Applications Of Deep Learning

There are specific neural network models for specific tasks and it depends on your purpose. For example, you need it for text processing, image recognition, object recognition and speech recognition.

There are dozens of application of deep learning, few of them are listed below.

Transfer learning: If you have ever shopped online there are some recommendations on side of the web page, these are the recommendation of past purchase behaviour of customers, but a deep network has revolutionised this aspect of the fashion world by learning how to match clothing combination that most customers find appealing. For example, if you had purchased a shirt it will automatically show you on the side that what kind pants or accessories will match to it.

The second application is related to image processing, it shows how deep network notice the key points of an image. For example, there is an application called neural storyteller, if you upload a picture of Earth from space it will recognise all key points like earth, cloud, stars and sunlight and it will make a story based on these key points.

Inceptionism is a neural network that hallucinates, the idea behind of this project was to showcase the feature learned by different levels of convolution neural network. For example in face detection, convolution network learns the complex features by assembling simpler ones. In the picture, it learns features like edges and colour contrast. These simple features forms complex features like nose and eyes which then combine to form a face but neural network does it automatically during its training process. In Inceptionism also known as deep dreaming, each layer and filter learn automatically so they can form a facial feature.

Next is related to speech recognition and it uses in smartphones, cars and home appliances. Tech giants Google, Apple, Samsung and Microsoft are already stepped up in the game by introducing their personal assistant google assistant, Siri, Bixby and Cortana respectively. Their speech to text feature is very fast and score higher in accuracy. The Same feature of these assistants you can find in your cars and home devices.
In object recognition, the neural network recognises the image and give a list of features of the object.

Tensor flow is one of kind of library that used in object recognition. The term library is a pre-made set of functions and modules that u can call through your own program. It is typically created by the highly qualified software team.
Google used tensor flow widely in its project like in its assistant, photos app, Gmail and search engine. However, Samsung has also introduced its new assistant called Bixby that also used tensor flow. For example: when you turn on your camera app and point to an object Bixby notice all kind of key features of an object and shows you result based on that.

There has been some advancement in application of deep learning since its introduction.
Toxicity detection for different chemical structure finds out the toxic properties from theses structure.
There is another application where a computer algorithm called deep q learning plays pong against itself and eventually, achieves expertise.

Next one is a Mitosis detection from large images. Mitosis means that cell nuclei are undergoing a different transformation that is quite harmful, and quite difficult to detect.
The best techniques out there are using a conventional neural network and are outperforming professional radiologist at their own task.
There is an application in above paragraph called hallucinations also called sequence generation. It looks at different video games and tries to learn how they work, and generates new footage out of thin air by using a recurrent neural network because of the imperfections of 3D scanning procedure, many 3D scanned pieces of furniture that are too noisy to be used as is. However, there are techniques to look at these really noisy models and try to figure out how they should look by learning the symmetries and other properties of real items of furniture.

These algorithms can also do an excellent job at different layouts predicting how different Fluids behave in time and are therefore excited to be super useful in physical simulation in the following years.
On the list of highly sophisticated scientific topics, there is the application that can find out what makes a good selfie and how good your photos are.

To Learn more on Deep Learning Click Here

Ritesh Kanjee has over 7 years in Printed Circuit Board (PCB) design as well in image processing and embedded control. He completed his Masters Degree in Electronic engineering and published a paper for IEEE called Vision-based adaptive Cruise control using Pattern matching (on Google Scholar). His work was implemented in LabVIEW. He works as an Embedded Electronic Engineer in defence research. He has experience in FPGA design with programming in both VHDL and Verilog.