%PDF-1.3 1 0 obj << /Kids [ 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 8 >> endobj 2 0 obj << /Subject (Neural Information Processing Systems http\072\057\057nips\056cc\057) /Publisher (Curran Associates) /Language (en\055US) /Created (2008) /Description-Abstract (In a variety of behavioral tasks\054 subjects exhibit an automatic and apparently sub\055optimal sequential effect\072 they respond more rapidly and accurately to a stimulus if it reinforces a local pattern in stimulus history\054 such as a string of repetitions or alternations\054 compared to when it violates such a pattern\056 This is often the case even if the local trends arise by chance in the context of a randomized design\054 such that stimulus history has no predictive power\056 In this work\054 we use a normative Bayesian framework to examine the hypothesis that such idiosyncrasies may reflect the inadvertent engagement of fundamental mechanisms critical for adapting to changing statistics in the natural environment\056 We show that prior belief in non\055stationarity can induce experimentally observed sequential effects in an otherwise Bayes\055optimal algorithm\056 The Bayesian algorithm is shown to be well approximated by linear\055exponential filtering of past observations\054 a feature also apparent in the behavioral data\056 We derive an explicit relationship between the parameters and computations of the exact Bayesian algorithm and those of the approximate linear\055exponential filter\056 Since the latter is equivalent to a leaky\055integration process\054 a commonly used model of neuronal dynamics underlying perceptual decision\055making and trial\055to\055trial dependencies\054 our model provides a principled account of why such dynamics are useful\056 We also show that near\055optimal tuning of the leaky\055integration process is possible\054 using stochastic gradient descent based only on the noisy binary inputs\056 This is a proof of concept that not only can neurons implement near\055optimal prediction based on standard neuronal dynamics\054 but that they can also learn to tune the processing parameters without explicitly representing probabilities\056) /Producer (Python PDF Library \055 http\072\057\057pybrary\056net\057pyPdf\057) /Title (Sequential effects\072 Superstition or rational behavior\077) /Date (2008) /Type (Conference Proceedings) /firstpage (1873) /Book (Advances in Neural Information Processing Systems 21) /Description (Paper accepted and presented at the Neural Information Processing Systems Conference \050http\072\057\057nips\056cc\057\051) /Editors (D\056 Koller and D\056 Schuurmans and Y\056 Bengio and L\056 Bottou) /Author (Angela J\056 Yu\054 Jonathan D\056 Cohen) /lastpage (1880) >> endobj 3 0 obj << /Type /Catalog /Pages 1 0 R >> endobj 4 0 obj << /Contents [ 12 0 R 13 0 R 14 0 R 15 0 R 16 0 R 17 0 R 18 0 R 19 0 R ] /Rotate 0 /Resources << /ExtGState << /GS0 20 0 R >> /Font << /T1_4 21 0 R /T1_2 26 0 R /T1_3 27 0 R /T1_0 29 0 R /T1_1 30 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 5 0 obj << /Contents 32 0 R /Rotate 0 /Resources << /ExtGState << /GS0 20 0 R >> /XObject << /Im0 33 0 R >> /Font << /T1_21 29 0 R /T1_20 27 0 R /T1_6 34 0 R /T1_7 37 0 R /T1_4 40 0 R /T1_5 44 0 R /T1_2 48 0 R /T1_3 51 0 R /T1_0 54 0 R /T1_1 56 0 R /T1_8 57 0 R /T1_9 61 0 R /T1_10 64 0 R /T1_11 68 0 R /T1_12 30 0 R /T1_13 21 0 R /T1_14 71 0 R /T1_15 75 0 R /T1_16 80 0 R /T1_17 84 0 R /T1_18 87 0 R /T1_19 90 0 R >> /ProcSet [ /PDF /Text /ImageC ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 6 0 obj << /Contents 93 0 R /Rotate 0 /Resources << /ExtGState << /GS0 20 0 R >> /ColorSpace << /CS0 94 0 R >> /XObject << /Im13 96 0 R /Im12 97 0 R /Im11 98 0 R /Im10 99 0 R /Im14 100 0 R /Im7 101 0 R /Im6 102 0 R /Im5 103 0 R /Im4 104 0 R /Im3 105 0 R /Im2 106 0 R /Im1 107 0 R /Im0 108 0 R /Im9 109 0 R /Im8 110 0 R >> /Font << /T1_6 56 0 R /T1_7 30 0 R /T1_4 40 0 R /T1_5 34 0 R /T1_2 44 0 R /T1_3 51 0 R /T1_0 48 0 R /T1_1 37 0 R /T1_8 21 0 R /T1_9 75 0 R /T1_10 71 0 R /T1_11 80 0 R /T1_12 87 0 R /T1_13 90 0 R /T1_14 84 0 R /T1_15 111 0 R /T1_16 112 0 R /T1_17 27 0 R >> /ProcSet [ /PDF /Text /ImageC /ImageI ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 7 0 obj << /Contents 116 0 R /Rotate 0 /Resources << /ExtGState << /GS0 20 0 R >> /Font << /T1_21 84 0 R /T1_20 117 0 R /T1_23 111 0 R /T1_22 112 0 R /T1_6 34 0 R /T1_7 40 0 R /T1_4 44 0 R /T1_5 51 0 R /T1_2 48 0 R /T1_3 121 0 R /T1_0 54 0 R /T1_1 56 0 R /T1_8 37 0 R /T1_9 57 0 R /T1_10 61 0 R /T1_11 30 0 R /T1_12 21 0 R /T1_13 75 0 R /T1_14 71 0 R /T1_15 90 0 R /T1_16 80 0 R /T1_17 87 0 R /T1_18 29 0 R /T1_19 27 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 8 0 obj << /Contents 125 0 R /Rotate 0 /Resources << /ExtGState << /GS0 20 0 R >> /Font << /T1_6 84 0 R /T1_7 87 0 R /T1_4 75 0 R /T1_5 90 0 R /T1_2 80 0 R /T1_3 117 0 R /T1_0 30 0 R /T1_1 21 0 R /T1_8 71 0 R /T1_9 126 0 R /T1_10 127 0 R /T1_11 111 0 R /T1_12 130 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 9 0 obj << /Contents 133 0 R /Rotate 0 /Resources << /ExtGState << /GS0 20 0 R >> /ColorSpace << /CS1 134 0 R /CS0 136 0 R >> /XObject << /Im7 138 0 R /Im6 139 0 R /Im5 140 0 R /Im4 141 0 R /Im3 142 0 R /Im2 143 0 R /Im1 144 0 R /Im0 145 0 R >> /Font << /T1_6 75 0 R /T1_7 48 0 R /T1_4 146 0 R /T1_5 148 0 R /T1_2 151 0 R /T1_3 29 0 R /T1_0 54 0 R /T1_1 56 0 R /T1_8 21 0 R /T1_9 30 0 R /T1_10 71 0 R /T1_11 84 0 R /T1_12 80 0 R /T1_13 27 0 R /T1_14 90 0 R /T1_15 87 0 R /T1_16 111 0 R >> /ProcSet [ /PDF /Text /ImageC /ImageI ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 10 0 obj << /Contents 153 0 R /Rotate 0 /Resources << /ExtGState << /GS0 20 0 R >> /Font << /T1_6 27 0 R /T1_4 90 0 R /T1_5 29 0 R /T1_2 71 0 R /T1_3 80 0 R /T1_0 30 0 R /T1_1 21 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 11 0 obj << /Contents 154 0 R /Rotate 0 /Resources << /ExtGState << /GS0 20 0 R >> /Font << /T1_4 29 0 R /T1_2 21 0 R /T1_3 71 0 R /T1_0 30 0 R /T1_1 27 0 R >> /ProcSet [ /PDF /Text ] >> /CropBox [ 0 0 612 792 ] /Parent 1 0 R /MediaBox [ 0 0 612 792 ] /Type /Page >> endobj 12 0 obj << /Length 589 /Filter /FlateDecode >> stream HSn0) bD4i\M/) (hؒGZ}Q|,wvf}"Om$@Y߄\~rlCxIƹf<&m1AɆLw?'ep'cRR, gLRq,W=ڧV3(ve~b @6m:WWO906 YٕDJ]|[@ _37Uʁi LJ $Bng*m$8eHt=a32XUgKb)Y"{jwhp^/^Vh`YV ,?RHJ~Truw5ʙ qj8듖I8˕ZW}Ι^!Kgt+VDn+[k6 ȸȹ»v_MP*$Mk+b͔#jwۀwӡb3܅by2@gmט;I`@gԄ9vqcЎ]vδE8oC'{VnptؾaYZo<y` (+1M endstream endobj 13 0 obj << /Length 571 /Filter /FlateDecode >> stream HlSM0+93q6.9cg%RʯGܲɊ>jZ4V×-U-|x>⛬zj*)wa,㒅̺=6L\dtS.Jl8zLܼ a츔hh} *8