Open-source tool by Deci for analyzing health of AI training dataset

Machine learning startup Deci has unveiled an open-source tool called DataGradients, which allows data scientists to analyze the health of training datasets for AI models. The tool aims to address the challenges faced by AI developers in terms of hardware limitations and dataset quality. By profiling datasets before creating models, data scientists can gain insights into the capabilities and performance of their models. DataGradients is particularly useful in computer vision, where the quality of the training data directly influences model capabilities. The tool helps identify issues such as corrupted data, distributional shifts, and duplicate annotations, allowing users to make informed decisions and mitigate these problems. The open-source nature of DataGradients may help it gain popularity among developers, according to Constellation Research Inc.’s Andy Thurai. This release marks the third open-source tool launched by Deci, following the SuperGradients PyTorch training library and the YOLO-NAS object detection foundation model.