- Automatic Generation of Handwriting: For a common context of a word or phrase, this Deep Learning application comprises the development of a fresh set of handwriting. Effectively, the handwriting is shown as a set of coordinates used by a pen to create the samples. It is revealed that pen motion and letter production are related, and more examples are generated.
- Playing Games Automatically: Here, fresh text has been written word by word or byte for character from a corpus of text. It is feasible to learn how to spell, pronounce, and even recognise the form of the text in corpus phrases using deep learning algorithms. Usually, to learn text creation from items in sequence of input strings, massive recurrent neural network are used. However, by using a persona model that generates one character at a time, LSTM recurrent neural network have recently demonstrated exceptional success in this challenge.
- Translations of languages: Technology companies are paying close attention to machine translation. Significant improvements in translation quality have been achieved as a result of this investment and current developments in deep learning. Google claims that switching to deep learning increased translation accuracy in Google Translate by 60% compared to the previous phrase-based approach. More than 100 languages can now be translated by Google and Microsoft with relatively close accuracy in a number of them.
- Restoration of pixels: Before Deep Learning, it was difficult to zoom into videos any further than their true resolution. In 2017, Google Brain researchers developed a Deep Learning network to identify faces from crappy images taken of people. This strategy, referred to as Pixel Iterators Super Resolution, makes use of pixels to produce super resolution. It significantly increases photo resolution, emphasising key details just enough to enable personality recognition.
- Election and demographic predictions: Gebru et al. used 50 million Streetview images to test the potential of a deep learning network. The results were outstanding as usual. The computer become adept at identifying and locating autos and their specifications. It was able to recognise over 22 million vehicles and provide information on their make, type, body style, and the year. Inspired by the success of these Deep Learning capabilities, the investigations continued. It was demonstrated that the system could estimate each location’s demographics based solely on the make and model of its vehicles.
- Lucid dreaming: An project called DeepDream displays patterns that a neural network has learned. DeepDream over-interprets and magnifies the patterns it discovers in a picture, intensifying them like a little child seeing clouds and trying to make sense of their illogical shapes.
By passing an image from across network and determining the gradient of it in proportion to the activation functions of a certain layer, it is able to achieve this. The image is then changed to emphasise these activations, which improves the network’s perception of the patterns and results in a surreal visual. This approach was known as “Inceptionism”.