Unique AI System Identifies Objects Via Similar Processes As Humans

By DocWire News Editors - Last Updated: April 11, 2023

A group of researchers from UCLA Samueli School of Engineering and Stanford have recently created an artificial intelligence (AI) system that can identify and discover objects it encounters using visual learning methods similar to a human’s.

Advertisement

This AI system is currently in a “computer vision” class of technology, which allows computers to efficiently analyze visual images. This is a very important step leading to AI systems that are capable of self-learning, intuition, reasoning, and decision making in a more human-like manner. With current AI computer visual systems becoming more powerful and able as complexity increases, they are very limited to specific tasks. These systems have the ability to identify things that they have been trained and programmed to see but lack the flexibility to adapt.

One skill that current AI visual systems lack is the ability to deduce a full image from only a partial image of an object- for example if you were to see an image of a leaf and a branch you could envision a whole tree from this. These systems also struggle to view objects in unfamiliar settings, with different aspects of the scene deceiving them.

READ MORE: Researchers Develop AI With Human-Like Hearing

Computer engineers have been aiming to create AI systems that have these abilities we see as unique to humans. Breaking through this barrier would allow computer systems to see a partial image of a human standing behind a wall and understand from the visible head and hand where the rest of the body is.

ai

In AI visual systems, the computer is not capable of making its own rational decisions or learn on its own. These systems are merely trained by thousands of images to pick up on certain cues that have been labelled as areas of interest. To overcome this challenge, the scientists devised a three-step system that was described in Proceedings of the National Academy of Sciences.

First, this AI system breaks an image up into small chunks that the researchers dubbed “viewlets”. Second, the computer then learns how each of these viewlets fit together to create the whole image. Lastly, the system looks at the objects in the surrounding area and distinguishes whether or not this data is relevant to the primary object or if it is extraneous.

To assist the AI in adapting more like a human, the researchers immersed the system into an internet replica of the environment we live in as humans.

“Fortunately, the internet provides two things that help a brain-inspired computer vision system learn the same way humans do,” said Vwani Roychowdhury, UCLA engineering professor and principal investigator of the study. “One is a wealth of images and videos that depict the same types of objects. The second is that these objects are shown from many perspectives–obscured, bird’s eye, up-close–and they are placed in different kinds of environments.”

READ MORE: IDx is Paving the Way for AI Use in Medicine

To create the structure for their system, the researchers collected insight from cognitive psychology and neuroscience.

“Starting as infants, we learn what something is because we see many examples of it, in many contexts,” Roychowdhury said. “That contextual learning is a key feature of our brains, and it helps us build robust models of objects that are part of an integrated worldview where everything is functionally connected.”

Testing their AI system on roughly 9,000 images that showed people and various objects, the researchers found that their platform was capable of building detailed models of the human body with no external guidance or labeling of the images. Successful results were yielded via images of various vehicles as well, with their system matching, or outperforming all traditional computer visual systems.

Source: UCLA Samueli School of Engineering

Advertisement