2014/11/14 by M. Zhao, Mingmin Zhao, Zhao, Mingmin +6
Computer Science · Mathematics · Neuroscience · #Artificial intelligence #Coding (social sciences) #Computer Vision and Pattern Recognition (cs.CV) #Computer science #Context (archaeology) #ENCODE #Encoding (memory) #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Image and Signal Denoising Methods #Inference #Interpretability #Machine Learning (cs.LG) #Machine learning #Mathematics #Neural and Evolutionary Computing (cs.NE) #Neural dynamics and brain function #Pattern recognition (psychology) #Perception #Predictability #Representation (politics) #Visual Attention and Saliency Detection #cs.CV #cs.LG #cs.NE
paper · pdf · doi:10.48550/arxiv.1411.3815
openalex publication_date 2014/11/14 · arxiv created 2015/04/16 · arxiv updated 2015/04/17 · openalex created_date 2025/10/24 · openalex updated_date 2026/07/28
We propose a new neurally-inspired model that can learn to encode the global\nrelationship context of visual events across time and space and to use the\ncontextual information to modulate the analysis by synthesis process in a\npredictive coding framework. The model learns latent contextual representations\nby maximizing the predictability of visual events based on local and global\ncontextual information through both top-down and bottom-up processes. In\ncontrast to standard predictive coding models, the prediction error in this\nmodel is used to update the contextual representation but does not alter the\nfeedforward input for the next layer, and is thus more consistent with\nneurophysiological observations. We establish the computational feasibility of\nthis model by demonstrating its ability in several aspects. We show that our\nmodel can outperform state-of-art performances of gated Boltzmann machines\n(GBM) in estimation of contextual information. Our model can also interpolate\nmissing events or predict future events in image sequences while simultaneously\nestimating contextual information. We show it achieves state-of-art\nperformances in terms of prediction accuracy in a variety of tasks and\npossesses the ability to interpolate missing frames, a function that is lacking\nin GBM.\n