Healthy subjects can learn the Pavlovian association of a tone conditioned stimulus (CS) that predicts the presentation of a puff of air to the eye (i.e. unconditioned stimulus, US). Eventually, the CS will elicit a conditioned eyeblink response after the tone onset and ahead of the airpuff onset. This ability may be lost due to aging or cerebellar traumatic brain injury. In order to reconstitute this behavioral function, a real-time analysis and control system was developed. It consists of a recording system (including headstage pre-amplifiers and a main biosignal amplifier with analog-to-digital conversion), a signal processing system, a model of the cerebellar functionality (Verschure & Mintz 2001) and a stimulation device for the excitation of the final motor pathway of the eyeblink response.

In this approach CS and US signals in the form of multiple unit activity are acquired from the pontine nucleus (PN) and the inferior olive (IO), respectively. To allow a correct identification of stimulus onsets in the online system, training data recorded during conditioning is used offline to calculate parameters for the detection algorithms. Concurrently, parameters for the cerebellar model are calculated. The online system reacquires biosignals, detects the onsets of the stimulation, and feeds these time-points into the cerebellar model which uses them to simulate cerebellar plasticity and learns the required behavior to avoid the airpuff. Once the model has learned, its output is used to control an electrical stimulator implanted in the facial nucleus of the animal. This output elicits an eye-blink shortly before the airpuff is applied.

Here the key technologies and methods for the correct operation of the real-time part of the system will be discussed. It is vital to ensure an optimal cooperation between the operators and the recording system in order to provide good data quality, correct CS and US detection with a tolerable rate of false positives and missed events, and correct learning speed of the cerebellar model. This is achieved by providing real-time audiovisual inspection of the various signals acquired and produced by the system. Experimental results from test runs performed on anaesthetized rats are shown, proving the correct functionality of the presented integrated approach.

• P. F. M. J. Verschure, M. Mintz: A real-time model of the cerebellar circuitry underlying classical conditioning: A combined simulation and robotics study. Neurocomputing 38-40: 1019-1024 (2001) Classical Conditioning. Advances in Neural Information Processing Systems (Vol. 17, pp. 577-584).