RESUMEN
Feedforward Convolutional Neural Networks (ffCNNs) have become state-of-the-art models both in computer vision and neuroscience. However, human-like performance of ffCNNs does not necessarily imply human-like computations. Previous studies have suggested that current ffCNNs do not make use of global shape information. However, it is currently unclear whether this reflects fundamental differences between ffCNN and human processing or is merely an artefact of how ffCNNs are trained. Here, we use visual crowding as a well-controlled, specific probe to test global shape computations. Our results provide evidence that ffCNNs cannot produce human-like global shape computations for principled architectural reasons. We lay out approaches that may address shortcomings of ffCNNs to provide better models of the human visual system.
Asunto(s)
Aglomeración , Percepción de Forma/fisiología , Procesamiento de Imagen Asistido por Computador , Redes Neurales de la Computación , Visión Ocular/fisiología , Simulación por Computador , HumanosRESUMEN
Recently, it was observed that protons in non-conducting solids doped with 1,3-bisdiphenylene-2-phenylallyl (BDPA) or its sulfonated derivative (SA-BDPA) can be polarized through Overhauser effects via resonant microwave irradiation. These effects were present under magic angle spinning conditions in magnetic fields between 5 and 18.8â¯T and at temperatures near 100â¯K. This communication reports similar effects in static samples at 6.7â¯T and, more importantly, at temperatures as low as 1.2â¯K, in a different dynamic regime than in the previous study. Our results provide new information towards understanding the mechanism of the Overhauser effect in non-conducting solids. We discuss possible origins of the fluctuations that can give rise to an Overhauser effect at such low temperatures.
RESUMEN
Hyperpolarization of metabolites by dissolution dynamic nuclear polarization (D-DNP) for MRI applications often requires fast and efficient removal of the radicals (polarizing agents). Ordered mesoporous SBA-15 silica materials containing homogeneously dispersed radicals, referred to as HYperPolarizing SOlids (HYPSOs), enable high polarization - P(1H) = 50% at 1.2 K - and straightforward separation of the polarizing HYPSO material from the hyperpolarized solution by filtration. However, the one-dimensional tubular pores of SBA-15 type materials are not ideal for nuclear spin diffusion, which may limit efficient polarization. Here, we develop a generation of hyperpolarizing solids based on a SBA-16 structure with a network of pores interconnected in three dimensions, which allows a significant increase of polarization, i.e. P(1H) = 63% at 1.2 K. This result illustrates how one can improve materials by combining a control of the incorporation of radicals with a better design of the porous network structures.