Deep networks are increasingly being applied to problems involving image synthesis, e.g., generating images from textual descriptions and reconstructing an input image from a compact representation. Supervised training of image-synthesis networks typically uses a pixel-wise loss (PL) to indicate the mismatch between a generated image and its corresponding target image. We propose instead to use a loss function that is better calibrated to human perceptual judgments of image quality: the multiscale structural-similarity score (MS-SSIM) [34]. Because MS-SSIM is differentiable, it is easily incorporated into gradient-descent learning. We compare the consequences of using MS-SSIM versus PL loss on training deterministic and stochastic autoencoders. For three different architectures, we collected human judgments of the quality of image reconstructions. Observers reliably prefer images synthesized by MS-SSIM-optimized models over those synthesized by PL-optimized models, for two distinct PL measures (L 1 and L 2 distances). We also explore the effect of training objective on image encoding and analyze conditions under which perceptually-optimized representations yield better performance on image classification. Finally, we demonstrate the superiority of perceptually-optimized networks for super-resolution imaging. Just as computer vision has advanced through the use of convolutional architectures that mimic the structure of the mammalian visual system, we argue that significant additional advances can be made in modeling images through the use of training objectives that are well aligned to characteristics of human perception.

One Sentence Summary: The meaning of concepts resides in relationships across encompassing systems that each provide a window on a shared reality. AbstractConcept induction requires the extraction and naming of concepts from noisy perceptual experience. For supervised approaches, as the number of concepts grows, so does the number of required training examples. Philosophers, psychologists, and computer scientists, have long recognized that children can learn to label objects without being explicitly taught. In a series of computational experiments, we highlight how information in the environment can be used to build and align conceptual systems. Unlike supervised learning, the learning problem becomes easier the more concepts and systems there are to master. The key insight is that each concept has a unique signature within one conceptual system (e.g., images) that is recapitulated in other systems (e.g., text or audio). As predicted, children's early concepts form readily aligned systems.A typical person can correctly recognize and name thousands of objects. By 24 months, children already exhibit an average vocabulary of 200-300 words (Fenson et al., 1994). However, it remains unclear what mechanism makes this feat possible. Here, we conduct an information analysis and demonstrate that it is theoretically possible to learn the labels for objects through purely unsupervised means. Our key insight is that objects embedded within a conceptual system (e.g., text, audio, images) have a unique signature that allows for entire conceptual systems to be aligned (e.g., images with text) absent any instruction. 1

No abstract

No abstract