Workshop at the Bibliotheca Hertziana BHMPI

The irruption of Machine Learning has changed the stakes in regards to architectural and urban research. Both the quantitative lens and the content agnostic flexibility these methods provide has caused a shift in attention towards the possibility of circumventing direct formulas by a supercharged probabilistic approach to urban phenomena.

Currently, we witness the rapid development and implementation of disruptive paradigms like IoT, and Smart City, which need large inputs of data coming from a new and pervasive form of data gathering: sensing. Either based on camera feeds from self-driving cars or drones, point cloud scans or digitized records, these forms of data constitute important keystones in the deployment of a new machine cognition of the city and will shape our present and future cities in pervasive ways. What does that mean to other types of information about heritage, experience or valuation? The image of the architectural and urban environment these large quantitative datasets bear is limited, not only due to the type of photographic images they contain, the framing they inherit, and the scenery they depict, but also by the limited ideas about which relevant objects should be segmented and annotated, and which categories are applied. The lack of incorporation of decanted epistemologies from the humanities and certain social sciences in the data, either explicitly modeled or implicit in the design of the datasets used, constitutes not only a challenge but a veritable urgency.

At the same time, there is a wealth of material in the collections on the history of art and architecture in many relevant research Institutions, which, like in the case of the Max Planck Institute for the History of Art – Bibliotheca Hertziana, are in the process of being digitized. This development, happening in parallel to the constant development of annotation and segmentation tools, ought to help us address the manufacturing of relevant datasets for architectural and urban research that capture relevant information and data, which are, for the time being, alarmingly scarce. Again, while certain attempts have been made in the form of architectural historical datasets, these are insufficient due largely to the different pace of development between technological innovation and the methods of the humanities.
Equally important is the realization that the city as an object and subject of study that has a history and a projection, and which guarantees a continuity and a heterogeneity which can help ground some of the problems in a broader context, and therefore facilitate an interdisciplinary reflection.

How can we leverage computational methods to address this continuum? And How can we devise and cross-fertilize comprehensive and innovative research methodologies that consider the city as an object of a longue durée without sacrificing the capabilities of both researchers and machines to attune to the significant fine grain specificities of different periods and focus of study? Can the city as an object of study be conducive for establishing a much needed interdisciplinary dialogue?

Participants: Frederick Chando Kim, doctoral researcher EPFL, Johanes Mikhael EPFL; Lucía Jalón Oyarzun postdoctoral researcher DVS; Shin Koseki, UNESCO Chair Professor in Urban Landscape, Université de Montréal; Myriam Guillemette, researcher Université de Montréal; Miryam Parent, Université de Montréal; Alejandro Cantera, researcher INSPIDE-World Bank, Valentine Bernasconi, doctoral researcher DVS; Eva Cetinić postdoctoral researcher DVS, and Jason Armitage, doctoral researcher DVS. The event will be closed to the public.