A woman has been working on her book, a young adult fantasy novel, for hours. At some point, she gets the familiar itch to check her email: She can't think of what to write next. She could bang her head against the wall, or maybe turn to a favorite book for inspiration, or lose her momentum to distraction. But instead she turns to an AI writing tool, which takes in her chapter so far and spits out some potential next paragraphs. These paragraphs are never quite what she wants, though they sometimes contain beautiful sentences or fascinating directions.

Human parsing is an important task for human-centric understanding. Generally, two mainstreams are used to deal with this challenging and fundamental problem. The first one is employing extra human pose information to generate hierarchical parse graph to deal with human parsing task. Another one is training an end-to-end network with the semantic information in image level. In this paper, we develop an end-to-end progressive cognitive network to segment human parts. In order to establish a hierarchical relationship, a novel component-aware region convolution structure is proposed. With this structure, latter layers inherit prior component information from former layers and pay its attention to a finer component. In this way, we deal with human parsing as a progressive recognition task, that is, we first locate the whole human and then segment the hierarchical components gradually. The experiments indicate that our method has a better location capacity for the small objects and a better classification capacity for the large objects. Moreover, our framework can be embedded into any fully convolutional network to enhance the performance significantly.

There is a trend in machine learning right now towards automated machine learning pipelines. These automated machine learning pipes condense the model selection and optimization phase into a single step. They perform a random (or some sort of optimized) search over hyperparameters and models to pick the "best" set based on some evaluation metric. Basically all you have to do is plug in data. This automated process might appear to make things easier, but it can be a harmful practice.