A Study on Image Classification using RecurNets

Category: Artificial Intelligence

In an era of abundance of evolution in deep learning models due to the value it provides and the massive increase in computation power due to GPUs and TPUs, the by-product is the massive environmental damage it is doing, the training time, the increasing demand for computation power, therefore, the increasing demand of very powerful devices. More often than not, researchers emphasize more on the accuracy of models rather than efficiency, resulting in carbon emissions of prolific proportions. We came up with recurrent models that save massive amounts of computation units with a bit of sacrifice in the accuracy so that it is much more efficient, can be used across lightweight devices such as IOT, robots or mobile phones and require less time. Recurrence was traditionally not used in image classification until recently.