As a software program engineer with a background in political science I’ve spent years working on the intersection of expertise and public coverage, focusing notably on progressive applied sciences resembling web3 and machine studying. Machine studying (ML) instruments have gotten more and more accessible, with libraries like fastai expediting the event and deployment course of. This elevated accessibility has expanded the horizon of helpful purposes of present ML expertise, and impressed me to discover novel methods to make use of open supply fashions to impression public coverage.For this challenge, I’ve utilized ResNet-18 by fastai’s Python library to create an indication software which identifies areas present process inhabitants densification and concrete development.But why merge machine studying with city research and political science? As AI expertise matures, it provides the chance to enhance coverage outcomes, from addressing earnings inequality to local weather change. Of course, it’s necessary to acknowledge that many challenges stay in utilizing AI and ML for public coverage, and efforts to grasp and mitigate bias in fashions shall be essential to this work.It’s necessary to grasp the nuances of city development and inhabitants densification, and the way they will impression a metropolis. This phenomenon poses challenges, resembling:Housing Crises: Cities experiencing development usually face rising housing prices, exacerbating homelessness and housing shortages.Infrastructure and Funding: A rising inhabitants requires enhanced public companies and bodily infrastructure.Relying solely on laborious inhabitants knowledge for policy-making can danger making coverage reactionary reasonably than preventative, which . This is the place machine studying, particularly pc imaginative and prescient, can help.Computer imaginative and prescient provides an answer for approximating inhabitants development extra promptly than ready for official census knowledge. Using a ResNet-18 mannequin, fine-tuned by way of fastai, I developed a device that may establish cities at the moment present process inhabitants densification. Examples embody Raleigh, NC; Austin, TX; and Phoenix, AZ. The thrilling a part of this experiment was that every one of this achieved with minimal coaching knowledge and computational effort!You can view the demo software utilizing ResNet-18 on my HuggingFace Spaces.Defining “Transitional” Areas: In order to tune ResNet-18 to acknowledge what constituted an city space present process inhabitants densification, gathered phrases and phrases from political science and social examine disciplines which might embody the essence of city transition . For this I curated a dataset of phrases usually used to explain areas present process city development resembling: “city sprawl aerial view”, “peri-urban space aerial view”, and “rural-urban interface”.Training Data Source: Images for coaching have been sourced from Bing Image Search and separated in to a few units of knowledge used to wonderful tune the mannequin: Urban Area, Rural Area, and Transitional Area.Model Training: The software skilled the ResNet-18 mannequin over seven epochs utilizing knowledge augmentation methods resembling resizing and re-cropping randomly. The mannequin boasts roughly a 90% accuracy charge for figuring out “transitional” areas based mostly on preliminary exams.While the outcomes of this demonstration software have been attention-grabbing, it’s necessary to notice the a number of limitations of such an method:Data Sourcing: Since the photographs are fetched from Bing, there’s a danger of encountering inaccurately labeled or outdated pictures. Data was gathered by a easy Bing picture search given the benefit of utilizing their API.Subjectivity in Definitions: My interpretation of what counts as a “transitional” space is private and will circuitously correspond with precise demographic knowledge. While definitions for city, rural and transitional areas have been gathered from related sources, there in fact stays a component of subjectivity and inaccuracy.With further time and sources, a number of enhancements may very well be made to this applicationFine-tuning with Verified Data: One means to enhance accuracy is to fine-tune the mannequin utilizing pictures of cities throughout instances of verified inhabitants development.Expert Data Labeling: Consult city planners and consultants for extra correct labeling of transitional areas.Machine studying, and pc imaginative and prescient particularly, holds nice promise in figuring out societal shifts earlier than laborious knowledge turns into out there. This lead time may very well be invaluable for policy-makers dealing with the fallout of fast inhabitants development. With the continued improvement of instruments like fastai and PyTorch, engineers will have the ability to prototype efficient options with minimal time and sources.So, whether or not you’re a metropolis planner or a knowledge scientist, the implications are price your consideration. Personally, I can’t wait to see how developments on this expertise will be leveraged in public coverage to enhance the human situation.
https://medium.com/@keatondkirkpatrick/detecting-urban-population-densification-with-machine-learning-82258cbe8369
Detecting Urban Population Densification with Machine Learning | by Keaton Kirkpatrick | Sep, 2023
