We saw it, machine learning is awesome and useful. Let's focus on one of its subset, Deep Learning, which basically is large neural networks. These algorithms became possible because computers are now fast enough to use and analyze the huge amount of data we have access to. Previous post here.
Years ago, all translations systems used the word-for-word technique, but it's hard to implement, and the results are not accurate at all, so it only works with simple sentences. You first need a dictionary for each language, you replace each word with its translation, and then apply some grammar rules that you previously programmed (like swapping nouns and adjectives order).
But new techniques using Deep Learning are currently achieving top results, check out the awesome and really impressive DeepL. It's fast, super accurate, and as easy to use as Google Translate.
Colorize black and white images
No, it's not magic. It's Deep Learning, again. Because of the enormous amount of data a neural network has been fed with, it can "guess" the color of each pixels of a black and white picture. Try it out.
This one as many applications, track a brand health, evaluate reactions to a new product, measure market opportunities, or compare yourself with your competitors. Any of these can be done with sentiment analysis, using natural language processing (NLP) algorithms like Word2vec.
Before Deep Learning kicks in, sentiment analysis was done this way: every word has a "score", and a sentence sentiment is the sum its words score. The issue with this is that it doesn't work with many sentences. Take a look at : "I want an ice cream so bad", even if there is the word "bad", the overall sentence is positive. The missing piece in this algorithm is that it doesn't "understand" context, precisely what Deep Learning currently does.
Face detection and recognition
If you use Facebook, you probably noticed how your friends are automatically tagged as soon as you upload a photo. This is achieved using Deep Leaning (again!), their algorithm has an accuracy of 97.35%, literally as good as humans.
You can use these face detection/recognition (computer vision) algorithms to extract a wide range of features — as age, sex, or mood — from just a picture.
Thank you for reading this post, checkout our homepage to know what our company do.