Humboldt-Universität zu Berlin - Collaborative Research Center for Theoretical Biology

In a few computational steps from sensory representation to feature extraction and object recognition

What are the computational algorithms that represent key stepping stones for cognitive functions of small brains? In crickets, acoustic communication signals carry essential information on short and long time scales, somewhat in coarse analogy to the words and phrases of human speech. Nevertheless, salient features of a signal are extracted for robust object representation in only few processing steps that are accessible to physiological methods. At present the computational algorithms of feature extraction on these two time scales are unsolved. Beyond these questions, the rules by which the two time scales are integrated into an object that serves as the basis for a decision process are unanswered.

The goal of the present project is to identify the general algorithm for feature extraction in crickets on the two - short and long - time scales and to determine its neuronal basis. The dependence of the processing algorithms on phase, time and filter tuning as predicted by a mathematical approach will be examined by behavioural choice tests. Involved neurons will be probed by both intracellular recordings as well as extracellular multi-electrode techniques at the known appropriate sites within the brain. In tight linkage with experimental data sets basic models of neurons and small networks will be employed to probe hypotheses and, conversely, statistic model predictions will be tested in redesigned experiments. By combination of these theoretical and experimental approaches we aim to elucidate general computational algorithms underlying acoustic object recognition.

description of the 2nd period german version