Meet I-JEPA: Meta’s ‘Human-like’ AI Image Creation Model

Meta revealed its all-new human-like AI image creation model called I-JEPA, which will use background knowledge about the world to complete images.

Mark Zuckerberg’s Meta recently announced that it will be bringing a human-like AI image creation model which will be superior to the existing AI models. The company said it would enable the researchers to access the components of this Artificial Intelligence model that could significantly help them in completing images through accurate image analysis.

Meta claims that the Image Joint Embedding Predictive Architecture model (I-JEPA) will technically use the background knowledge of the world to complete the missing parts in an image. It is a more intellectual approach when compared to the existing approach used by the current Generative AI models based on the identification of nearby pixels.

The company’s blog post says, “This model, the Image Joint Embedding Predictive Architecture (I-JEPA), learns by creating an internal model of the outside world, which compares abstract representations of images (rather than comparing the pixels themselves).”

The company mentioned that the model is inspired by the fact that humans in general gain extensive background knowledge about the world merely by observing things. Hence, learning algorithms that can grasp common sense and incorporate this general knowledge are essential for these AI image models to facilitate intelligent behaviors.

The advanced AI model is aimed at creating images that are free from common errors that repeatedly occur in AI-generated images through the existing Generative AI image models.

It is the vision of the Chief Artificial Intelligence scientist at Meta, Yann LeCun behind the development of this AI image model, and the company expressed its enthusiasm while unveiling the same in their blog post. Yann LeCun aims to create machines that can learn internal models of how the world works to increase their efficiency and accuracy while performing complex tasks.

While highlighting its efforts toward building human-like Artificial Intelligence, the company also mentioned, “We look forward to working to extend the JEPA approach to other domains, like image-text paired data and video data. In the future, JEPA models could have exciting applications for tasks like video understanding. This is an important step towards applying and scaling self-supervised methods for learning a general model of the world.”

Meta has been a constant publisher of open-sourced AI research through its in-house research lab. The company believes in sharing models developed by its researchers to stimulate innovation, spotting safety gaps, and lowering costs

Mark Zuckerberg also stated,

“For us, it’s way better if the industry standardizes on the basic tools that we’re using and therefore we can benefit from the improvements that others make”.