Duration: 0:00

48 views

0

Alex Brandsen, Iris Kramer, Wouter Verschoof-van der Vaart

For over two decades there have been sporadic presentations of diverse machine learning (ML) applications to digital archaeology at the CAA. In recent years there is a notable increase of papers using ML in archaeology, which may be ascribed to the success of Deep learning and Convolution Neural Networks (CNNs) across various disciplines that were previously described as being too complex for using machine learning. Applications using deep learning now show high performance on challenging tasks ranging from computer vision to natural language processing. In digital archaeology we have seen and foresee applications of these techniques including automated object detection in remote sensing data, artefact image classification, use-wear analysis, text mining, paleography, predictive modeling, 3D shape analysis and recognition, and typology development. Our aim for this session is to bring together the previously scattered ML research to discuss practical as well as theoretical approaches for ML in digital archaeology. For practical approaches we would encourage a critical dialogue to identify individual and shared problems, opportunities, and solutions. We invite authors to provide a thorough explanation on their approach and engage on some of the following questions: How do you structure archaeological datasets which are often small, incomplete, and noisy? What considerations applied to your choice of ML technique and how did was this technique tuned to your particular research? Which threshold do you find appropriate to determine the success of your method? What was your desired outcome and how did your final results compare to this? If your outcome resulted in a lot of new data that needs further manual validation, how do you plan to verify this? Do you foresee other applications for your method within archaeology or in other fields? Our request for theoretical approaches can be more broadly interpreted. Some examples include: creation of annotated benchmark datasets, sharing of developed methods, data (or data structure), and code, data science challenges, conventions for data structure and performance metrics, need for collaboration or special interest groups, insights from ML fields outside of archaeology, ethics of ML in archaeology, education of ML in archaeology, rapid publishing of new ideas, future gazing.