Micro-Outsourcing for Mechanical CAD/CAM

最近，使用众包来交付点击已经证明了一种可行的方式，提供廉价的，强大的，基于内容的图像分析。这项提案寻求资金，以调查是否可以使用类似的方法来解决机械CAD/CAM中的几何推理问题。微外包，或众包，是一个新词，指的是将一项传统上由员工或承包商执行的任务，以公开召集的形式，将其外包给一个不确定的、通常是很大的群体。例如，公众可能会被邀请开发一项新技术，执行一项设计任务，改进一种算法，或帮助捕获、系统化或分析大量数据。人类智能任务(HIT)是一个人类认为简单，但计算机发现极其困难的问题。例如，与一张照片相关的点击可能是：这张照片中有一只狗吗？许多制造操作需要几何推理来对2D和3D形状中的各种图案或约束进行排序或识别。找到这些问题的最佳解决方案将提高许多行业的生产率，并直接影响它们的利润。然而，这些类型的问题通常是无法计算的(即NP完全的)，因此目前的做法是由CAD/CAM软件生成良好的而不是最优的解决方案。如果这种难题的众包方法被证明是有效的，它将展示制造业可以解决多少类似的模式识别和优化问题，并提供一个令人信服的示范，说明数字经济如何分配工作和数据。在其历史上的大部分时间里，CAD/CAM研究的动机都是希望通过算法来提高系统的智能性，这些算法可以计算出对人类来说很明显的形状属性(例如。薄片或孔的位置)。然而，这被证明是困难的，在取得进展的情况下，它通常解决了特殊案件(例如。2.5维几何图形)，而不是提供通用解决方案。经过几十年的学术努力，几何推理问题仍在研究议程上的例子不胜枚举，例如：路径规划、元件布局、工艺规划、局部对称检测和形状特征识别。最根本的困难是如何赋予计算机对人类毫不费力就能获得的物体整体形状的鉴赏能力。有趣的是，在图像和语音识别中也遇到了类似的困难，其中自动化系统仍然无法再现人类的表现水平。因为这种几何推理代表了一个主要的技术瓶颈，需要工程师手动完成许多相对琐碎的任务，这一过程可能既耗时又次优(例如。通常不可能穷尽地探索所有备选路径、顺序或计划)。因此，消除这一几何理解瓶颈将导致广泛行业的显着生产率提高。这项提议旨在调查分布式方法(俗称众包或微外包)的潜力，这种方法已经证明有能力为许多经典的人工智能问题提供实际解决方案，如图像和语音翻译。研究将使用两个范例应用程序来支持对研究问题的系统调查。第一个研究将集中在定义明确的任务上，结果很容易量化(部分嵌套)，而第二个研究将集中在一个容易陈述但很难量化的问题(形状相似)上。该项目将使用商业众包平台(即亚马逊的机械Turk)的API创建一个实验软件平台，以支持对这两种不同类型的命中系统性能的系统调查。

Recently the use of Crowdsourcing to deliver HITs has demonstrated a feasible way of providing cheap, robust, content based, Image analysis. This proposal seeks funding to investigate if a similar approach can be used to solve the geometric reasoning problems found in Mechanical CAD/CAM.Micro-outsourcing, or crowdsourcing, is a neologism for the act of taking a task traditionally performed by an employee or contractor, and outsourcing it to an undefined, generally large group of people, in the form of an open call. For example, the public may be invited to develop a new technology, carry out a design task, refine an algorithm or help capture, systematize or analyze large amounts of data. A Human Intelligence Task (HIT) is a problem that humans find simple, but computers find extremely difficult. For example a HIT related to a photograph could be: Is there a dog in this photograph? Many manufacturing operations require geometric reasoning to sequence, or recognize, various patterns, or constraints, in 2D and 3D shapes. Finding the best solutions to these problems would increase the productivity of numerous industries and impact directly on their profits. However frequently these types of problems are effectively incomputable (i.e. NP-complete) and so current practise is for CAD/CAM software to generate good , rather than optimum, solutions. If a Crowdsourcing approach to such difficult problems proves to be effective it would demonstrate how many similar pattern recognition and optimization problems manufacturing industry could be solved and provide a compelling demonstration of how a digital economy can distribute work, as well as, data. For much of its history CAD/CAM research has been motivated by the desire to increase the intelligence of systems by means of algorithms that could compute shape properties readily apparent to humans (eg. location of thin sections or holes). However this has proved to be difficult and where progress has been made it has generally solved special cases (eg. 2.5D geometry) rather than providing generic solutions. Examples of geometric reasoning problems still on the research agenda after decades of academic effort are numerous, for example: path planning, component packing, process planning, partial symmetry detection and shape feature recognition. Essential the difficulty is one of endowing computers with the appreciation of an object's overall form that humans gain so effortlessly. Interestingly similar difficulties have been encountered in image and speech recognition where automated systems still fail to reproduce human levels of performance.Because of this Geometric Reasoning represents a major technological bottleneck requiring many relatively trivial tasks to be done manually by engineers, a process that can be both time-consuming and sub-optimal (eg. frequently it will be infeasible to exhaustively explore all the alternatives paths, sequences or plans). Consequently removal of this geometric comprehension bottleneck would result in significant productivity gains across a wide range of industries. This proposal seeks to investigate the potential of a distributed approach (know colloquially as CrowdSourcing or Micro-outsourcing ) that has already proved its ability to provide practical solutions to many classic AI problems, such as image and speech interpretation. Research will use two exemplar applications to support a systematic investigation of the research issues. The first study will focus on a well defined task with easily quantifiable results (part nesting), while the second study will focus on a problem (shape similarity) easily stated but difficult to quantified.The project will create an experimental software platform, using the API of a commercial Crowdsourcing platform (i.e. Amazon's mechanical turk), to support the systematic investigation of the system's performance for these two different types of HIT.