Tuesday, 22 December 2015

How Boring is Your Face? Ask the 'MemNet' Algorithm


MemNet
Ever thought of knowing how good your face looks in the pictures captured by the selfie cameras. A group of scientists at the MIT’s Computer Science and Artificial Intelligence Laboratory has developed a unique computer algorithm which can easily measure ‘’how memorable an image is’. One is just required to feed the photo from the Instagram feed and the MemNet algorithm will simply show the portion of the images, which are most likely to stick in the memories of viewers. Scientist has even uploaded a version of this algorithm online for everyone to try out.

What is MemNet? 

MmeNet is a deep-learning algorithm, which is designed by the scientist at MIT. Deep learning algorithm is quite hard to develop as they are built to acquire and incorporate new information, which helps in improving their abilities without any manipulation or control of human programmers. Deep learning technologies work in its own unique manner as essentially mimics the neural pathways which area associated with the human learning. This helps deep learning algorithm in repurposing the same logic and understanding for various different skill sets.

MemNet to get better at reading images

MemNet is expected to improve its ability to read images and predict an images’ indelibility by simply processing more and more information over time. These computer scientists have released the MemNet version for the common users to try and with prediction; it will improve its ability of reading faces with more accuracy.

MemNet was first endowed with the basic required skill sets by the MIT scientists through uploading of thousands of images and associated data. The metadata fed to the MemNet basically included the information about each image’s popularity and the emotional impact as determined by the online viewers. When asked to predict the memorable parts within a picture MemNet performed as good as human.

The lead study author, Aditya Khosla, stated that understanding the memorability will help us in making systems, which can capture the most important information, and it can even help in storing the information which are most likely to be forgotten by the humans. Visual images are mostly preferred by a large number of institutions to send across the messages but MemNet can help it making more memorable than before.

Future prospect of the MemNet algorithm

MemNet is not much better at the task of measuring the memorability than any other human but researchers are hopeful it will get better over time. It creates a heatmap to show the memorable part and boring part of each image, which happens to the key elements in understanding the memorability level of the pictures. This algorithm can help the marketers and even the movie makers in editing their images in such a manner than it stays inside the head of the customers and viewers. MemNet can also help people in learning various things easily and efficiently too. People tend to assimilate and forget things more often but the research in developing MemNet opens up a new avenue. It has the potential to improve the people’s memory they are presented with the memorable images.

No comments:

Post a comment

Note: only a member of this blog may post a comment.