Informative Neural Codes to Separate Object Categories

摘要: Abstract In order to develop object recognition algorithms, which can approach human-level performance, researchers have been studying how the human brain performs in past five decades. This has already in-spired AI-based such as convolutional neural networks, are among most successful platforms today and performance specific tasks. However, it is not yet clearly known recorded activations convey information about category processing. One main obstacle lack of large feature sets, evaluate contents multiple aspects activations. Here, we compared a set 25 features, extracted from time series electroencephalography (EEG) participants doing an task. We could characterize informative categories. Among evaluated event-related potential (ERP) components N1 P2a were features with highest Theta frequency bands. Upon limiting analysis window, observed more for detecting temporally patterns signals. The results this study constrain previous theories codes information.