College of Art and Architecture
875 Perimeter Drive MS 2461
Moscow, ID 83844-2461

phone: (208) 885-4409
fax: (208) 885-9428
email: caa@uidaho.edu


Urban Design Center
322 E Front Street, Suite 120
Boise, ID 83702

phone: (208) 334-2999
email: arch@uidaho.edu
website: uidaho.edu/caa/facilities/udc

Integrated Design Lab
306 S 6th St.
Boise, ID 83702

phone: (208) 429-0220
email: kevinv@uidaho.edu
website: www.idlboise.com

Architecture Design Recognition

By Karen Hunt

Spies in action movies know what tools to use when they need to get somewhere that requires some sort of face, voice or fingerprint recognition program.

While most of the world may not be breaking into somewhere using the technology, most people use recognition technology every day. They simply don’t realize it.

Recognition technology is used most commonly in Pandora, Facebook and Google, and is designed to find other items similar to what its users search.

But what if it was possible to use recognition technology in designing a floor plan or recognizing a specific type of building? That’s exactly what several students in Frank Jacobus’ Hierarchical Temporal Memory – or HTM – class are learning how to do. They are experimenting with a program that recognizes objects the same way the human brain does, uncovering similarities and differences between characteristics of similar objects.

Jacobus, assistant professor of design in the College of Art and Architecture, is working with colleagues and students conducting experiments in HTM.

“HTM is a machine learning device modeled after the human brain,” says Jacobus.

Imagine a pen. Think of all the different ways the pen can be viewed. No matter what direction the human eye sees the object, it still recognizes that it is, in fact, a pen. HTM works in the same way. The only difference is that the human eye scans images as whole objects, but the HTM program scans images as a series of shapes in space and time.

Jacobus’ graduate research course tested the program in four projects: two of the projects tested the program’s ability to recognize architectural styles while the other two tested its ability to recognize plan typologies. Each student pushed the program to see how accurately it could distinguish between types of images.

For the program to be able to identify what an object is, hundreds of images of one object has to be entered into the program. Ultimately, the more images that are entered into the program, the more accurate the program’s output becomes.

“We worked on generating images for the first month,” says Amanda Green, a graduate student in architecture. “Gradually each week, we would populate our tests.”

Garrett Lumens, a graduate student in architecture, set out to find if the program could tell the difference between Gothic and Greek architecture.

“I created an extensive photo library of Gothic cathedrals and Greek temples,” says Lumens. “The hardest part was identifying the places online to find the photos of the buildings.”

Unlike Pandora or Google that suggest music or articles based upon similarities, HTM works on a percentage scale, allowing the viewer to know what percentage of similarity the image has to a specific type.

Both Green and Lumens were able to achieve 100 percent accuracy with HTM in identifying the correct object.

Green also is working to compare levels of human recognition to that of the machine learning device.

“I have the human element where I’m testing the human involvement with the software,” says Green. “I’m looking to see if both humans and HTM recognize the same things.”

Now that Jacobus and his students know that the program can tell what an image is, the next step is to see where they can apply it in the creative design process.

“I am researching a number of issues that involve the computer's impact on architectural practice and production,” says Jacobus. “It has become my job to determine what questions to ask about how this tool can be used to positively impact architecture thought and design.”

He believes many architects may be afraid of the HTM program because they think it will eliminate part of their job or they will lose creativity. But the program is intended to be a design companion, making certain aspects of the architect’s role more efficient and effective.

“If the tool can recognize different types of space, can it recognize the building codes that go with those spaces?” says Jacobus. “This could ensure that the codes are being met while the design is being produced.”

If HTM is successful for architecture, the program may be integrated into other research areas, such as art or natural resources. But for now, Jacobus and his students are hoping that HTM will aid the creative design process.

“We’re moving into a technological age,” says Lumens. “There’s always going to be something new. We can either embrace what’s coming or be left behind.”