This week, the top international AI will NeurIPS 2019 grand opening in Vancouver, Canada. On this NeurIPS, Alipay articles selected papers show results, and held a workshop exchange of experts gathered to share.
In NeurIPS 2019 booth, Alipay exhibited garbage recycling bins based on intelligent video image recognition technology and intelligent classification applet. Viewers can easily identify pictures of garbage trash items and the corresponding category by phone, on-site experience from the East "Life Little Helper."
It is understood that a small program for site experience for Alipay launched in July, behind based on image recognition, environmental protection and other technical knowledge map. On line two months of total users has exceeded 4 million, can identify more than 11,000 kinds of junk items. After that, Alipay and upgraded technologies, the development of intelligent recycling bins using the AI + IoT technology that allows automatic separation of bottles for all kinds of different materials, classify, improve the recovery rate at the same time, solve the problem of recyclable waste.
Why is the birth of this little program it?
This is from the status quo garbage literacy Chinese Speaking: 73% of the garbage in landfills treatment, only 15% recycling in China; and the same number in the world's average level, the proportion of recyclable waste recycling is up to 38%. This is actually a waste of resources, caused by the huge differences in key reason is that in most parts of China has not the implementation of waste policy, and public awareness of waste separation is also not as prevalent as in developed countries: in daily life we could face as many as 2,000 kinds of junk items, 90% of people will throw wrong. Therefore, the correct classification is the key to treatment of recyclable garbage. The traditional garbage, we need a sound social input of expert knowledge, but also need to invest huge social promotion and popularization of cost. Based on this, Alipay engineers born to help people solve the garbage problem with the idea of artificial intelligence techniques.
How smart garbage achieve?
基于智能分类小程序，即使毫无垃圾分类经验的老人和儿童也能在日常投递垃圾过程中正确分类。看似简单的操作流程，其背后是一整套完善的人工智能技术在做支撑。其中最核心的图像分类模型来自于蚂蚁金服自建的金融视觉平台。蚂蚁金融视觉平台提供了一站式建模能力，从而支撑图像模型开发与应用过程所需的数据工程、算法选型、模型调参、技术指标收敛、工程部署、线上效果迭代环节，真正实现从数据到模型“端到端”的全套建模能力。
此外，由于垃圾物品性状的特殊性，基于图像的分类准确率提升难点之一就是如何让机器能够快速地“学习”这些新的知识。一方面我们建立了环保知识图谱，从而可以针对各类不同材质的物品进行自动分离、归类，更好地适应对于某些非常见物品和变形物品分类的诉求；另一方面，通过AI数据中台，打通模型训练与图像标注环节，建立了一个完善且高效的“模型冷启动—>模型上线服务—>真实数据回流—>模型迭代”链路。得益于这个模型自迭代链路，即使短时间内迭代几十个版本算法模型也毫无压力，使得识别类目数量和识别准确率能够不断提升。
目前，这套图像识别技术已逐步开放给行业伙伴。利用支付宝的海量用户覆盖能力和城市服务的便民心智，结合AI+IoT技术，实现人工智能识别垃圾和分类，通过小程序及智能终端设备的移动互联网解决方案，可链接用户与回收人员，如实现一键预约、上门回收，并可以广泛应用于旧物回收、租赁等回收处理环节。用科技推动个人与企业单位参与垃圾分类回收的积极性和处理效能。
