386 



John R. Platt 



finger tracing of block letters and large patterns by children and newly-sighted 

 adults suggests its early importance. Studies of the developmental pathology of 

 pattern-perception with partial oculomotor paralysis might be instructive.^ 



Null Detectors and Delay Lines 



What kinds of neural connections might be needed in the mosaic to deter- 

 mine addresses as these operations are performed ? 



The discriminating self-congruence information is always of the form 

 'constant repetition of the same pattern' or 'repetition after a time delay'. 

 What is needed from the photodetectors is the information 'absence of change'. 

 They are being used as null detectors. Unstable detector elements in the labora- 

 tory are often used in the same way whenever the utmost accuracy of measure- 

 ment and simplicity of interpretation is wanted. The rather complex relation- 

 ships that can be established when using mosaic receptors as null detectors seem 

 not to have been explored before. 



To signal 'no change' we could use a null cell of the type already described. 

 But another good way to examine stability of pattern would be to have a cell 

 with two input channels of different lengths, like two of the channels in Fig. 4, 



Image 



Fig. 4. Cone addresses from delayed coincident pulses to a 

 null-transmitting velocity-detector cell. 



or of different diameters and travel times, where the cell sensitivity is such as to 

 require simultaneous spikes from both channels in order to produce an outgoing 

 spike in its axon. (The word 'channels' is used to avoid the experimentally 

 unsettled question as to whether these could be all-or-none dendrites of the cell, 

 if such exist, or two excitatory synapses with different delays in the axons, or 

 collateral processes from the cells of the preceding stage.) 



Such input channels would be delay lines like those used in nuclear physics 

 to distinguish certain events and particles, to eliminate random spurious 

 counts, and to measure velocities. They might be used for all these purposes 

 here. The axon output of such a cell, as shown in Fig. 4, combines three im- 

 portant properties. It is (a) a null indicator, firing only for those patterns that 

 are identical in the input channels. It is (b) an indicator of a particular delay 

 time or difference of times in the channels. And it is (c) a pattern transmitter, 

 since the pattern is not lost. Let us call this a null transmitter cell, and if the 



