Laura Narvaez Zertuche is an Associate Partner at Foster + Partners. She is a registered Architect and MSc in Urban Design and Sustainable Development from Tec de Monterrey; she also holds a MSc and PhD in Architecture and the Built Environment from Space Syntax Laboratory, Bartlett School of Architecture UCL.

The journey of Data Science has evolved into different analytical approaches with the aim  to answer different questions about the kind of data being created. Data comes in many  ways: sensors used to gather climate information, social media sites, purchase transaction  records, mobile phone GPS signals, or customer marketing trends to name a few. In spatial science, I often describe that if mobility and the way people move in cities was the lifeblood  of the 20th century city, data is the very beating heart of the 21st century, making cities  places of transactions [1].

In the same way cities have evolved to be places of transactions, the field of data science  has also evolved its analytical perspectives to understand the built environment, from  questioning what the data tells us (descriptive analytics), to ask why certain phenomena  happens the way it does (diagnostic analytics), what could happen in the future (predictive  analytics), and what actions can be taken next (prescriptive analytics). And now cognitive  analytics, which relies on cognitive computing, has further strengthened the predictive power  of machines by making them act like the human brain.

To understand what cognitive computing is designed to do, we can benefit from first  understanding the quality it seeks to emulate: Cognitive intelligence. Cognitive intelligence  is the human ability to think in abstract terms to reason, plan, create solutions to problems,  and learn. Cognitive computing systems works in a similar way, like the human brain. They  are able to process asynchronous information, to adapt and respond to events, and to carry  out multiple cognitive processes simultaneously to solve a specific problem.

Whilst some experts see cognitive computing as a sub-field of Artificial Intelligence (AI) [2], others see the distinction differently. According to IBM, whilst AI powers machines to do  human tasks, cognitive computing goes several steps further to create machines that can  actually think like humans [3]. Another key difference lies in the way AI and cognitive  computing are designed to tackled real-world problems. Whilst AI is meant to replace human  intelligence, cognitive computing is meant to supplement and facilitate it.

In this article it is presented how cognitive analytics exemplifies the best possible blend of  AI, machine learning (ML), deep learning (DL) and semantic technologies applied in three  variants of urban science.

Applications in GIS

Cognitive analytics has made a lot of creative applications possible even within the field of  spatial science. For example, one of them has been the intersection of GIS and AI. Esri GIS  technology with the Microsoft Cognitive Toolkit (CNTK) have advanced in applying computer  vision to geospatial analysis. Some of the deep learning use cases (image below) include:

Image classification used for pedestrian and traffic management planning, object detection  for infrastructure mapping and feature extraction, semantic segmentation to create land cover classification layers, and instance segmentation to help reconstructing 3D buildings  from lidar data [4].

Applications in Urban Research and Neuroscience

Another unique and different approach of combining semantic processing and urban analysis is a paper published in Nature, linking formal analysis of the street network and  brain activation during navigation [5]. The research uses fMRI and a simulation of London (using MATLAB Cogent2000 Toolbox, semantics processing) including 26 streets as their testing area.

One of the ways to measure how and when the brain processes topological structures to  guide future behaviour during everyday life was to combine the use of fMRI technology with Space Syntax methods (graph-theoretic analysis), which relate human behaviour to the  spatial layout of the built environment.

Space Syntax methods provided a formal way of analysing the topological properties of the  street network applied at different scales (e.g. local or global). This meant measuring properties of centrality in the selected 26 streets and how these are examined in relation to  the fMRI analysis.

The image below shows an example of the study when new streets are entered during  navigation of the city, showing right posterior hippocampal activity indexes the change in the  number of local topological connections available for future travel and right anterior  hippocampal activity reflects global properties of the street entered.

Applications in Urban Design Computing

The third application of cognitive analytics is the Cognitive Design Computing (CoDeC) [6]  workstream of the Big Data Informed Urban Design and Governance project at the Future  Cities Lab in Singapore in 2018. The idea of CoDeC was “to combine unique human design  competences with computational methods for the generation, analysis, and exploration of  urban designs”. Such computational methods enable an urban designer to approach a  computational planning process from the perspective of what performance design solution can be achieved by generating a set of possible solutions.

The cognitive urban design computing system integrates available simulation methods and  combines them with optimization and machine learning approaches, with the future aim  being to mimic the was a designer’s brain works. “Given the complex nature of urban design  problems, an additional aspect is the provision of models for human-computation interaction.  Since urban design tasks cannot be fully automated, the computational burden of an urban  design problem needs to be distributed between computer and designer.”

The workstream argues that one of the ways in which digital tools can support designers is  by generating design proposals. Learning how designers use such computational design  support systems, in combination with machine learning methods, it may be possible to derive  to more complex artificial solution strategies that can help computers make more efficient suggestions in the future.

Below is an extract of the automatization of the spatial synthesis process, using multi-criteria  optimization algorithms. For full information about the design space exploration methods,  please visit: https://youtu.be/A79UNp9pAr4

The three applications of cognitive analytics reveal how much potential this evolving  approach in data science has had in urban technologies and research. In the GIS world, deep  learning is rapidly evolving allowing data scientists to leverage cutting-edge research whilst  taking advantage of industrial-strength GIS applications. The latest ArcGIS API for Python  workflow has emerged from these deep learning capabilities. On the other hand,  Neuroscience has played a key role in the evolution of AI systems. The interdisciplinary  research of combining spatial cognition attributes with the use of AI systems, either in spatial analytics or computer science, are certainly ways to keep the journey of Data Science  evolving to new and innovative analytical approaches.

Laura Narvaez Zertuche 

 

References

[1] Sambiasi, S. 7 Architectural Considerations that are Shaping Future Cities (2019), ArchDaily. Available from: https://www.archdaily.com/927262/7-architectural-considerations-that-are-shaping-future-cities [Accessed: 9 February 2021].

 

[2] Roe, C. A Brief History of Cognitive Computing (2014), Dataversity. Available from: https://www.dataversity.net/brief-history-cognitive-computing/#:~:text=The%20first%20underpinnings%20of%20modern,was%20coined%20by%20the%2 0late [Accessed: 11 February 2021].

 

[3] Schroeer, T. Cognitive computing: Hello Watson on the shop floor (2017), IBM. Available from: https://www.ibm.com/blogs/internet-of-things/iot-cognitive-computing-watson/#:~:text=Cognitive%20computing%20refers%20to%20systems,their%20experiences%20with%2 0their%20environment [Accessed: 11 February 2021].

 

[4] Singh, R. Where Deep Learning Meets GIS (2019), ESRI. Available from : https://www.esri.com/about/newsroom/arcwatch/where-deep-learning-meets-gis/ [Accessed: 11 February  2021].

 

[5] Javadi, AH., Emo, B., Howard, L. et al. Hippocampal and prefrontal processing of network topology to  simulate the future. Nat Commun 8, 14652 (2017). https://doi.org/10.1038/ncomms14652

 

[6] The Virtual and the Real in Planning and Urban Design: Perspectives, Practices and Applications (pp.288) Chapter: 2.1. (2017), Routledge. Editors: Claudia Yamu, Alenka Poplin, Oswald Devisch, Gert De Roo. Also available from: https://toolbox.decodingspaces.net/cognitive-urban-design-computing-fcl/ [Accessed:  13 February 2021].