Lezing | Seminar
AI & Data Science @ Archeologie
- vrijdag 31 januari 2020
- Followed by drinks and snacks
- Van Steenis
2333 CC Leiden
- Central hall
Het Data Science Onderzoeksprogramma, SAILS en de Faculteit Archeologie hebben hun krachten gebundeld om een uniek evenement te organiseren: een seminar op het snijvlak van Data Science, AI en Archeologie. Kom luisteren naar lezingen over hoe Deep Learning, Imaging, Text Mining en ander computergebaseerd onderzoek Archeologie vooruit kan helpen. Het seminar is vrij toegankelijk voor iedereen. Voertaal Engels. Meer informatie op de Engelstalige pagina.
Graag registreren via this link.
Wouter Verschoof - van der Vaart
"The use of R-CNNs in the automated detection of archaeological objects in LiDAR data"
A common practice in present-day archaeology and heritage management is to manually analyze remotely sensed data for the occurrence of archaeological (landscape) objects. However, the amount of high quality, (freely) available remotely sensed data is growing at an overwhelming rate, which results in new challenges to effectively and efficiently analyze these data by hand. To cope with this problem, computer-aided methods for the (semi-) automatic detection of archaeological objects are needed.
This research project explores the implementation of Deep Learning R-CNNs (Region-based Convolutional Neural Networks) in order to develop a generic, flexible, and robust automated detection method for archaeological objects in LiDAR data.
In this paper a workflow (called WODAN 2.0) is presented. This has been trained and tested on LiDAR data gathered from a forested area in the central part of the Netherlands, the Veluwe. This area contains a multitude of archaeological objects, including (Prehistoric) barrows, Celtic fields, and (Medieval) charcoal kilns. By implementing this improved workflow we have been able to develop a method to automatically detect and categorize these archaeological objects. We will present the results of experiments done with the workflow on a newly developed reference dataset. These results will be compared to the performance of a prior workflow and the results of a large-scale citizen science project, called Heritage Quest. In this project, citizen researchers were tasked with marking barrows, Celtic fields, and charcoal kilns in LiDAR images from the Veluwe, comparable to the task of WODAN 2.0.
"Digging in Documents - Creating a Search Engine for Excavation Reports using Text Mining"
Over 70.000 Dutch fieldwork reports are available online, and this number is growing rapidly. The information in this grey literature can be of immense value, but is underused at the moment. Currently it is only possible to search through the metadata of these documents. However, these metadata are often limited and sometimes inconsistent, and don't capture the ‘by-catch opportunity’; i.e. a single Bronze Age find within a large Medieval excavation. In this talk, we will present our results using Text Mining to create an intelligent search engine (AGNES), that allows search on the full text and specific concepts (e.g. artefacts, time periods) found in the text.
"Estimating Sex from Calcaneus Measurements: A Machine Learning Approach"
The estimation of sex is considered to be one of the fundamental steps in the identification of human skeletal remains in archaeological, forensic, and disaster victim identification (DVI) settings and typically involves scoring various features of the pelvis and the cranium. Though this method has proven to be highly accurate, in practice it fails to take into account aspects such as fragmentation due to post-depositional processes and taphonomy. The calcaneus (heel bone) is the largest bone in the foot and is often recovered fairly intact, which is mainly due to it being a weight bearing bone and its protection from taphonomic changes with the use of footwear.
Based on seven calcaneal measurements of both the left and right side of 96 adult individuals from the 19th century osteoarchaeological Middenbeemster collection from the Netherlands, cross-validated machine learning algorithms for supervised learning using binomial logistic regression were developed. In total, three multiple variable models were computed with accuracy rates ranging from 80.95% to 83.33%. Single variable models for each of the variables separately were computed as well, with nine out of fourteen variables yielding an accuracy rate of 70% or more, and four of these nine variables scored 75%.
Though this study is the first to develop a calcaneus-based metric method for estimating sex in human skeletal remains in a Dutch population using a machine learning approach, the results of this research are in fact extremely promising. Introducing such a valid adjunctive method for sex estimation in Dutch individuals could be vastly beneficial in the analysis of human skeletal remains in archaeological, forensic, and DVI settings.
“Improving Neanderthals with Machine Learning”
Neanderthals were robust Late Pleistocene hominins not very distinct from ourselves living in Eurasia for considerable time periods until their sudden disappearance from the archaeological record around 40 kya. The presented research has identified many unexpected lacunae in our knowledge of Neanderthals which lead to the construction of a multi-parameterized Agent-Based Model of Neanderthal dispersal in north-western Europe. Results of simulations with this model through time were compared to the archaeological record detailing the absence and presence of these enigmatic hominins in the research area.
After identifying the need for an unbiased parameterization of the model a Genetic Algorithm was implemented to explore the vast parameter space looking for those Neanderthal agents that produced the best match with the archaeology. This modelling and simulation system was aptly named HomininSpace. Starting with a randomly generated set of Neanderthals, the tool automatically improves promising individuals by modifying parameter values according to rules inspired by Nature. Depending on the research question that is addressed the resulting best Neanderthal species does not always conform to expectations and results can be used to explore ideas about Neanderthal behavior in our deep past.