%PDF-1.3 1 0 obj << /Kids [ 3 0 R 4 0 R 5 0 R 6 0 R 7 0 R 8 0 R 9 0 R 10 0 R 11 0 R ] /Type /Pages /Count 9 >> endobj 2 0 obj << /Title (MDL\055CW\072 A Multimodal Deep Learning Framework With Cross Weights) /Producer (PyPDF2) /Author (Sarah Rastegar\054 Mahdieh Soleymani\054 Hamid R\056 Rabiee\054 Seyed Mohsen Shojaee) /Subject (2016 IEEE Conference on Computer Vision and Pattern Recognition) >> endobj 3 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 13 0 R /Resources << /XObject << /x8 14 0 R /x6 17 0 R /x12 20 0 R /x10 23 0 R >> /ExtGState << /s9 26 0 R /s11 29 0 R /R16 32 0 R /a0 << /CA 1 /ca 1 >> /s5 33 0 R /s7 36 0 R >> /Font << /F2 39 0 R /F1 40 0 R /R17 41 0 R /R21 45 0 R /R19 48 0 R /R23 52 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 56 0 R 57 0 R 58 0 R 59 0 R 60 0 R 61 0 R 62 0 R 63 0 R ] >> endobj 4 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 64 0 R /Resources << /ExtGState << /R16 32 0 R >> /Font << /F2 65 0 R /F1 66 0 R /R38 67 0 R /R17 41 0 R /R19 48 0 R /R23 52 0 R /R44 71 0 R /R42 75 0 R /R40 79 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 82 0 R 83 0 R 84 0 R 85 0 R 86 0 R 87 0 R 88 0 R 89 0 R ] >> endobj 5 0 obj << /Parent 1 0 R /Rotate 0 /Contents 90 0 R /Resources << /XObject << /R65 91 0 R /R66 92 0 R /R63 93 0 R /R62 94 0 R >> /ExtGState << /R16 32 0 R >> /Font << /R58 95 0 R /R44 71 0 R /R42 75 0 R /R40 79 0 R /F2 98 0 R /F1 99 0 R /R60 100 0 R /R38 67 0 R /R17 41 0 R /R19 48 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 103 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 104 0 R 105 0 R 106 0 R 107 0 R 108 0 R 109 0 R ] /Type /Page >> endobj 6 0 obj << /Parent 1 0 R /Rotate 0 /Contents 110 0 R /Resources << /XObject << /R78 111 0 R /R79 112 0 R >> /ExtGState << /R16 32 0 R >> /Font << /R82 113 0 R /R40 79 0 R /F2 117 0 R /F1 118 0 R /R60 100 0 R /R38 67 0 R /R17 41 0 R /R19 48 0 R /R23 52 0 R /R80 119 0 R /R84 123 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 103 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 126 0 R 127 0 R 128 0 R 129 0 R 130 0 R ] /Type /Page >> endobj 7 0 obj << /Parent 1 0 R /Rotate 0 /Contents 131 0 R /Resources << /XObject << /R101 132 0 R /R100 133 0 R >> /ExtGState << /R16 32 0 R >> /Font << /F2 134 0 R /R19 48 0 R /F1 135 0 R /R17 41 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /Group 103 0 R /MediaBox [ 0 0 612 792 ] /Annots [ 136 0 R 137 0 R 138 0 R 139 0 R 140 0 R 141 0 R 142 0 R ] /Type /Page >> endobj 8 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 143 0 R /Resources << /XObject << /R113 144 0 R >> /ExtGState << /R16 32 0 R >> /Font << /R40 79 0 R /F2 145 0 R /F1 146 0 R /R17 41 0 R /R19 48 0 R /R23 52 0 R /R80 119 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 147 0 R 148 0 R 149 0 R 150 0 R 151 0 R 152 0 R ] >> endobj 9 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 153 0 R /Resources << /XObject << /R134 154 0 R >> /ExtGState << /R16 32 0 R >> /Font << /R17 41 0 R /R40 79 0 R /F2 155 0 R /F1 156 0 R /R82 113 0 R /R19 48 0 R >> /ProcSet [ /Text /ImageC /ImageB /PDF /ImageI ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 157 0 R 158 0 R 159 0 R 160 0 R 161 0 R 162 0 R 163 0 R 164 0 R 165 0 R 166 0 R 167 0 R 168 0 R 169 0 R 170 0 R ] >> endobj 10 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 171 0 R /Resources << /ExtGState << /R16 32 0 R >> /Font << /R40 79 0 R /F2 172 0 R /F1 173 0 R /R17 41 0 R /R19 48 0 R /R80 119 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 174 0 R 175 0 R 176 0 R 177 0 R 178 0 R 179 0 R 180 0 R 181 0 R 182 0 R 183 0 R 184 0 R 185 0 R 186 0 R 187 0 R 188 0 R 189 0 R 190 0 R 191 0 R 192 0 R 193 0 R 194 0 R ] >> endobj 11 0 obj << /Parent 1 0 R /Rotate 0 /Type /Page /Contents 195 0 R /Resources << /ExtGState << /R16 32 0 R >> /Font << /F2 196 0 R /R19 48 0 R /R23 52 0 R /F1 197 0 R /R17 41 0 R >> /ProcSet [ /ImageC /Text /PDF /ImageI /ImageB ] >> /MediaBox [ 0 0 612 792 ] /Annots [ 198 0 R 199 0 R 200 0 R 201 0 R 202 0 R 203 0 R 204 0 R 205 0 R 206 0 R 207 0 R 208 0 R 209 0 R 210 0 R 211 0 R 212 0 R 213 0 R 214 0 R 215 0 R 216 0 R 217 0 R 218 0 R 219 0 R 220 0 R 221 0 R 222 0 R 223 0 R 224 0 R 225 0 R 226 0 R 227 0 R 228 0 R 229 0 R 230 0 R 231 0 R 232 0 R 233 0 R 234 0 R 235 0 R 236 0 R ] >> endobj 12 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 13 0 obj << /Length 18799 >> stream q q q 0.10000 0 0 0.10000 0 0 cm /R16 gs 0 g q 10 0 0 10 0 0 cm BT /R17 14.34620 Tf 1 0 0 1 72.85200 675.06700 Tm [ (MDL\055CW) 54.99890 (\072) -249.98900 (A) -250.00200 (Multimodal) -250.01000 (Deep) -250.00800 (Lear) 14.98930 (ning) -249.99300 (Framew) 9.99795 (ork) -250 (with) -250.01500 (Cr) 17.99460 (oss) -250.00800 (W) 65.00110 (eights) ] TJ /R19 11.95520 Tf 9.80195 -37.85820 Td [ (Sarah) -250.01300 (Raste) 14.99740 (g) 4.98790 (ar) 40.00330 (\054) -250.01200 (Mahdieh) -249.99300 (Sole) 15.00560 (ymani) -250.01400 (Baghshah\054) -250 (Hamid) -250 (R\056) -249.98700 (Rabiee\054) -250.01400 (Se) 15.00560 (yed) -249.98500 (Mohsen) -250.01000 (Shojaee) ] TJ -17.32890 -13.94690 Td [ (AICT) -250.01600 (Inno) 15.01480 (v) 24.98440 (ation) -250.00200 (Center) 39.98490 (\054) -250.01200 (Department) -250 (of) -250.01400 (Computer) -250.01400 (Engineering\054) -249.98500 (Sharif) -250.01000 (Uni) 24.99570 (v) 14.98510 (ersity) -249.98900 (of) -250.01400 (T) 70.01640 (echnology) ] TJ 203.16400 -13.94800 Td [ (T) 70.01640 (ehran\054) -249.99700 (Iran) ] TJ /R21 8.96640 Tf -16.60510 -13.94800 Td (rabiee\100sharif\056edu) Tj /R17 11.95520 Tf -105.88900 -41.04610 Td (Abstract) Tj /R23 9.96260 Tf -83.92770 -23.92890 Td [ (Deep) -198.98700 (learning) -198.99000 (has) -199.01000 (r) 37.01960 (eceived) -198.99600 (muc) 14.98160 (h) -198.98200 (attention) -198.98200 (as) -199.01300 (of) -199.01300 (the) -199.01800 (most) ] TJ -11.95510 -11.95510 Td [ (powerful) -325 (appr) 44.99370 (oac) 14.98400 (hes) -324.99700 (for) -324.98600 (multimodal) -326.00900 (r) 37.01830 (epr) 36.98160 (esentation) -325.00500 (learn\055) ] TJ 11.95590 TL T* [ (ing) -379.01800 (in) -380.00300 (r) 37.01960 (ecent) -378.99700 (year) 10 (s\056) -699.01200 (An) -378.98700 (ideal) -379.01500 (model) -380 (for) -379.01800 (multimodal) -380 (data) ] TJ 11.95510 TL T* [ (can) -424.98200 (r) 37.01960 (eason) -423.99100 (about) -425 (missing) -424.98800 (modalities) -424 (using) -424.99800 (the) -425.02000 (available) ] TJ T* [ (ones\054) -247.00400 (and) -245.98300 (usually) -246.00700 (pr) 44.98390 (o) 10.00320 (vides) -245.98800 (mor) 36.98770 (e) -246.00300 (information) -246.01300 (when) -246.01800 (multiple) ] TJ T* [ (modalities) -201.01600 (ar) 36.98520 (e) -202.01000 (being) -201.00700 (consider) 37 (ed\056) -293.98600 (All) -201.00700 (the) -201.99800 (pr) 36.98650 (e) 15.01220 (vious) -201.00600 (deep) -202.00300 (mod\055) ] TJ T* [ (els) -318.99600 (contain) -317.99500 (separ) 14.99630 (ate) -318.99800 <6d6f64616c6974792d73706563690263> -319.00300 (networks) -317.99100 (and) -318.99100 <026e64> -318.99100 (a) ] TJ T* [ (shar) 36.99630 (ed) -285.99100 (r) 37.01960 (epr) 36.98100 (esentation) -285.01100 (on) -285.98200 (top) -286.01400 (of) -286.01800 (those) -284.99300 (networks\056) -418.01100 (Ther) 36.99630 (efor) 36.98890 (e) 9.99343 (\054) ] TJ 11.95630 TL T* [ (the) 30.00300 (y) -341.00600 (only) -340.98900 (consider) -341.01500 (high) -341.01900 (le) 14.98030 (vel) -341.98300 (inter) 14.99870 (act) 0.98758 (ions) -342 (between) -340.98500 (modal\055) ] TJ 11.95470 TL T* [ (ities) -334 (to) -332.99100 <026e64> -334.00700 (a) -334.01800 (joint) -333.99100 (r) 37.01960 (epr) 36.98160 (esentation) -333.00300 (for) -334.00400 (them\056) -560.99200 (In) -333.98100 (this) -333.99800 (paper) 111.01800 (\054) ] TJ T* [ (we) -287.98200 (pr) 44.98390 (opose) -286.99400 (a) -287.98600 (multimodal) -286.99600 (deep) -287.99000 (learning) -287.01300 (fr) 14.99010 (ame) 14.98030 (work) -288.01100 (\050MDL\055) ] TJ T* [ (CW\051) -265.00300 (that) -264.99000 (e) 19.99240 (xploits) -264.00100 (the) -265.00700 (cr) 45.01330 (oss) -264.99700 (weights) -265.01500 (between) -264.99700 (r) 37.01830 (epr) 36.98160 (esentation) ] TJ T* [ (of) -263.98300 (modalities\054) -267.98300 (and) -264.02000 (try) -265.00600 (to) -263.98300 (gr) 14.98890 (adually) -264.00400 (learn) -264.01500 (inter) 15 (actions) -264.00300 (of) -265.00200 (the) ] TJ 11.95590 TL T* [ (modalities) -205.01500 (in) -205.99300 (a) -204.98000 (deep) -204.98400 (network) -204.99600 (manner) -205.98500 (\050fr) 44.98270 (om) -205.01700 (low) -205 (to) -205.01200 (high) -205.98300 (le) 14.98030 (vel) ] TJ 11.95510 TL T* [ (inter) 14.99870 (actions\051\056) -765.99800 (Mor) 36.98650 (eo) 10 (ver) 110.99900 (\054) -440.00300 (we) -402.00300 (theor) 36.99260 (etically) -403.01100 (s) 0.98513 (how) -403.00900 (that) -401.98700 (con\055) ] TJ T* [ (sidering) -287.99300 (these) -287.00100 (inter) 14.99870 (actions) -288 (pr) 44.98390 (o) 10.00320 (vide) -287.00700 (mor) 36.98890 (e) -287.99600 (intr) 15.00240 (a\055modality) -287.99400 (in\055) ] TJ T* [ (formation\054) -245.98500 (and) -245.00300 (intr) 44.99860 (oduce) -243.99200 (a) -245.01300 (multi\055sta) 10.00320 (g) 10.00320 (e) -244.98400 (pr) 36.98650 (e\055tr) 14.98280 (aining) -244.98400 (method) ] TJ T* [ (that) -296.98500 (is) -296.01600 (based) -296.99300 (on) -297 (the) -295.98200 (pr) 44.98390 (operties) -296.98100 (of) -296.01700 (multi\055modal) -296.99700 (data\056) -450.01800 (In) -297.00700 (the) ] TJ T* [ (pr) 44.98390 (oposed) -250.99500 (fr) 14.99140 (ame) 14.98160 (work\054) -250.99500 (as) -249.98600 (opposed) -250.99000 (to) -251.00500 (the) -249.99000 (e) 19.99180 (xisting) -250.98300 (deep) -251.01400 (meth\055) ] TJ 11.95590 TL T* [ (ods) -191.99100 (for) -193.00800 (multi\055modal) -191.99500 (data\054) -203.98100 (we) -191.99900 (try) -193.01800 (to) -191.99500 (r) 37.01960 (econstruct) -192.01700 (the) -193.01900 (r) 37.01830 (epr) 36.98160 (esen\055) ] TJ 11.95510 TL T* [ (tation) -242.00100 (of) -241.98700 (eac) 15.01340 (h) -241.99400 (modality) -242.00300 (at) -241.98700 (a) -241.99400 (given) -241.99000 (le) 14.98030 (vel\054) -243.99900 (with) -242.00600 (r) 37.01830 (epr) 36.98160 (esentation) ] TJ T* [ (of) -316.99400 (other) -316.00700 (modalities) -316.99700 (in) -316.01300 (the) -316.99900 (pr) 36.98520 (e) 15.01220 (vious) -316.00700 (layer) 111.01100 (\056) -509.99100 (Extensive) -316.99400 (e) 19.99180 (xper) 20 (\055) ] TJ T* [ (imental) -319.99500 (r) 37.01960 (esults) -319.99800 (show) -319.99600 (that) -320 (the) -320.01800 (pr) 44.98390 (oposed) -320.00300 (model) -320.01000 (outperforms) ] TJ T* [ (state\055of\055the\055art) -322.01400 (information) -322.00100 (r) 37.01960 (etrie) 14.99010 (val) -322.01700 (methods) -321.99800 (for) -322.98800 (both) -322.00300 (im\055) ] TJ T* [ (a) 10.00320 (g) 10.00320 (e) -404.01500 (and) -405.01500 (te) 20 (xt) -404.00800 (queries) -405.01000 (on) -404.00100 (the) -404.98400 (P) 89.98870 (ASCAL\055sentence) -404.00600 (and) -405.01600 (SUN\055) ] TJ 11.95590 TL T* [ (Attrib) 19.99420 (ute) -249.99000 (databases\056) ] TJ /R17 11.95520 Tf 35.92230 TL T* [ (1\056) -249.99000 (Intr) 18.01460 (oduction) ] TJ /R19 9.96260 Tf 11.95510 -18.94800 Td [ (In) -454.98100 (real\055w) 9.99343 (orld) -453.98900 (applications\054) -505.98900 (we) -454.98400 (usually) -454.01000 (encounter) -455.01200 (data) ] TJ -11.95510 -11.95470 Td [ (consisting) -358.98800 (of) -360.01700 (dif) 24.98480 (ferent) -359.00100 (modalities\056) -639.02000 (Images) -358.98700 (annotated) -359.99600 (with) ] TJ 11.95510 TL T* [ (tags) -352.98600 (and) -352.99400 (videos) -353.99500 (containing) -353.00100 (audio) -352.98200 (signal) -353.01200 (along) -354.00200 (wi) 0.99003 (th) -354.00700 (visual) ] TJ T* [ (signal) -423.00100 (are) -423.98000 (e) 15.01280 (xamples) -423 (of) -424.00600 (multimodal) -423.00300 (data\056) -830.01300 (Although) -424.00500 (each) ] TJ 11.95590 TL T* [ (modality) -283.98700 (has) -285.00500 (i) 0.98758 (ts) -284.99800 (o) 24.98110 (wn) -283.98400 (information) -283.98800 (and) -285.00500 (stati) 1.00228 (stical) -284.99200 (properties\054) ] TJ 11.95510 TL T* [ (dif) 24.98540 (ferent) -287.01400 (modalities) -287.00200 (usually) -287.02000 (share) -287.01200 (high) -286.98900 (le) 25.01790 (v) 14.98280 (el) -287.00800 (concepts\056) -421.99100 (The) ] TJ T* [ (rationale) -205.99000 (for) -206.00600 (using) -206.99600 (dif) 24.98480 (ferent) -206.00700 (modalities) -205.99600 (to) -205.99300 (learn) -206.00500 (a) -206.99000 (sha) 0.98758 (red) -206.98800 (rep\055) ] TJ T* [ (resentation) -405.98400 (is) -405.99700 (that) -405.99600 (those) -405.99200 (modalities) -406.00200 (may) -405.99800 (pro) 14.98520 (vide) -405.99800 (comple\055) ] TJ T* [ (mentary) -191.99000 (information) -192.00300 (about) -191.98900 (a) -192.01200 (common) -192.01100 (concept\056) -290.99800 (Moreo) 14.99260 (v) 14.98280 (er) 39.98600 (\054) -204.00300 (if) ] TJ 258.75000 473.32000 Td [ (we) -224.00400 (could) -225.00400 (model) -224.01700 (and) -225.01600 (learn) -224.00200 (the) -223.99200 (cross\055modal) -224.99200 (relations\054) -228.99100 (we) -224.98200 (can) ] TJ T* [ (reason) -259.01100 (about) -260.01800 (missing) -259.00600 (modalities) -259.98600 (using) -258.98600 (the) -259.00600 (a) 19.99180 (v) 24.98110 (ailable) -259.98100 (modal\055) ] TJ 11.95590 TL T* [ (ities) -246.99600 (or) -247.99500 (e) 25.01050 (v) 14.98280 (en) -247.01800 (impro) 15.00480 (v) 14.98280 (e) -246.98300 (the) -247.99000 (unimodal) -247.01300 (models\056) -309.01200 (In) -247.01800 (recent) -248.00300 (years\054) ] TJ 11.95510 TL T* [ (there) -368.01700 (has) -366.99000 (been) -368.01500 (a) -367.98300 (gro) 24.98360 (wing) -367.01700 (interest) -367.99700 (in) -366.98800 (using) -367.98800 (deep) -368.01700 (netw) 10.00810 (orks) ] TJ T* [ (for) -222.98500 (multi\055modal) -223 (learning\056) -300.98900 (Ngiam) ] TJ /R23 9.96260 Tf 132.79500 0 Td [ (et) -223.01700 (al\056) ] TJ /R19 9.96260 Tf 22.65310 0 Td (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 467.62800 506.49900 Tm (13) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 477.59000 506.49900 Tm [ (\135) -222.98700 (used) -223.01200 (deep) -222.98200 (auto\055) ] TJ -168.72800 -11.95510 Td [ (encoder) -210.00700 (to) -209.01200 (e) 15.01220 (xtract) -209.99700 (high) -209.98200 (le) 25.01790 (v) 14.98280 (el) -208.98000 (features) -210.00400 (from) -209.01400 (speech) -210 (and) -210 (video) ] TJ 11.95510 TL T* [ (signals\054) -289.99600 (and) -281.98700 (then) -282.01900 (aggre) 15.01470 (g) 4.98446 (ated) -281.00700 (these) -281.98200 (features) -281.99200 (to) -282.02200 <026e64> -282.00700 (a) -281.99700 (shared) ] TJ T* [ (representation\056) -446.01600 (Sri) 24.99340 (v) 24.98110 (asta) 20.00160 (v) 24.98110 (a) ] TJ /R23 9.96260 Tf 106.86800 0 Td [ (et) -295.00500 (al\056) ] TJ /R19 9.96260 Tf 24.81680 0 Td (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 443.86400 470.63400 Tm (20) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 453.82700 470.63400 Tm [ (\135) -295.01400 (introduced) -296 (the) -295 (multi\055) ] TJ -144.96500 -11.95630 Td [ (modal) -212.99900 (deep) -212.98200 (Boltzman) -212.98200 (machine) -213.00900 (that) -213.00900 (learns) -213.01800 (a) -212.98900 (deep) -212.98400 (generati) 24.98600 (v) 14.98280 (e) ] TJ 11.95470 TL T* [ (model) -292.00500 (o) 14.98280 (v) 14.98280 (er) -291.99800 (joint) -291.99800 (space) -291.99300 (of) -291.01000 (image) -292.01500 (and) -291.98600 (te) 14.98280 (xt) -292.02000 (inputs\056) -436.00600 (Recently) 64.99410 (\054) ] TJ (Sohn) ' /R23 9.96260 Tf 24.27300 0 Td [ (et) -380.01300 (al\056) ] TJ /R19 9.96260 Tf 28.21170 0 Td (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 364.66400 434.76800 Tm (19) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 374.62700 434.76800 Tm [ (\135) -380.01700 (proposed) -381.01000 (a) -380.02000 (training) -381.00500 (approach) -380.01000 (for) -380.99500 (mult) 1.00964 (i\055) ] TJ -65.76480 -11.95510 Td [ (modal) -308.00300 (deep) -307.98500 (learning) -308.01000 (that) -306.99300 (is) -308.01200 (based) -308.01700 (on) -308.01700 (minimizing) -308.00300 (v) 24.98110 (ariation) ] TJ 11.95510 TL T* [ (of) -253.01400 (information) -251.99200 (instead) -252.99500 (of) -251.99500 (maximizing) -253.00200 (the) -251.98700 (lik) 10.01790 (elihood\056) -317.99100 (All) -252.99700 (of) ] TJ T* [ (these) -372.00700 (methods) -371 (for) -372.01900 (multimodal) -371.99200 (data) -372.01100 (ha) 19.99670 (v) 14.98280 (e) -371.00200 (a) -371.98200 (common) -371.98200 (strat\055) ] TJ 11.95590 TL T* [ (e) 15.01220 (gy) -225.01100 (in) -225.00900 (which) -225.00700 (the) 14.98520 (y) -225.01600 <02727374> -224.99700 (learn) -226.00100 (la) 0.98513 (yers) -225.99700 (of) -225.02100 <6d6f64616c6974792d73706563690263> -225.00200 (rep\055) ] TJ 11.95510 TL T* [ (resentations) -365.00500 (and) -364.99300 (then) -364.00500 (learn) -364.99800 (a) -365.00300 (shared) -364.99300 (representation) -365.01300 (across) ] TJ T* [ (multiple) -249.99000 (modalities) -249.98800 (at) -249.99500 (the) -249.98800 (top) -250.01700 (layer) -249.99700 (of) -249.99700 (the) -249.98800 (deep) -249.99700 (netw) 10.00810 (ork\056) ] TJ T* [ (Besides) -224.99200 (deep) -226 (multimodal) -224.99700 (learning) -225.00200 (approaches\054) -229.99100 (some) -225.98200 (proba\055) ] TJ T* [ (bilistic) -212.00100 (methods) -212.00600 (ha) 19.99670 (v) 14.98280 (e) -212.00900 (also) -211.98900 (been) -212.00400 (recently) -213.01300 (introduce) 1.00473 (d) -213.01800 (for) -212.00400 (mul\055) ] TJ 11.95590 TL T* [ (timodal) -408.01000 (data\056) -783.98200 (Xing) ] TJ /R23 9.96260 Tf 85.39410 0 Td [ (et) -408.00800 (al\056) ] TJ /R19 9.96260 Tf 29.30590 0 Td (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 426.88000 327.17100 Tm (24) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 436.84200 327.17100 Tm [ (\135) -408.01500 (proposed) -407.98600 (dual\055wing) -407.98100 (har) 19.99180 (\055) ] TJ -127.98000 -11.95510 Td [ (moniums) -370.00400 (to) -368.98500 (learn) -370.01700 (a) -369.98200 (joint) -369.98700 (representation) -369.01200 (of) -370.01700 (the) -370.00700 (image) -370.00200 (and) ] TJ 11.95510 TL T* [ (te) 14.98030 (xt) -346.98900 (modalities\056) -602.00500 (Zhen) ] TJ /R23 9.96260 Tf 92.18520 0 Td [ (et) -346.99800 (al\056) ] TJ /R19 9.96260 Tf 26.88790 0 Td (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 431.25300 303.26100 Tm (26) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 441.21500 303.26100 Tm [ (\135) -347.00600 (introduced) -347.01100 (a) -347.98600 (probabilis\055) ] TJ -132.35300 -11.95510 Td [ (tic) -227.01100 (generati) 24.98360 (v) 14.98280 (e) -228.00600 (approach) -227.01600 (called) -227.99600 (multimodal) -226.99600 (latent) -227.00100 (binary) -227.98600 (em\055) ] TJ 11.95470 TL T* [ (bedding\056) -518.98200 (Ozdemir) ] TJ /R23 9.96260 Tf 77.79880 0 Td [ (et) -319.00300 (al\056) ] TJ /R19 9.96260 Tf 25.78320 0 Td (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 415.76200 279.35200 Tm (14) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 425.72400 279.35200 Tm [ (\135) -319.99100 (presented) -318.99600 (a) -319.99100 (model) -318.98100 (based) -320.01500 (on) ] TJ -116.86200 -11.95510 Td [ (Bayesian) -421.98300 (nonparametric) -421.98600 (frame) 25.01050 (w) 10.00320 (ork) -422.00300 (to) -421.99800 (learn) -422.98800 (the) -421.99800 (underly\055) ] TJ 11.95590 TL T* [ (ing) -340.98500 (semantically) -339.98200 (meaningful) -341.00700 (and) -340.99700 (abstract) -340.01200 (features) -341.00200 (of) -341.00200 (mul\055) ] TJ 11.95510 TL T* [ (timodal) -243.99900 (data\056) -307.98300 (Ho) 24.98360 (we) 25.01300 (v) 14.98280 (er) 39.98600 (\054) -245.01500 (since) -243.98900 (dif) 24.98600 (ferent) -244.00300 (data) -243.99400 (modalities) -243.98900 (ha) 19.99670 (v) 14.98280 (e) ] TJ T* [ (dif) 24.98600 (ferent) -295.01200 (st) 0.99248 (atistical) -294.98000 (properties\054) -306.00300 (shallo) 25 (w) -293.98500 (models) -295.01900 (are) -294 (not) -294.99000 (usu\055) ] TJ T* [ (ally) -427.99200 (able) -426.98500 (to) -427.99500 (e) 15.01220 (xtract) -428 (high\055le) 24.99580 (v) 14.98280 (el) -426.98300 (concepts) -427.99200 (from) -427.99700 (multimodal) ] TJ (data\056) ' T* [ (One) -634.00900 (of) -634.99000 (the) -634.00500 (other) -634.00200 (recent) -634.01700 (challenges) -633.99200 (about) -635.01200 (multimodal) ] TJ 11.95590 TL T* [ (datasets) -393.99300 (is) -394 (their) -394 (huge) -394.00300 (size\056) -741.00900 (In) -394.01300 (order) -394.01300 (to) -394.00300 (achie) 25.01540 (v) 14.98280 (e) -394.01800 (real) -394.01800 (time) ] TJ 11.95510 TL T* [ (retrie) 25.01540 (v) 24.98110 (al\054) -267.00900 (some) -264.01500 (approaches) -263.99000 (such) -263.98500 (as) -263.99500 (multimodal) -264.01000 (hashing) -264.00500 (has) ] TJ T* [ (been) -267.99200 (introduced) -266.98700 (which) -268.01900 (encodes) -267.00400 (the) -267.98500 (high\055dimensional) -268.00400 (input) ] TJ T* [ (v) 14.98280 (ectors) -251.01000 (into) -250.99200 (compact) -251.98000 (binary) -251.00200 (strings) -250.98300 (while) -250.99700 (trying) -252.01200 (t) 0.98513 (o) -251.99200 (increase) ] TJ T* [ (similarity) -211.99200 (between) -211.99600 (dif) 24.98600 (ferent) -212.00900 (modalities) -211.99400 (that) -211.98900 (ha) 19.99670 (ving) -211.97900 (the) -211.99400 (same) ] TJ T* [ (concept) -251.98700 (in) -251.98500 (the) -251.99000 (result) 0.99738 (ed) -251.99700 (space\056) -315.98600 (F) 14.99260 (or) -251.99700 (e) 15.01220 (xample\054) -252.01200 (Bronstein) ] TJ /R23 9.96260 Tf 216.30700 0 Td [ (et) -251.99200 (al\056) ] TJ /R19 9.96260 Tf -216.30700 -11.95590 Td (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 312.18000 111.97900 Tm (2) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 317.16100 111.97900 Tm [ (\135) -363.98100 (applied) -365.01800 (a) -363.98300 (boosting) -364.99100 (procedure) -364.01800 (for) -364.01800 (cross\055modality) -364.98800 (simi\055) ] TJ -8.29883 -11.95510 Td [ (larity) -300.98900 (learning\056) -463.98100 (Zhen) ] TJ /R23 9.96260 Tf 87.26410 0 Td [ (et) -301.00600 (al\056) ] TJ /R19 9.96260 Tf 25.05510 0 Td (\133) Tj ET Q 0 1 0 rg q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 424.49900 100.02300 Tm (25) Tj ET Q 0 g q 10 0 0 10 0 0 cm BT /R19 9.96260 Tf 1 0 0 1 434.46100 100.02300 Tm [ (\135) -301.01300 (proposed) -300.98400 (a) -301.99400 (co\055re) 15.00730 (gularized) ] TJ -125.59900 -11.95510 Td [ (hashing) -391.98400 (based) -392.00600 (on) -392.00300 (a) -392.99800 (boos) 1 (ted) -392.98300 (co\055re) 15.00730 (gularization) -391.97900 (frame) 25.01050 (w) 10.00320 (ork\056) ] TJ -13.74100 -36.95590 Td (1) Tj ET Q Q Q q q 1 1 1 rg /a0 gs 48.40600 786.42200 515.18800 -52.69900 re f q /s5 gs /x6 Do Q q /s7 gs /x8 Do Q q /s9 gs /x10 Do Q q /s11 gs /x12 Do Q Q Q Q q 1 0 0 1 0 0 cm BT /F1 12 Tf 14.40000 TL ET 1 1 1 rg n 270 47 72 14 re f* 0.50000 0.50000 0.50000 rg BT /F2 9 Tf 10.80000 TL ET BT 1 0 0 1 297 50 Tm (2601) Tj T* ET Q endstream endobj 14 0 obj << /Filter /FlateDecode /Resources << /ExtGState << /a0 << /CA 1 /ca 1 >> >> /XObject << /x18 15 0 R >> >> /Length 28 /Group << /Type /Group /S /Transparency /CS /DeviceRGB /I true >> /BBox [ 78 746 96 765 ] /Type /XObject /Subtype /Form >> stream x+O4PH/VЯ0Pp 0 endstream endobj 15 0 obj << /Filter /FlateDecode /Resources 16 0 R /Length 107 /Type /XObject /BBox [ 78 746 96 765 ] /Subtype /Form >> stream xe AC̬wʠ =p,?]%+H-
Jc "82w8VSnGW;"
endstream
endobj
16 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
17 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x15 18 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 67 752 84 775 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ04Up
0
endstream
endobj
18 0 obj
<<
/Filter /FlateDecode
/Resources 19 0 R
/Length 228
/Type /XObject
/BBox [ 67 752 84 775 ]
/Subtype /Form
>>
stream
xeQKn!s ?FPav6R٪TS.
b];15YyR
{7QL.\:Rv/x9l+L7h%1!}i/AI(kz"U&,YO![R hg{3}4/GyYF:!w}Gn+'xJcO9i뽼_-:`
endstream
endobj
19 0 obj
<<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
>>
endobj
20 0 obj
<<
/Filter /FlateDecode
/Resources <<
/ExtGState <<
/a0 <<
/CA 1
/ca 1
>>
>>
/XObject <<
/x24 21 0 R
>>
>>
/Length 28
/Group <<
/Type /Group
/S /Transparency
/CS /DeviceRGB
/I true
>>
/BBox [ 132 751 480 772 ]
/Type /XObject
/Subtype /Form
>>
stream
x+O4PH/VЯ02Qp
0
endstream
endobj
21 0 obj
<<
/Filter /FlateDecode
/Resources 22 0 R
/Length 53223
/Type /XObject
/BBox [ 132 751 480 772 ]
/Subtype /Form
>>
stream
xtI:6%Q㨈?7rA= u%6 ?Y(WbWo{B>9
x`Znϳ|8{3?0x*z ǃ|,@:w>`c|*ϻⳅKO3`g
:_|}}><.6`Z{{3]#<_o"~:ͺgk7/Ұ@|K yp ]03ʷCmş8˽Y?>(3!Bwqs.Z8,~~=rMT̩y+/*w: uBZ_`ߵp`%M?ɝ1ɳw=vDۉy&xb4Q>d@ sg~lA