A Learning Algorithm for Synfire Chains

Neu robiological studies ind icate very p recise tempo ral behavior of neuron firings. Abeles [1] has recorded spik e timing of different co rtical cells and, in particular, has observed th e following level of precisio n: when a neuron A fires, neuron B w ould fire 15 1ms later an d neu ron C would fire p recisely 289 ms after that—with a precision across trials of 1 ms! Such lon g delays req uire d ozens of com bined transm ission delay s from the p resynaptic n euron (A) to th e postsynaptic neu ron (C ). Th e mech anism proposed by Abeles for g enerating su ch precise d elayed synch ronization h as been called syn fire chains. Ho w could synfire ch ains dev elop? What learning pro cedure could generate su ch precise temporal chains? How cou ld a connectionist netwo rk of spikin g neurons learn sy nfire chains? An algorithm fo r a network o f spik ing neurons that learns synfire chains will be p resented.