Internal Models for Active Sensing in Freely Swimming Electric Fish

Animals move their sensory organs in order to acquire information about their external environment, a process known as Active Sensing. This necessitates an internal model predicting and compensating for the sensory consequences of motor behavior. Studies of the electrosensory lobe (ELL) of weakly electric fish have provided a detailed account of the mechanisms underlying such internal models; recordings in immobilized fish have shown that ELL, a cerebellar-like structure, uses corollary discharge (copies of the motor commands) and anti-Hebbian synaptic plasticity to generate a ‘negative image’ of these self-generated sensations. It is unknown, however, whether this mechanism holds for free natural exploration.

To test this, we developed methods for long-term neural recordings in freely swimming fish. This method allows us to simultaneously observe both the inputs and the outputs of ELL, as well as to track the animal’s behavior in high resolution. We show that the effects of multiple motor variables are indeed accurately canceled-out in the activity of ELL output cells. However, we demonstrate that motor-based negative image is insufficient to explain this cancellation since the sensory effects of motion are determined by the fish’s surrounding environment. We show that sensory feedback can be utilized to dynamically adapt the internal model to changing contexts, thus enabling precise and reliable perception of the external environment. Our findings suggest that neural feedback plays a crucial role in generalization of internal models in other sensory modalities, as well as in motor control.