COMPOUND CONDITIONING UNDER TEMPORAL UNCERTAINTY Robert J. Polewan & John W. Moore* University of Massachusetts Amherst Neuroscience & Behavior Program Compound Conditioning Under Temporal Uncertainty
Download ReportTranscript COMPOUND CONDITIONING UNDER TEMPORAL UNCERTAINTY Robert J. Polewan & John W. Moore* University of Massachusetts Amherst Neuroscience & Behavior Program Compound Conditioning Under Temporal Uncertainty
COMPOUND CONDITIONING UNDER TEMPORAL UNCERTAINTY Robert J. Polewan & John W. Moore* University of Massachusetts Amherst Neuroscience & Behavior Program Compound Conditioning Under Temporal Uncertainty Eyeblink conditioning has long been a model for understanding behavioral and physiological processes of learning, memory, and performance. The present research extends our previous studies of rabbit eyeblink conditioning under temporal uncertainty to compound conditioning. Neuroscience & Behavior Program 2 Temporal Uncertainty Training Rabbits were trained to make eyeblink conditioned responses (CRs) to a compound conditioned stimulus (CS) consisting of a tone (T) and a light (L) presented simultaneously an reinforced with an unconditioned stimulus (US). This training involved a mixture of two CS-US intervals. On some trials, the US occurred 300 ms after CS onset; on other trials, the US occurred 700 ms after CS onset. Neuroscience & Behavior Program 3 Temporal Uncertainty Training Randomly mixing trials with these CS-US intervals produced bimodal CR waveforms with amplitude peaks located at the two temporal loci of the US, temporal windows centered at 300 and 700 ms. Neuroscience & Behavior Program 4 TD (CSC) model Sutton and Barto’s (1990) TD (CSC) model is a representational system capable of describing the complex conditioned response waveforms instilled through training under temporal uncertainty. The model assumes a delay-line timing structure. Neuroscience & Behavior Program 5 Delay-line Timing Structure Basic tapped delay-line. Injection of CS input begins sequential propagation of signal through a delay-line. Each synapse (—<) introduces a delay; the total delay from activation of the first element in the delay-line to the last element is a direct function of the number of sequential synapses. Taps from the delay-line units send timing information to higher-order processing units. Neuroscience & Behavior Program 6 Compound Conditioning Under Temporal Uncertainty Like the Rescorla-Wagner model, the TD (CSC) model assumes that CR performance to a compound stimulus is the sum of the “associative strengths” of the components. In order to test this assumption, it is necessary to specify how the theoretical indices of CR associative strength map onto real measures of performance such as CR amplitude. Neuroscience & Behavior Program 7 Compound Conditioning Under Temporal Uncertainty Summation of CR amplitudes to component stimuli should reconstitute the CR waveforms obtained under compound conditioning. Deviations from a “simple summation” rule should indicate shortcomings and point the way to improving the model. Neuroscience & Behavior Program 8 Compound Conditioning Under Temporal Uncertainty Factors that could challenge a simple summation rule for reconstituting a compound CR waveform from its components include • Configuring/patterning • Overshadowing • Transfer from prior training. In addition, floor effects (thresholds) and ceiling effects (saturation) could complicate assessment of the model in terms of CR amplitudes. Neuroscience & Behavior Program 9 Pretraining Prior to compound conditioning training half of the 24 rabbits were pretrained to one CS (predominantly a light) at one of the two CS-US intervals. Pretrained rabbits were run concurrently with yoked control rabbits. Neuroscience & Behavior Program 10 Averaged Topographical CR Waveforms ALL RABBITS Sessions 16-20 EYELID POSITION (MM) 8 7 6 5 4 3 2 1 0 TL- + 2SEM n = 24 Peak 1 = 5.7543 mm @ 356 ms; Peak 2 = 5.1005 mm @ 724 ms 8 7 6 5 4 3 2 1 0 L- + 2SEM Peak 1 = 1.9541 mm @ 432 ms; Peak 2 = 1.9379 mm @ 792 ms n = 24 8 7 6 5 4 3 2 1 0 T- + 2SEM n = 24 0 200 Peak 1 = 3.4884 mm @ 384 ms; Peak 2 = 3.2151 mm @ 772 ms 400 600 800 1000 1200 TIME FROM CS ONSET (MS) Neuroscience & Behavior Program 11 Averaged Topographical CR Waveforms Neuroscience & Behavior Program 12 Peak Amplitudes in Decomposition Peak Amplitude (mm) Pretrained & Controls 8 7 6 5 4 3 2 1 0 TLLT- PRE Peak 1 PRE Peak 2 CONT Peak 1 CONT Peak 2 Group and Peak Position Mean peak amplitudes (+SE) to TL-, L-, and T- for the 12 pretrained (PRE) and 12 control rabbits (CONT) at both peak locations in the fourth session-block of training. Neuroscience & Behavior Program 13 Decomposition Peak Latencies Shifts Pretrained & Controls Peak Latency (ms) 900 800 700 600 500 TL- 400 300 T- L- 200 100 0 PRE Peak 1 PRE Peak 2 CONT Preak 1 CONT Peak 2 Group and Peak Location Mean peak latencies (+SE) to TL-, L-, and T- for 12 pretrained (PRE) and 12 control rabbits (CONT) at both peak locations in the fourth session-block of training. Neuroscience & Behavior Program 14 Proportional Contribution Proportion Proportional Contribution 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 PRE TCONT T- 1 2 3 4 Session Blocks Mean proportional contribution of T- to the compound waveform for both pretrained subjects (PRE) and control subjects (CONT). Neuroscience & Behavior Program 15 Gain Factors Gain Factors (N = 24) 3.0 Gain Factor 2.5 2.0 PRE 1.5 CONT 1.0 0.5 0.0 1 2 3 4 Session Blocks Mean gain factors of T- and L- for pretrained subjects (PRE) and control subjects (CONT). Neuroscience & Behavior Program 16 Sum of Squared Deviations Sum of Squared Deviation for the Combination Rules Sum of Squared Deviation 4000 3500 Pretrained 3000 Control 2500 2000 1500 1000 500 0 LSR MRR OSR SSR MPR TR WSR SAR LR Comnination Rules Mean (+SE) Sum of Squared Deviations (SSDs) for pretrained (n = 12) and control rabbits (n = 12) for nine combination rules in the last block of sessions. Neuroscience & Behavior Program 17 Summary of Findings Decomposition slowed the initiation of the motor program representing CR waveforms. The latency of initiation was greater for the light than the tone, consistent with the tone’s greater salience, as indexed by the tones greater proportional contribution to the compound. Decomposition did not affect other features of component CR waveforms, as inter-peak intervals remained unchanged. Neuroscience & Behavior Program 18 Summary of Findings The slower initiation of the motor program did not result in a later “catching up” such that the second amplitude peak appeared within the 700ms temporal window. Nor did the slower initiation result a greater temporal separation of amplitude peaks. In terms of the spreading activation account of CR topography proposed by the TD (CSC) model, the speed of propagation remained the unchanged. Neuroscience & Behavior Program 19 Discussion The slower initiation of component CR waveforms following compound conditioning may reflect a “processing cost.” Pearce has suggested that compound CSs are gestalts, and that changes in performance to component stimuli are “generalization decrements.” If so, the costs of decomposition did not extend to CR amplitudes, as amplitudes adhered to a summation combination rule. Neuroscience & Behavior Program 20 Acknowledgments Moore Lab • • • • • • Vanessa Castagna Jamy Gaynor Jordan Marks Tony Rauhut June-Seek Choi Marcy Rosenfield Thank You Neuroscience & Behavior Program 21 Neuroscience & Behavior Program 22 Summation Experiment Rabbits were trained with individual CSs, with each CS trained at a different CS-US interval (light at 300 ms and tone at 700 ms). The two CSs were only presented together on probe trials. Neuroscience & Behavior Program 23 Averaged Summation CR Waveforms ALL RABBITS Sessions 16-20 EYELID POSITION (MM) 8 7 6 5 4 3 2 1 0 TL- + 2SEM n=4 Peak = 4.6648 mm @ 772 ms 8 7 6 5 4 3 2 1 0 L- + 2SEM Peak = 5.2006 mm @ 368 ms n=4 8 7 6 5 4 3 2 1 0 T- + 2SEM n=4 0 Peak = 3.8265 mm @ 824 ms 200 400 600 800 1000 1200 TIME FROM CS ONSET (MS) Neuroscience & Behavior Program 24 Pretrained & Control Summation Rabbits CONTROL Subject B Sessions 21-25 PRETRAINED Subject A Sessions 21-25 A TL- + 2SEM 8 7 6 5 4 3 2 1 0 A L- + 2SEM 8 7 6 5 4 3 2 1 0 A T- + 2SEM 0 200 8 7 6 5 4 3 2 1 0 peak = 5.0234 mm @ 756 ms peak = 6.3253 mm @ 368 ms 600 800 TIME FROM CS ONSET (MS) Neuroscience & Behavior Program 1000 A TL- + 2SEM 8 7 6 5 4 3 2 1 0 1200 peak = 7.0509 mm @ 692 ms A L- + 2SEM 8 7 6 5 4 3 2 1 0 peak amplitude = 3.9108 mm @ 840 ms 400 EYELID POSITION (MM) EYELID POSITION (MM) 8 7 6 5 4 3 2 1 0 A T- + 2SEM 0 200 400 600 peak = 6.8979 mm @ 352 ms peak amplitude = 6.8169 mm @ 716 ms 800 1000 1200 TIME FROM CS ONSET (MS) 25 Summation Results Waveforms to compound CS showed a unimodal peak that was between the component peaks in both amplitude and latency. The peak amplitude favored the more salient tone CS resulting in a “Performance Overshadowing Effect.” Neuroscience & Behavior Program 26 Performance Overshadowing/ Summated Generalization The characteristic waveform for one stimulus (tone) is dominated the characteristic waveform for the other stimulus (light) because of the tone’s higher salience, even though the light stimulus is pretrained. One possible explanation for the intermediate temporal position of the peak is summated generalization, similar to summated generalization along dimensions such as auditory frequency (Moore, 1972). Neuroscience & Behavior Program 27 Neuroscience & Behavior Program 28 Effects of Pretraining Pretrained & Controls Peak Amplitude (mm) 8 7 6 5 TL- 4 L- 3 T- 2 1 0 PRE Peak 1 Pre Peak 2 CONT Peak 1 CONT Peak 2 Group and Peak Location Mean peak amplitudes (+SE) to TL-, L-, and T- for the 6 rabbits pretrained to light at the 300-ms ISI (PRE) and their 6 yoked controls (CONT) at both peak locations in the fourth session-block of training. Neuroscience & Behavior Program 29