Many scenarios come to mind when we think about how existing training models for dynamic human situations lack consideration of scene adaptability. Consider the marauders who infiltrated the US Capitol building this week — how will investigators determine what went wrong and devise future protocols to make sure such a debacle never happens again? Clearly, many assumptions reduce the predictive accuracy of crowd behaviors, but data-driven methods do enhance the visual realism of crowd simulation. Trajectories of crowd movements and social attributes in real imagery can make a real difference. What if science takes the next step and incorporates crowd-driven image classification into artificial intelligence (AI)? Researchers would be able to quickly and accurately train algorithms.
Rapid advances in computing power, the availability of big data, and improvements in machine learning algorithms mean AI is changing the world as we know it. Computer vision, which entails AI technology to understand and label images, is used in activities as diverse as driverless car testing, medical diagnostics, and the monitoring of livestock or tree canopies. The Internet based cyber-physical world has profoundly changed the information environment for the development of AI, bringing a new wave of research. A new and salient characteristic of AI, crowd-driven intelligence, has attracted much attention from both industry and academic communities.
There is considerable human work involved in AI — tuning the algorithms, gathering the data, deciding what should be modeled in the first place, and using the outcomes of machine learning in the real world. As much research indicates, the accuracy of machine learning tasks critically depends on high quality ground truth data. Therefore, in many cases, producing good ground truth data typically involves trained professionals; however, this can be costly in time, effort, and money. Specifically, crowd-driven intelligence provides a novel problem-solving paradigm through gathering the intelligence of crowds to address challenges and has become increasingly popular to generate a large number of training data of good quality. Many computational tasks, such as image recognition and classification, are very trivial for human intelligence but pose grand challenges to current AI algorithms.
This week, the International Institute for Applied Systems Analysis (IIASA) announced the development of the new Picture Pile Platform, which aims to provide users with the opportunity to set up and run their own crowd-driven image classification campaigns. Those campaigns can quickly and accurately train AI algorithms.
While there are many image databases that can be used to train machine learning algorithms to perform computer vision tasks, there is a lack of datasets containing more specific features of interest, for example, crop or building types. The new Picture Pile Platform will address this by building upon the existing Picture Pile crowd-driven application that allows users to classify or help sort through piles of pictures.
These can be very high resolution satellite images, geo-tagged photographs, or any other images (e.g., images from medical applications) that require sorting. After a pile has been sorted, the image classifications can be made publicly available with FAIR (Findable, Accessible, Interoperable, and Reusable) metadata so that they can be freely used by anyone. The FAIR principles emphasize machine-actionability (i.e., the capacity of computational systems to find, access, interoperate, and reuse data with none or minimal human intervention) because humans increasingly rely on computational support to deal with data as a result of the increase in volume, complexity, and creation speed of data.
The Picture Pile Platform will provide quality control mechanisms to guarantee the accuracy of the data collected.
Picture Pile was initially developed as part of the pioneering research and development activities within the ERC Consolidator Grant, “CrowdLand: Harnessing the Power of Crowdsourcing to Improve Land Cover and Land-Use Information,” and has significantly contributed to the emerging field of citizen science. To date, there have been 34 Picture Pile campaigns, involving 10,130 people who have classified over 15 million images.
“We have often been approached by institutions asking if we could make a pile, in other words, specific image classifications, in Picture Pile,” explains IIASA Strategic Initiatives Program Director Steffen Fritz, who will lead the project. “The new platform will address the gap that currently exists in the market for a platform that allows users to build their own tailored, quality controlled crowd-driven campaigns to collect image classifications in an efficient, engaging, and fair way, and then possibly make the data collected openly and freely available. Once the platform has been built, the running costs will be low, and the overall benefit for society will be tremendous.”
Eventually, premium services will be added to make the platform commercially self-sustaining.
Mobile crowdsourcing is an extension of human computation from the virtual digital world to the physical world. The gamified version of the Picture Pile annotation tool is accessible as an online version as well as a mobile app in both IOS and Android (name: Picture Pile).
“If it is possible for everyone to easily, quickly, and freely run their own Picture Pile campaigns, and choose for the resulting data to be made openly and freely available to everyone, scientists and application developers from many different fields will be able to train AI models that can solve tasks faster, more reliably, and more cost effectively than humans. The opportunities for applying this innovation to a broad range of sectors promises far-reaching benefits to society and scientific research,” says IIASA researcher Tobias Sturn, lead developer on the Picture Pile Platform.
IIASA has collaborated with numerous institutions including the European Space Agency, the Earth Day Network, the Wilson Center, remote sensing companies, and universities to create piles with Picture Pile to train machines to detect degraded housing from satellite and ground photos, marine litter from aerial photos, and classifications of different crops in order to tackle food security issues. Currently IIASA and SAS are utilizing Picture Pile to power algorithms to detect deforestation in the Amazon rainforest.
This highly competitive ERC Proof of Concept grant is one of three awarded to Austria-based institutions in the latest annual rounds. IIASA researchers have been awarded a number of ERC grants over the last year to fund frontier research in the fields of equitable pension policies, climate change and population trends, and negative emissions technologies.