NSF Award
9260156
Low cost workstations and increasingly powerful PCs now make it feasible for a computer to record experimental data and make high-level analyses of the results for many classes of neuroscience experiments; the computer should further be able to determine autonomously when statistically valid results have been accumulated, recognize and repeat defective or failed measurements, appropriately alter experiment parameters and protocol, and even recognize unexpected opportunities and adjust protocol to take advantage of them. While the best of current generation of experiment control software supports on-line analysis and decision making, it requires sophisticated programming skills to implement these capabilities, and the experiment must pause to analyze results and evaluate control logic. This causes a loss of experiment time. We propose research on a Real-Time Expert System Experiment Controller that will: 1) capture data from a laboratory's present control computer; 2) perform sophisticated analyses on incoming results; 3) make experiment control decision of the sort presently made by the experimenter (basing them on the experimenter's priorities, reasoning methods, statistical criteria and rules of thumb, which are encoded in the expert system's knowledge base as a hierarchical set of interdependent hypotheses; 4) either directly command the present control computer to implement its decisions, allowing the experimenter to manipulate priorities and high level control variables to influence the way the controller allocates effort; 5) or convey recommendations to the experimenter so he may make the ultimate decisions. Such a system should improve experiment flow decision by increasing the sophistication of analysis on which they are based. The proposed research will evaluate the Controller's performance, suggesting design improvements and validating the concept of expert system experiment control. Success will lead to improved productivity and cost-effectiveness of neuroscience research that is amenable to computer control.
