2021 Call for Code Awards: Live from New York, with SNL’s Colin Jost! Learn more

Overcoming the Three Major AI Challenges in Computer Vision Projects

The past decade has witnessed a constant growth of IT projects engaging computer vision. Although IBM’s AI implementations provide a great potential to increase process efficiency, most AI projects are facing three major challenges. These challenges can prevent the expected results from being achieved. The first challenge is caused by a lack of data due to missing IT infrastructure. Secondly, most provided data is of poor quality. For example, the resolution of scanned documents may be less than 100dpi caused by missing standards. Thirdly, many employees have reservations about AI which diminish the adaption rate of this technology.

There is limited understanding of how to overcome these AI challenges in the data science field. In this short talk, we present a machine learning (ML) architecture based on Openshift in the context of fraud analysis of documents. This ML architecture was developed and applied for a public sector client. It shows great performance for small amounts of data with poor quality. In addition, it provides a high level of explainability to build up employee’s trust. By utilizing this ML architecture, other Data Scientists and IT Architects are enabled to overcome the above-mentioned challenges to achieve the expected outcomes from AI applications.