We propose a novel deep learning model for joint document-level entity disambiguation, which leverages learned neural representations. Key components are entity embeddings, a neural attention mechanism over local context windows, and a differentiable joint inference stage for disambiguation. Our approach thereby combines benefits of deep learning with more traditional approaches such as graphic...

Recently, encoder-decoder neural networks have shown impressive performance on many sequence-related tasks. The architecture commonly uses an attentional mechanism which allows the model to learn alignments between the source and the target sequence. Most attentional mechanisms used today is based on a global attention property which requires a computation of a weighted summarization of the who...

Human communication emerges from cortical processing, known to be implemented on a regular repetitive neuronal substratum. The supposed genericity of cortical processing has elicited a series of modeling works in computational neuroscience that underline the information flows driven by the cortical circuitry. In the minimalist framework underlying the current theories for the embodiment of cogn...

Data from brain-damaged and neurologically intact populations indicate hemispheric asymmetries in the temporo-parietal cortex for discriminating an object's global form (e.g. the overall shape of a bicycle) versus its local parts (e.g. the spokes in a bicycle tire). However, it is not yet clear whether such asymmetries reflect processes that (i) bias attention toward upcoming global versus loca...

We investigated neural substrates of global/local processing of bilateral hierarchical stimuli using functional magnetic resonance imaging (fMRI). Subjects were presented with two compound letters that were displayed simultaneously in the left and right visual fields, respectively. In a steady-state, block-design paradigm, hemodynamic responses were recorded while subjects detected infrequent g...

Spatial frequency (SF) selection has long been recognized to play a role in global and local processing, though the nature of the relationship between SF processing and global/local perception is debated. Previous studies have shown that attention to relatively lower SFs facilitates global perception, and that attention to relatively higher SFs facilitates local perception. Here we recorded eve...

OBJECTIVE Examine the neural mechanisms of global/local processing of multiple hierarchical stimuli. METHODS Event-related brain potentials (ERPs) were recorded from adults who selectively attended to the global or local level of two compound letters that were simultaneously presented in the left and right visual fields, respectively. The compound stimuli were either broadband in spatial freq...

The neural mechanisms of hierarchical stimulus processing were investigated using a combined event-related potentials (ERPs) and positron emission tomography (PET) approach. Healthy subjects were tested under two conditions that involved selective or divided attention between local and global levels of hierarchical letter stimuli in order to determine whether and where hemispheric differences m...

The trade-off between feature representation power and spatial localization accuracy is crucial for the dense classification/semantic segmentation of aerial images. High-level features extracted from late layers a neural network are rich in semantic information, yet have blurred details; low-level early contain more pixel-level but isolated noisy. It therefore difficult to bridge gap high due t...