The new digital revolution, deep learning

Recognize a photo, a song, a user’s habit. With artificial intelligence it is already possible. But why is it important and how does it affect our lifestyle?

Before answering this question, we must take a step back to explain the difference Artificial Intelligence (IN THE), Machine Learning (ML) and Deep Learning (DL), terms that are often confused but with precise meaning.

To explain the basic idea, let’s use an image (opening) from the NVIDIA website.

From the picture it is obvious that the concept of artificial intelligence is a general concept of ML which in turn is a more general concept of DL. But not only. In fact, as we see in his first algorithms deep learning were born just 10 years ago in contrast to artificial intelligence born around the 1950s with early languages ​​such as LISP and PROLOG aiming to mimic the potential of human intelligence.

The first artificial intelligence algorithms were limited to performing a certain probable number of actions according to a specific logic set by the developer (as in the game of checkers or chess).

Throughout machine learningArtificial intelligence has evolved through so-called supervised and unsupervised learning algorithms to create mathematical automated learning models based on large volumes of input data that constitute the “experience” of artificial intelligence.

In supervised learning, in order to create the model, it is necessary to train the artificial intelligence by assigning a label to each element: for example, if I want to sort the fruit, I will take pictures of many different apples and label the model “apple”. “. so for pear, banana etc.

In unsupervised learning, the process will be reversed: a model will have to be created starting from different fruit images and the model will have to export the labels according to the common characteristics of apples, pears and bananas.

His models machine learning supervisors are already used by antivirus, spam filters, but also in marketing such as the products offered by amazon.

The example of the spam filter

The idea behind a spam filter is to train a model that “learns” from hundreds of thousands (if not millions) of emails, labeling each email as spam or legal. Once the model is trained, the sorting function includes:

Export peculiar features (called features) such as text words, email sender, source IP address, etc.

Think of a “weight” for each exported attribute (for example, if there are 1000 words in the text, some of them may be more discreet than others, such as the word “viagra”, “porn” etc. will have a different weight than good morning, universities, etc.)

Perform a mathematical function, which, taking as input elements (words, sender, etc.) and their corresponding weights, returns a numeric value

Check if this value is above or below a certain threshold to determine if the email is legal or unsolicited (sort).

Artificial neurons

As mentioned, the Deep learning is his branch machine learning. The difference with machine learning It’s the computational complexity that brings huge amounts of data into the game with a “flat” learning structure made of artificial neural networks. To understand this concept, we start with the idea of ​​replicating the unique human neuron as in the figure below.

As we saw earlier for machine learning, we have a series of input signals (to the left of the image) to which we associate different weights (Wk), we add a cognitive “bias” which is a kind of distortion and finally we apply a function activation, i.e. a mathematical function such as a sigmoid function, an over-tangent, ReLU, etc. which by taking a series of weighted inputs and taking into account a bias, returns an output (yk).

This is the only artificial neuron. To create a neural network, the outputs of the individual neuron are connected to one of the inputs of the next neuron, forming a dense network of connections as shown in the figure below that represents the actual Deep Neural Network.

Deep Learning

As we can see from the above figure we have a set of inputs that must be provided to the neural network (input level), after the intermediate levels called hidden layers that represent the “layers” of the model and finally an output level capable of distinguishing (or recognize) one object on top of another. We can think of each hidden level as a learning ability: the greater the number of intermediate levels (ie the deeper the model), the more accurate the comprehension will be and the more complex the calculations to be made.

Note that the output level represents a set of output values ​​with a certain degree of probability, for example 95% an apple, 4.9% a pear and 0.1% a banana and so on.

Imagine a DL model in its field computer vision: the first layer can recognize the edges of the object, the second layer starting from the edges can recognize the shapes, the third layer starting from the shapes can recognize complex objects consisting of many shapes, the fourth layer starting from complex shapes can recognize details and so on. There is no exact number when defining a model hidden layerbut the limit is imposed by the power required to train the model at a specific time.

Without going into too much detail, the training of a neural network aims to calculate all the weights and biases that must be applied to all the individual neurons in the model: it is therefore obvious that complexity increases exponentially to the intermediate levels increase (hidden layer). For this reason, graphics card processors (GPUs) have been used for education: these cards are suitable for more demanding workloads as, unlike CPUs, they can perform thousands of functions in parallel using SIMD (Single Instruction Multiple Data) architectures as well as modern technologies such as e.g. Tensor core allowing matrix functions on hardware.

Deep Learning Applications

By processing huge amounts of data, these models have a high tolerance for errors and noise despite incomplete or inaccurate data. Therefore, they provide fundamental support in every field of science. Let’s look at some of them.

Image classification and security

In the case of crime, it allows a person to be identified starting with the image captured by a surveillance camera and comparing it to a database of millions of people: this function, if done manually by humans, could take days, if not months or even years. In addition, through image reconstruction, some models allow coloring of missing parts, with an accuracy close to 100% of the original color.

Natural Language Editing

The ability of a computer to understand written and spoken text in the same way as humans. The most famous Alexa and Siri systems can not only understand but answer questions of a different nature.

Other models can do it emotion analysisalways using systems of extraction and views from the text or words.

Medical diagnoses

In the medical field, these models are now used to make diagnoses, including computed tomography or magnetic resonance imaging. Outcomes that are 90-95% confident at the exit level, in some cases, can predict a cure for the patient without human intervention. Able to work 24 hours a day, every day, they can also provide support during the patient’s screening phase, significantly reducing waiting times in an emergency room.

Autonomous driving

Autonomous driving systems require continuous real-time monitoring. More advanced models provide for vehicles capable of handling any driving situation independently of a driver whose presence on board is not foreseen, providing for the presence of only transportable passengers.

Predictions and Profiles

Deep learning financial models allow us to make assumptions about future market trends or to know the insolvency risk of an institution more accurately than people today with interviews, studies, questionnaires and manual calculations.

These models used in marketing let us know the tastes of people to suggest new products, for example, based on correlations made with other users who have a similar shopping history.

Adaptive developments

Based on the “experiences” that have been downloaded, the model is able to adapt to situations that occur in the environment or due to user input. Adaptive algorithms cause an update of the entire neural network based on new interactions with the model. For example, let’s imagine how YouTube offers videos with a specific theme, depending on the season, adapting day by day and month by month to our new personal tastes and interests.

In conclusion, the Deep Learning is still a rapidly growing field of research. Universities are also updating their curricula on this subject, which still requires a solid foundation in mathematics.

The benefits of applying DL to industry, research, health and everyday life are undeniable.

However, we must not forget that this must provide support to the person and that only in certain limited and very specific cases can it replace the person. To date, in fact, there are no “general purpose” models that are capable of solving any kind of problem.

Another aspect is the use of these templates for illegal purposes, such as creating videos DeepFake (see article), ie techniques used to overlay other images and videos with original images or videos with the aim of creating fake news, scams or revenge porn.

Another illegal way to use these models is to create a series of techniques aimed at endangering a computer system, such as reverse machine learning. Through these techniques it is possible to cause the model to be incorrectly classified (and thus motivate the model to make the wrong choice), obtain information about the data set used (introduce privacy issues) or clone the model (causing copyright issues) .

bibliographical references

https: //blogs.nvidia.com/blog/2016/07/29/whats-difference-artificial-int …

https://it.wikipedia.org/wiki/Lisp

https://it.wikipedia.org/wiki/Prolog

https://it.wikipedia.org/wiki/Learning_supervisioned

https: //www.enjoyalgorithms.com/blog/email-spam-and-non-spam-filtering-u …

https://foresta.sisef.org/contents/?id=efor0349-0030098

https://towardsdatascience.com/training-deep-neural-networks-9fdb1964b964

https://hemprasad.wordpress.com/2013/07/18/cpu-vs-gpu-performance/

https://it.wikipedia.org/wiki/Analisi_del_sentiment

https: //www.ai4business.it/intelligenza-artificiale/auto-a-guida-autonom …

https: //www.linkedin.com/posts/andrea-piras-3a40554b_deepfake-leonardodi …

https://arxiv.org/abs/1712.03141

Leave a Comment