Research Breakthrough: Artificial Intelligence (AI) Can Now Use Its Imagination

Researchers at the University of Southern California (USC)  are helping the use of Artificial Intelligence (AI) imagine what has not been seen, a technique that would help better AI, new and improved medicines and of course autonomous vehicle safety.

The Research team which has in its conglomeration Professor Laurent Itti, a computer science professortogether with PhD students, Yunhao Ge, Sami Abu-El-Haija and Gan Xin developed an AI that uses human-like capabilities to imagine a never-before-seen object with different attributes.

The research paper dubbed: Zero-Shot Synthesis with Group-Supervised Learning, was first published in the 2021 International Conference on Learning Representations on May 7.

The study’s lead author, a PhD student, YunhaoGe talked about the inspiration behind the project when he said:

“We were inspired by human visual generalization capabilities to try to simulate human imagination in machines. Humans can separate their learned knowledge by attributes—for instance, shape, pose, position, color—and then recombine them to imagine a new object. Our paper attempts to simulate this process using neural networks.”

Analysing the Artificial Intelligence (AI) generalisation problem.

Let’s have a look at this scenario. We want to create an AI system that generates the pictures of cars and it’s a requirement that algorithms of a few car images would be provided to be able to generate many types of cars, let’s say from Porsches, Pontiacs, to pick-up trucks, in any color and generated from multiple angles.

One of AI long-sought goals is creating models that can extrapolate, which implies that with few specimen examples, a model should have the capacity to extract the underlining format and apply them to vast range of other examples it hasn’t seen before.

But here comes the problem of Artificial Intelligence, the generalization issue. AI machines are mostly configured in a way that they take in abstracts like sample pictures without taking into cognizance the attributes of the objects. This connotes that they cannot think on their own.

The Researchers with the New AI Scienceof Imagination

A concept called ‘disentanglement’ was used by the researchers to solve the AI’s limitation of imagination.

Then how does disentanglement works?

Yunhao Ge enumerated how disentanglement can be used to generate deepfakes, and in the case of images, by disentangling human face movements and identity.

He added that by doing that, “people can synthesize new images and videos that substitute the original person’s identity with another person, but keep the original movement.”

The new method will rather take a group of sample images instead of one sample at a time as traditional algorithms have come to be accustomed with, mining the similarity between them to produce “controllable disentangled representation learning.”

It afterward recombine the ensuing knowledge to have what is called a ‘controllable novel image synthesis’, better summed as ‘imagination’

Ge went on to highlight the benefits of this when he said;

“For instance, take the Transformer movie as an example. It can take the shape of Megatron car, the color and pose of a yellow Bumblebee car, and the background of New York’s Times Square. The result will be a Bumblebee-colored Megatron car driving in Times Square, even if this sample was not witnessed during the training session.”

This can be said to be akin to the process of human extrapolation wherein a human sees an object and apply it to another object by substituting the original color with the new one.

The research team, having being availed of this technique, generated 1.56 million images of new sets of data that would be beneficial in further future research.

Implications and understanding the effect of the Study

According to the team of researchers, the framework of the study can be in alignment with nearly any type of data or knowledge, widening the opportunity for different applications. For example, race and gender-related knowledge can be disentangled by removing sensitive attributes from the equation altogether, ensuring fairer and more effective Artificial Intelligence.

In the medical line, it has possibility of helping doctors and biologists discover more drugs by the process of disentangling the medicine function of a drug from its other properties and then recombining them to make new medicine.

In the automobile industry, safer AI can be created with ‘imaginative’ Imbuing machines, giving autonomous vehicles the leverage to imagine and avoid dangerous scenarios that were not seen at the period of manufacturing and training.

An elated Professor Laurent Itti opined:

“Deep learning has already demonstrated unsurpassed performance and promise in many domains, but all too often this has happened through shallow mimicry, and without a deeper understanding of the separate attributes that make each object unique. This new disentanglement approach, for the first time, truly unleashes a new sense of imagination in A.I. systems, bringing them closer to humans’ understanding of the world.” 

Reference: “Zero-shot Synthesis with Group-Supervised Learning” by Yunhao Ge, Sami Abu-El-Haija, Gan Xin and Laurent Itti, 7 May 2021, 2021 International Conference on Learning Representations.
Hyperlink