Spike-timing-dependent plasticity (STDP) is ubiquitously observed at synapses of the mammalian brain, and is considered to be critical for both developmental and adulthood neural plasticity. Previous studies suggest that STDP enables neurons to capture the principal component of input stimulus patterns and to learn the major feature of stimuli. However, it is still unclear how neurons capture higher-order components, especially when these features are encoded by synchronization of neural activity, not by firing rate. For example, the brain may extract faint sounds from a mixture of non-independent auditory sources, but the underlying mechanism remains unknown. Furthermore, it is unknown how the structure of spike correlation and various types of noise influence the learning process. In addition, recent experimental studies suggest that inhibitory synapses also show spike-timing-dependent synaptic plasticity, although its functional role is still uncertain. Here, we constructed a computational circuit model with linear Poisson neuron model to analytically investigate how and when STDP can detect higher-order structures. We show that lateral inhibition and inhibitory STDP are crucial for improving the performance of learning process.