Goto

Collaborating Authors

Meet the world's first female AI news anchor

#artificialintelligence

In November, Xinhua and Sogou unveiled the first AI news anchors of any gender, a pair of male AIs trained to deliver the news in either English or Chinese. The same day they unveiled Xin, Xinhua and Sogou announced that they'd given these AI anchors the ability to stand and talk simultaneously, and they'll show off that new ability while covering the Two Sessions alongside their female counterpart. When Xinhua debuted their first AI anchors in November, the news agency claimed that each anchor could "work 24 hours a day on its official website and various social media platforms, reducing news production costs and improving efficiency." Since then, the anchors have delivered 3,400 news reports while logging 10,000 minutes of screen time, according to Tencent News.


MetaAnchor: Learning to Detect Objects with Customized Anchors

Neural Information Processing Systems

We propose a novel and flexible anchor mechanism named MetaAnchor for object detection frameworks. Unlike many previous detectors model anchors via a predefined manner, in MetaAnchor anchor functions could be dynamically generated from the arbitrary customized prior boxes. Taking advantage of weight prediction, MetaAnchor is able to work with most of the anchor-based object detection systems such as RetinaNet. Compared with the predefined anchor scheme, we empirically find that MetaAnchor is more robust to anchor settings and bounding box distributions; in addition, it also shows the potential on the transfer task. Our experiment on COCO detection task shows MetaAnchor consistently outperforms the counterparts in various scenarios.


MetaAnchor: Learning to Detect Objects with Customized Anchors

Neural Information Processing Systems

We propose a novel and flexible anchor mechanism named MetaAnchor for object detection frameworks. Unlike many previous detectors model anchors via a predefined manner, in MetaAnchor anchor functions could be dynamically generated from the arbitrary customized prior boxes. Taking advantage of weight prediction, MetaAnchor is able to work with most of the anchor-based object detection systems such as RetinaNet. Compared with the predefined anchor scheme, we empirically find that MetaAnchor is more robust to anchor settings and bounding box distributions; in addition, it also shows the potential on transfer tasks. Our experiment on COCO detection task shows that MetaAnchor consistently outperforms the counterparts in various scenarios.


Watch China's new AI anchor read the news

#artificialintelligence

Xinhua is boasting that their new artificial intelligence news anchor is a world's first and that "he" is now considered a regular member of the reporting team and, even better, never needs a break. Since the AI anchor can work 24 hours a day, Xinhua says that means production costs associated with human anchors can be reduced and efficiency improved. Xinhua also says that the virtual anchor can self-learn from watching live broadcasting videos and "can read texts as naturally as a professional news anchor." Take a look at the video below and you'll see his voice sounds highly synthesized. Still, the achievement is fascinating, creepy, and horrifying at the same time.


Learning Anchor Planes for Classification

Neural Information Processing Systems

Local Coordinate Coding (LCC) [18] is a method for modeling functions of data lying on non-linear manifolds. It provides a set of anchor points which form a local coordinate system, such that each data point on the manifold can be approximated by a linear combination of its anchor points, and the linear weights become the local coordinate coding. In this paper we propose encoding data using orthogonal anchor planes, rather than anchor points. Our method needs only a few orthogonal anchor planes for coding, and it can linearize any (\alpha,\beta,p)-Lipschitz smooth nonlinear function with a fixed expected value of the upper-bound approximation error on any high dimensional data. In practice, the orthogonal coordinate system can be easily learned by minimizing this upper bound using singular value decomposition (SVD).