Machine learning to understand images and text

机器学习研究的目的是建立从数据中提取有用信息的计算机系统。在学习的两个主要领域已经取得了相当大的进展。监督方法，它依赖于每个输入的目标标签，已被证明是非常强大和广泛适用的，特别是在分类问题，其中每个输入的例子被分配到一个小集合的类之一。这些方法通常随着训练样本和类的数量而伸缩性差。无监督方法利用未标记的数据，其中没有提供目标，并试图构建对许多输入输出映射有用的表示。这些方法很容易扩展，但不是强大的，因为没有标签相当大的结构必须施加使algorithms.This建议侧重于开发的学习方法，利用标记和unlabeledexamples，解决两个基本问题。第一个是提取结构化对象的信息摘要，其中包含几个相互依赖的组件。两种非常常见的结构化对象是图像和文档。这个领域的一个示例任务是场景分析，它需要在视觉输入中挑选出多个项目。例如，一个忙碌街道的视频被数字记录为彩色像素地图流，可以根据汽车和行人及其运动来重新表示。这是一个基本的和困难的问题，涉及到低层次的图像属性与高层次的，对象特定的知识相结合。人类从少数几个例子中获得视觉概念的能力，以及在复杂场景中识别实例的能力，对计算机视觉，认知科学和机器学习领域提出了核心挑战。提高机器快速获取新概念的能力不仅可以应用于视觉，使其能够分析包含相对新颖物体的场景，而且可以应用于在线交互领域，因为快速学习系统可以通过几个问题或对用户的网络浏览历史或产品偏好的有限了解来建立用户的准确模型。

Machine learning research aims to build computer systems that extract useful information from data.Considerable progress has been made in the two main areas of learning. Supervised methods, which relyon having target labels for each input, have proved to be very powerful and widely applicable,particularly on classification problems, in which each input example is assigned to one of a small setof classes. These methods generally scale poorly with the number of training examples and classes.Unsupervised methods utilize unlabeled data, where no targets are provided, and attempt to constructrepresentations useful for many input-output mappings. These methods scale readily but are not aspowerful, because without labels considerable structure must be imposed to make the algorithms work.This proposal focuses on the development of learning methods that utilize both labeled and unlabeledexamples, addressing two fundamental problems. The first is extracting informative summaries ofstructured objects, which contain several inter-dependent components. Two very common structured objectsare images and documents. A sample task in this domain is scene analysis, which entails picking outmultiple items in visual input. For example, a video of a busy street recorded digitally as a stream ofcolor pixel-maps may be re-represented in terms of the cars and pedestrians and their motions. This isan essential and difficult problem, involving a combination of low-level image properties withhigh-level, object-specific knowledge.The second problem is rapid learning. The human ability to acquire a visual concept from a fewexamples, and to recognize instances in a complex scene, poses a central challenge to the fields ofcomputer vision, cognitive science, and machine learning. Improving machines' ability to acquire newconcepts readily has applications not only to vision, enabling analysis of scenes containing relativelynovel objects, but also to the domain of online interaction, as a rapid learning system can build anaccurate model of a user with just a few questions, or limited knowledge of the person's web-surfinghistory or product preferences.