0 R /Prev 15 0 R /First 17 0 R /Last 45 0 R /Count -29 /Next 46 0 R >> endobj 46 0 obj << /Title(PREFACE) /A<< /S/GoTo /D(section*.1) >> /Parent 11 0 R /Prev 16 0 R /Next 47 0 R >> endobj 47 0 obj << /Title(1.0 INTRODUCTION) /A<< /S/GoTo /D(chapter.1) >> /Parent 11 0 R /Prev 46 0 R /Next 48 0 R >> endobj 49 0 obj << /Title(2.1 Markov Decision Processes) /A<< /S/GoTo /D(section.2.1) >> /Parent 48 0 R /Next 50 0 R >> endobj 51 0 obj << /Title(2.2.1 Value Iteration) /A<< /S/GoTo /D(subsection.2.2.1) >> /Parent 50 0 R /Next 52 0 R >> endobj 52 0 obj << /Title(2.2.2 Policy Iteration) /A<< /S/GoTo /D(subsection.2.2.2) >> /Parent 50 0 R /Prev 51 0 R /Next 53 0 R >> endobj 53 0 obj << /Title(2.2.3 Linear Programming) /A<< /S/GoTo /D(subsection.2.2.3) >> /Parent 50 0 R /Prev 52 0 R >> endobj 50 0 obj << /Title(2.2 Solving Markov Decision Processes) /A<< /S/GoTo /D(section.2.2) >> /Parent 48 0 R /Prev 49 0 R /First 51 0 R /Last 53 0 R /Count -3 /Next 54 0 R >> endobj 55 0 obj << /Title(2.3.1 Factored Transition and Reward Models) /A<< /S/GoTo /D(subsection.2.3.1) >> /Parent 54 0 R /Next 56 0 R >> endobj 57 0 obj << /Title(2.3.2.1 Approximate Value Iteration) /A<< /S/GoTo /D(subsubsection.2.3.2.1) >> /Parent 56 0 R /Next 58 0 R >> endobj 58 0 obj << /Title(2.3.2.2 Approximate Policy Iteration) /A<< /S/GoTo /D(subsubsection.2.3.2.2) >> /Parent 56 0 R /Prev 57 0 R /Next 59 0 R >> endobj 59 0 obj << /Title(2.3.2.3 Approximate Linear Programming) /A<< /S/GoTo /D(subsubsection.2.3.2.3) >> /Parent 56 0 R /Prev 58 0 R /Next 60 0 R >> endobj 60 0 obj << /Title(2.3.2.4 Learning Basis Functions) /A<< /S/GoTo /D(subsubsection.2.3.2.4) >> /Parent 56 0 R /Prev 59 0 R >> endobj 56 0 obj << /Title(2.3.2 Parametric Value Function Approximations) /A<< /S/GoTo /D(subsection.2.3.2) >> /Parent 54 0 R /Prev 55 0 R /First 57 0 R /Last 60 0 R /Count -4 /Next 61 0 R >> endobj 61 0 obj << /Title(2.3.3 Parametric Policy Approximations) /A<< /S/GoTo /D(subsection.2.3.3) >> /Parent 54 0 R /Prev 56 0 R /Next 62 0 R >> endobj 54 0 obj << /Title(2.3 Solving Large-Scale Markov Decision Processes) /A<< /S/GoTo /D(section.2.3) >> /Parent 48 0 R /Prev 50 0 R /First 55 0 R /Last 62 0 R /Count -4 >> endobj 62 0 obj << /Title(2.3.4 Grid-Based Approximations) /A<< /S/GoTo /D(subsection.2.3.4) >> /Parent 54 0 R /Prev 61 0 R >> endobj 48 0 obj << /Title(2.0 BACKGROUND AND RELATED WORK) /A<< /S/GoTo /D(chapter.2) >> /Parent 11 0 R /Prev 47 0 R /First 49 0 R /Last 54 0 R /Count -3 /Next 63 0 R >> endobj 64 0 obj << /Title(3.1 Factored Transition and Reward Models) /A<< /S/GoTo /D(section.3.1) >> /Parent 63 0 R /Next 65 0 R >> endobj 66 0 obj << /Title(3.2.1 Error Bounds) /A<< /S/GoTo /D(subsection.3.2.1) >> /Parent 65 0 R /Next 67 0 R >> endobj 68 0 obj << /Title(3.2.2.1 Factored State Relevance Density Functions) /A<< /S/GoTo /D(subsubsection.3.2.2.1) >> /Parent 67 0 R /Next 69 0 R >> endobj 67 0 obj << /Title(3.2.2 Expectation Terms) /A<< /S/GoTo /D(subsection.3.2.2) >> /Parent 65 0 R /Prev 66 0 R /First 68 0 R /Last 69 0 R /Count -2 >> endobj 69 0 obj << /Title(3.2.2.2 Factored Basis Functions) /A<< /S/GoTo /D(subsubsection.3.2.2.2) >> /Parent 67 0 R /Prev 68 0 R >> endobj 65 0 obj << /Title(3.2 Hybrid Approximate Linear Programming) /A<< /S/GoTo /D(section.3.2) >> /Parent 63 0 R /Prev 64 0 R /First 66 0 R /Last 67 0 R /Count -2 /Next 70 0 R >> endobj 71 0 obj << /Title(3.3.1 MC-HALP) /A<< /S/GoTo /D(subsection.3.3.1) >> /Parent 70 0 R /Next 72 0 R >> endobj 73 0 obj << /Title(3.3.2.1 Error Bounds) /A<< /S/GoTo /D(subsubsection.3.3.2.1) >> /Parent 72 0 R /Next 74 0 R >> endobj 74 0 obj << /Title(3.3.2.2 Grid Resolution) /A<< /S/GoTo /D(subsubsection.3.3.2.2) >> /Parent 72 0 R /Prev 73 0 R >> endobj 72 0 obj << /Title(3.3.2 e-HALP) /A<< /S/GoTo /D(subsection.3.3.2) >> /Parent 70 0 R /Prev 71 0 R /First 73 0 R /Last 74 0 R /Count -2 /Next 75 0 R >> endobj 75 0 obj << /Title(3.3.3 Cutting Plane Method) /A<< /S/GoTo /D(subsection.3.3.3) >> /Parent 70 0 R /Prev 72 0 R /Next 76 0 R >> endobj 77 0 obj << /Title(3.3.4.1 Compact Representation of Constraints) /A<< /S/GoTo /D(subsubsection.3.3.4.1) >> /Parent 76 0 R /Next 78 0 R >> endobj 78 0 obj << /Title(3.3.4.2 Separation Oracle) /A<< /S/GoTo /D(subsubsection.3.3.4.2) >> /Parent 76 0 R /Prev 77 0 R /Next 79 0 R >> endobj 76 0 obj << /Title(3.3.4 MCMC-HALP) /A<< /S/GoTo /D(subsection.3.3.4) >> /Parent 70 0 R /Prev 75 0 R /First 77 0 R /Last 79 0 R /Count -3 >> endobj 79 0 obj << /Title(3.3.4.3 Constraint Satisfaction) /A<< /S/GoTo /D(subsubsection.3.3.4.3) >> /Parent 76 0 R /Prev 78 0 R >> endobj 70 0 obj << /Title(3.3 Constraint Space Approximations) /A<< /S/GoTo /D(section.3.3) >> /Parent 63 0 R /Prev 65 0 R /First 71 0 R /Last 76 0 R /Count -4 /Next 80 0 R >> endobj 81 0 obj << /Title(3.4.1 A Simple Example) /A<< /S/GoTo /D(subsection.3.4.1) >> /Parent 80 0 R /Next 82 0 R >> endobj 82 0 obj << /Title(3.4.2 Scale-up Potential) /A<< /S/GoTo /D(subsection.3.4.2) >> /Parent 80 0 R /Prev 81 0 R /Next 83 0 R >> endobj 83 0 obj << /Title(3.4.3 The Curse of Treewidth) /A<< /S/GoTo /D(subsection.3.4.3) >> /Parent 80 0 R /Prev 82 0 R /Next 84 0 R >> endobj 84 0 obj << /Title(3.4.4 Hybrid Approximate Policy Iteration) /A<< /S/GoTo /D(subsection.3.4.4) >> /Parent 80 0 R /Prev 83 0 R >> endobj 80 0 obj << /Title(3.4 Experiments) /A<< /S/GoTo /D(section.3.4) >> /Parent 63 0 R /Prev 70 0 R /First 81 0 R /Last 84 0 R /Count -4 /Next 85 0 R >> endobj 85 0 obj << /Title(3.5 Conclusions) /A<< /S/GoTo /D(section.3.5) >> /Parent 63 0 R /Prev 80 0 R >> endobj 63 0 obj << /Title(3.0 SOLVING HYBRID FACTORED MDPS BY LINEAR PROGRAMMING) /A<< /S/GoTo /D(chapter.3) >> /Parent 11 0 R /Prev 48 0 R /First 64 0 R /Last 85 0 R /Count -5 /Next 86 0 R >> endobj 87 0 obj << /Title(4.1 Factored Transition and Reward Models) /A<< /S/GoTo /D(section.4.1) >> /Parent 86 0 R /Next 88 0 R >> endobj 88 0 obj << /Title(4.2 Generalized Hybrid ALP) /A<< /S/GoTo /D(section.4.2) >> /Parent 86 0 R /Prev 87 0 R /Next 89 0 R >> endobj 89 0 obj << /Title(4.3 Error Bounds) /A<< /S/GoTo /D(section.4.3) >> /Parent 86 0 R /Prev 88 0 R /Next 90 0 R >> endobj 90 0 obj << /Title(4.4 Expectation Terms) /A<< /S/GoTo /D(section.4.4) >> /Parent 86 0 R /Prev 89 0 R /Next 91 0 R >> endobj 91 0 obj << /Title(4.5 Constraint Space Approximations) /A<< /S/GoTo /D(section.4.5) >> /Parent 86 0 R /Prev 90 0 R /Next 92 0 R >> endobj 93 0 obj << /Title(4.6.1 Experimental Setup) /A<< /S/GoTo /D(subsection.4.6.1) >> /Parent 92 0 R /Next 94 0 R >> endobj 94 0 obj << /Title(4.6.2 Experimental Results) /A<< /S/GoTo /D(subsection.4.6.2) >> /Parent 92 0 R /Prev 93 0 R >> endobj 92 0 obj << /Title(4.6 Experiments) /A<< /S/GoTo /D(section.4.6) >> /Parent 86 0 R /Prev 91 0 R /First 93 0 R /Last 94 0 R /Count -2 /Next 95 0 R >> endobj 95 0 obj << /Title(4.7 Conclusions) /A<< /S/GoTo /D(section.4.7) >> /Parent 86 0 R /Prev 92 0 R >> endobj 86 0 obj << /Title(4.0 SOLVING HYBRID FACTORED MDPS WITH EXPONENTIAL-FAMILY TRANSITION MODELS) /A<< /S/GoTo /D(chapter.4) >> /Parent 11 0 R /Prev 63 0 R /First 87 0 R /Last 95 0 R /Count -7 /Next 96 0 R >> endobj 97 0 obj << /Title(5.1 Optimization of Relaxed HALP) /A<< /S/GoTo /D(section.5.1) >> /Parent 96 0 R /Next 98 0 R >> endobj 98 0 obj << /Title(5.2 Scoring Basis Functions) /A<< /S/GoTo /D(section.5.2) >> /Parent 96 0 R /Prev 97 0 R /Next 99 0 R >> endobj 99 0 obj << /Title(5.3 Optimization of Basis Functions) /A<< /S/GoTo /D(section.5.3) >> /Parent 96 0 R /Prev 98 0 R /Next 100 0 R >> endobj 100 0 obj << /Title(5.4 Overfitting on Active Constraints) /A<< /S/GoTo /D(section.5.4) >> /Parent 96 0 R /Prev 99 0 R /Next 101 0 R >> endobj 102 0 obj << /Title(5.5.1 Experimental Setup) /A<< /S/GoTo /D(subsection.5.5.1) >> /Parent 101 0 R /Next 103 0 R >> endobj 103 0 obj << /Title(5.5.2 Experimental Results) /A<< /S/GoTo /D(subsection.5.5.2) >> /Parent 101 0 R /Prev 102 0 R >> endobj 101 0 obj << /Title(5.5 Experiments) /A<< /S/GoTo /D(section.5.5) >> /Parent 96 0 R /Prev 100 0 R /First 102 0 R /Last 103 0 R /Count -2 /Next 104 0 R >> endobj 104 0 obj << /Title(5.6 Conclusions) /A<< /S/GoTo /D(section.5.6) >> /Parent 96 0 R /Prev 101 0 R >> endobj 96 0 obj << /Title(5.0 LEARNING BASIS FUNCTIONS IN HYBRID DOMAINS) /A<< /S/GoTo /D(chapter.5) >> /Parent 11 0 R /Prev 86 0 R /First 97 0 R /Last 104 0 R /Count -6 /Next 105 0 R >> endobj 106 0 obj << /Title(6.1 Semi-Markov Decision Processes) /A<< /S/GoTo /D(section.6.1) >> /Parent 105 0 R /Next 107 0 R >> endobj 107 0 obj << /Title(6.2 Partially-Observable Markov Decision Processes) /A<< /S/GoTo /D(section.6.2) >> /Parent 105 0 R /Prev 106 0 R >> endobj 105 0 obj << /Title(6.0 CONCLUSIONS AND FUTURE WORK) /A<< /S/GoTo /D(chapter.6) >> /Parent 11 0 R /Prev 96 0 R /First 106 0 R /Last 107 0 R /Count -2 /Next 108 0 R >> endobj 108 0 obj << /Title(APPENDIX. PROOFS) /A<< /S/GoTo /D(appendix.A) >> /Parent 11 0 R /Prev 105 0 R /Next 109 0 R >> endobj 110 0 obj [5 0 R/XYZ 72 720] endobj 111 0 obj [5 0 R/XYZ 72 720] endobj 112 0 obj << /Type/Encoding /Differences[1/dotaccent/fi/fl/fraction/hungarumlaut/Lslash/lslash/ogonek/ring 11/breve/minus 14/Zcaron/zcaron/caron/dotlessi/dotlessj/ff/ffi/ffl 30/grave/quotesingle/space/exclam/quotedbl/numbersign/dollar/percent/ampersand/quoteright/parenleft/parenright/asterisk/plus/comma/hyphen/period/slash/zero/one/two/three/four/five/six/seven/eight/nine/colon/semicolon/less/equal/greater/question/at/A/B/C/D/E/F/G/H/I/J/K/L/M/N/O/P/Q/R/S/T/U/V/W/X/Y/Z/bracketleft/backslash/bracketright/asciicircum/underscore/quoteleft/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y/z/braceleft/bar/braceright/asciitilde 130/quotesinglbase/florin/quotedblbase/ellipsis/dagger/daggerdbl/circumflex/perthousand/Scaron/guilsinglleft/OE 147/quotedblleft/quotedblright/bullet/endash/emdash/tilde/trademark/scaron/guilsinglright/oe 159/Ydieresis 161/exclamdown/cent/sterling/currency/yen/brokenbar/section/dieresis/copyright/ordfeminine/guillemotleft/logicalnot/hyphen/registered/macron/degree/plusminus/twosuperior/threesuperior/acute/mu/paragraph/periodcentered/cedilla/onesuperior/ordmasculine/guillemotright/onequarter/onehalf/threequarters/questiondown/Agrave/Aacute/Acircumflex/Atilde/Adieresis/Aring/AE/Ccedilla/Egrave/Eacute/Ecircumflex/Edieresis/Igrave/Iacute/Icircumflex/Idieresis/Eth/Ntilde/Ograve/Oacute/Ocircumflex/Otilde/Odieresis/multiply/Oslash/Ugrave/Uacute/Ucircumflex/Udieresis/Yacute/Thorn/germandbls/agrave/aacute/acircumflex/atilde/adieresis/aring/ae/ccedilla/egrave/eacute/ecircumflex/edieresis/igrave/iacute/icircumflex/idieresis/eth/ntilde/ograve/oacute/ocircumflex/otilde/odieresis/divide/oslash/ugrave/uacute/ucircumflex/udieresis/yacute/thorn/ydieresis] >> endobj 113 0 obj << /Encoding 112 0 R /Type/Font /Subtype/Type1 /Name/F1 /BaseFont/Times-Bold >> endobj 114 0 obj << /Encoding 112 0 R /Type/Font /Subtype/Type1 /Name/F2 /BaseFont/Times-Roman >> endobj 116 0 obj << /Filter[/FlateDecode] /Length 451 >> stream x}Sn0WxiK&aG23U 38gRW=8Aܱ_U~M bJ@q4$֫7seiAe!L3>.nECd HwRPLL(z~Yޯ? #g|@hh]֢%{?QQ' nP77T^ծF0u# >B{v/šGaHm"GiTؖSyhU,oۆ6\JQQ/*]^|!j^KmD)}qj0R^:ŘdBn9)_ϑwN*(]dJ^YHըC}a:ڶ(=" l5"3Xв3vW?ַQh endstream endobj 118 0 obj << /F1 113 0 R /F2 114 0 R >> endobj 6 0 obj << /ProcSet[/PDF/Text/ImageC] /Font 118 0 R >> endobj 121 0 obj [119 0 R/XYZ 72 720] endobj 122 0 obj [119 0 R/XYZ 72 720] endobj 123 0 obj << /Encoding 112 0 R /Type/Font /Subtype/Type1 /Name/F3 /BaseFont/Times-Italic >> endobj 124 0 obj << /Filter[/FlateDecode] /Length 623 >> stream xUˎ0+x0O]$RiTuAXC gF3__%) iJ#u8s9edbD;S X("QlQ~hE}Q7,'оMC@rz 8aQL.lNNjcQaty-[62Z3W`UńL$/rEEN*Z V\MFQV.D"y%: gpY'Ƥ x J߰-7*zETDf0GLkmL}EMJ1s& ]:Cphw'_c^Su#pp@GL
yRH6o,Q.)I-;JP=A4LĹGVI[uء= g,9wp딨qd_."Ey#Y[NfW
^
ȩ48Yz#FX2ކqRYO0}aHڌR4{Z caכfM *:
endstream
endobj
125 0 obj
<<
/F2 114 0 R
/F1 113 0 R
/F3 123 0 R
>>
endobj
120 0 obj
<<
/ProcSet[/PDF/Text/ImageC]
/Font 125 0 R
>>
endobj
128 0 obj
[126 0 R/XYZ 72 720]
endobj
129 0 obj
[126 0 R/XYZ 72 720]
endobj
130 0 obj
<<
/Filter[/FlateDecode]
/Length 900
>>
stream
x}UKo8WHkU1 E쁖iD
$l#;zgf8I}'1WETf,)⢊rb^?YWdHˇyXS6'wW7
e˻ۋUSYg,]lb^8M_^(ˉpZv=Ae(кOJ4CVatohwtn(h([};L