SBIR Phase I: Machine Vision for Content-based Video Marketing Analytics

这个小型企业创新研究（SBIR）第一阶段项目的更广泛的影响/商业潜力是保护消费者隐私，同时继续支持广告支持的互联网模式。目前基于跟踪的消费者定位方法本质上会侵蚀消费者隐私，在许多不同的网站上秘密跟踪用户，以收集人口统计和行为数据。另一方面，营销人员需要收集这些数据才能成功地接触到他们的受众，营销人员投入在线广告的收入已成为互联网经济的重要组成部分。如今，这种利弊竞争的微妙平衡受到广告拦截软件兴起的进一步威胁，这侵蚀了互联网广告投放的价值。该项目开发的视频营销分析功能将限制营销人员对侵入性消费者数据的需求，同时改善消费者体验。在商业领域，营销人员会珍惜机会，以最情绪化的方式定位他们的广告，最不具破坏性，最有吸引力的方式。这项技术将为营销人员提供通过算法观看数百万个视频的能力，从而实现比以往任何时候都更精简和定制的观众体验。这个小企业创新研究第一阶段项目旨在为感知注释开发商业应用，感知注释是一种由NSF资助开发的技术，可以将人类表现的详细测量注入机器学习过程，使机器学习者能够表现得更好，并以与人类更一致的方式表现。通过将这种新的人类来源的监督信号类别添加到机器学习过程中，提案者已经证明，有可能显着提高机器视觉性能，使机器能够更好地推广到新的，以前看不见的图像。虽然该公司的技术已经在大规模“野外”学术数据集上得到了严格的验证，但拟议的SBIR第一阶段活动的主要技术驱动力将是将公司的努力转移到分析“实时”，巨大和不断扩大的数据集，如在线视频。拟议的第一阶段工作的第二个主要驱动力将是构建“第二阶段”机器学习模型，该模型将基于感知注释的机器评级作为输入和输出可操作的营销决策。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is to protect consumer privacy while continuing to enable the ad-supported Internet model. Current tracking-based consumer targeting approaches inherently erode consumer privacy, surreptitiously tracking users across many different web sites in an effort to gather demographic and behavioral data. On the flip side of the coin, marketers need to collect such data to successfully reach their audiences, and the revenue that marketers pour into advertising online has become an essential component of the economics of the internet. Today, this delicate balance of competing pros and cons is further threatened by the rise of ad-blocking software, which erodes the value of internet ad placement. The video marketing analytics capability developed in this project will limit marketers' need for invasive consumer data, while improving consumer experience. In the commercial realm, marketers would value the opportunity to target their ads in the most emotionally consonant, least disruptive, and most engaging manner possible. This technology will provide marketers with the capability to watch millions of videos algorithmically, thus enabling a more streamlined and customized viewer experience than has ever before been possible on television or on the Internet. This Small Business Innovation Research Phase I project seeks to develop commercial applications for Perceptual Annotation, a technology developed with NSF funding that allows detailed measurements of human performance to be infused into a machine learning process, allowing the machine learner to both perform better and to perform in a way that is more consistent with humans. By adding this new category of human-derived supervisory signal into a machine learning process, the proposers have demonstrated that it is possible to significantly boost machine vision performance, allowing machines to generalize better to new, previously unseen images. While the company's technology has been rigorously validated on large-scale "in the wild" academic datasets, a major technical drive in the proposed SBIR Phase I activities will be to shift the company's efforts to the analysis of "live," enormous, and ever-expanding data sets such as online videos. A second major drive of the proposed Phase I work will be the construction of "second stage" machine learning models that take perceptual-annotation-based machine ratings as an input and output actionable marketing decisions.