ABSTRACT

The approaching era of expanded bioengineering will deal with many issues that relate to the improvement, recovery, and even replacement of neural functions. An obvious requirement of such ‘‘neurobehavioral engineering’’ is to determine how populations of neurons represent information relevant to task performance. A commonly applied method for quantifying information representation in neural ensembles is to determine how different behavioral conditions can be inferred from simultaneously recorded neural firing (1-8). In rat, position and trajectory representations have been decoded from ensembles of hippocampal neurons (9-12). Although several of these investigations applied nonlinear decoding methods including Bayesian algorithms and artificial neural networks, linear approaches were also successful in reconstructing behavioral contingencies from firing activity in small ensembles of simultaneously recorded neurons (1,3,4,7,8,11,13). Several studies have also investigated the reliability of detected neural representations on single trials (3,14,15), often in real time (4,5,7). Successful efforts to extract sensory, motor, and spatial information from neural activity support the conclusion that the identified neural representations are present on individual trials across sessions.