In this letter, we consider the problem of signal detection in generalized spatial modulation (GSM) using deep neural networks (DNN). We propose a novel modularized DNN architecture that uses small sub-DNNs to detect the active antennas and complex modulation symbols, instead of using a single large DNN to jointly detect the active antennas and modulation symbols. The main idea is that using small sub-DNNs instead of a single large DNN reduces the required size of the NN and hence requires learning lesser number of parameters. Under the assumption of i.i.d Gaussian noise, the proposed DNN detector achieves a performance very close to that of the maximum likelihood detector. We also analyze the performance of the proposed detector under two practical conditions: i) correlated noise across receive antennas and ii) noise distribution deviating from the standard Gaussian model. The proposed DNN-based detector learns the deviations from the standard model and achieves superior performance compared to that of the conventional maximum likelihood detector.