Beyond Text: Adaptive Data for the Multimodal Era – Adaption (adaptionlabs.ai)

Adaption has announced the launch of Adaptive Data for vision, extending their innovative platform beyond text and documents to manage and enhance image data. This significant development addresses the limitations of static datasets in AI systems, particularly in the realm of vision, where the images encountered in production often differ from training data. Recognizing that a vision model's effectiveness is heavily reliant on its underlying data, Adaptive Data aims to tackle challenges encountered when external factors, such as new medical imaging technologies or varied document layouts, disrupt model performance. The Adaptive Data platform allows teams to seamlessly manage visual tasks like visual question answering, image captioning, and classification, while supporting multiple data formats. It claims to enhance data quality by an average of 82% across 242 languages for text, and aims to replicate this success with images. The early results show that this adaptive approach leads to an impressive average win rate of 67% across five vision datasets. As Adaption plans to expand into other modalities, teams utilizing this platform can expect not just improved workflows but also transformative capabilities for their systems in vision and beyond.