Emergent Neural Network Mechanisms for Generalization to Objects in Novel Orientations
This addresses the problem of understanding generalization in DNNs for computer vision researchers, but it is incremental as it builds on existing knowledge of neural mechanisms.
The study investigated how Deep Neural Networks (DNNs) generalize to objects in novel orientations, finding that they achieve this by disseminating orientation-invariance from familiar objects, with capability strengthening when trained with more objects but limited to 2D rotations.
The capability of Deep Neural Networks (DNNs) to recognize objects in orientations outside the distribution of the training data is not well understood. We present evidence that DNNs are capable of generalizing to objects in novel orientations by disseminating orientation-invariance obtained from familiar objects seen from many viewpoints. This capability strengthens when training the DNN with an increasing number of familiar objects, but only in orientations that involve 2D rotations of familiar orientations. We show that this dissemination is achieved via neurons tuned to common features between familiar and unfamiliar objects. These results implicate brain-like neural mechanisms for generalization.