TRADITIONAL COMPUTER VISION ALONG WITH DEEP LEARNING MAKES AI BETTER

A deep learning system draws textual clues from the context of images to define them without the requirement for prior human interpretations. Since its beginnings, deep learning, as both a scientific discipline and an industry, has come a long way. From cellphone assistants to pattern recognition software system, security solutions, and other applications, deep learning has become a multi-billion dollar industry poised for exponential growth over the few next years. However to attain their full potential, these softwares have to “learn” how to learn on their own.

Self-Supervised Deep Learning

The power and application of deep learning is all about its ability to identify different kinds of patterns like voices, faces, images, objects, and codes. AI software doesn’t understand what these things actually are, and all they perceive is digital data, and they’re quite good at that. The great computer vision competence of deep learning algorithms assists them to set these things apart and classify them. To do so, however, this software needs to be supervised. They need human manual input in the form of interpretations to guide them before they generalize and form what they learned into new, parallel situations. Building and labeling large datasets is a complex and time-consuming job. Unsupervised machines will be totally autonomous as all they require is data taken straight from their environment. From there, they would take the info to develop predictions and reap the desired results. To create self-supervised deep learning systems, computer experts draw inspiration from how human intelligence works.

Now, an international team of computer vision experts has invented a technique to allow deep learning software to learn the visual features of images without any need for annotated examples.

Researchers from Universitat Autonoma de Barcelona (Spain), Carnegie Mellon University (U.S.), and the International Institute of Information Technology (India), worked on the study.

Unsupervised Algorithms, it’s a Matter of Semantics

In the study, the scientists created computational models that utilize textual information about images found on websites, such as Wikipedia, and connect them to the visual structures of these images. “We aim to give computers the capability to read and understand textual information in any type of image in the real world,” said Dimosthenis Karatzas, a research team member.

In the subsequent step, researchers implemented the models to train deep learning algorithms to choose suitable visual features that textually define images. Instead of labeled info about the content of a specific image, the algorithm takes non-visual hints from the semantic textual info discovered around the image. “Our experiments demonstrate state-of-the-art performances in image classification, object detection, and multi-modal retrieval compared to recent self-supervised or naturally-supervised approaches,” wrote researchers in the paper.

This is not a completely unsupervised system as algorithms still require models to train on, but the technique reveals that deep learning algorithms can tap into the internet to improve their unsupervised learning abilities.

Enthusiastic to pursue your career in the field of Deep Learning? If so, enrolling yourself in a deep learning course at any good institute would be the ideal way to set you on the right path.