“School of Cognitive Sciences”
Back to Papers HomeBack to Papers of School of Cognitive Sciences
| Paper IPM / Cognitive Sciences / 14468 |
|
||||||||
| Abstract: | |||||||||
|
Recent work in neural decoding has highlighted the importance of high-dimensional neural representations in cognitive functions. Independently, investigations into interactions between neurons have pinpointed reductions in the noise correlations between neurons as a mechanism for enhancing sensory and motor processes. However, the interactions or relative contributions of high-dimensional representations and changes in noise correlations have yet to be explored within a single population dataset. To explore the role of these coding properties during a cognitive task, we examined neural activity in populations of simultaneously recorded Frontal Eye Field (FEF) neurons in monkeys trained to perform a memory-guided saccade task. In this task, the monkeys were trained to saccade to the remembered location of a visual target, which appeared in 1 of 16 possible locations (2 eccentricities and 8 angles). Importantly, this simple task requires multiple cognitive operations, including encoding of the location of a visual target (stimulus encoding), maintenance of this information over time (working memory), and finally generating a saccadic eye movement to the remembered location (saccadic preparation). We applied a combination of decoding and encoding methods to the simultaneously recorded neural data. We found that the information content of the neural response increased between sensory encoding and saccade preparation, but differently for the single-cell and population responses. Specifically, we found that population of FEF neurons greatly enhance their ability to encode locations far beyond their original response field, but only during saccade preparation only when population activity is taken into account. In other words, the information content of the population response has expanded across space during saccade preparation. We showed that this expansion relies on a high-dimensional neural representation. Finally, we found that a reduction in the noise correlation can further increase the information content of the population; however, this contribution can be detected only when the change in signal correlation is accounted for since the latter is much larger and affects both the single-cell and population responses. Overall, our results provide new insights into which properties of neural population codes are most relevant for encoding information and different cognitive processes.
Download TeX format |
|||||||||
| back to top | |||||||||
