AI & Deep learning

Architecture : the organization of the model. It range from very simple structures like standard decision trees, Support Vector Machine (SVM) and  neural network to highly complex ones involving among others GRU, LSTM and CNN networks, GAN structures ,…

Artificial Intelligence (AI) : refers to the ability of a computer or a machine to mimic or even over-perform human intelligence in complex cognitive functions and tasks

Backpropagation : algorithm widely used with neural networks for training  feedforward models

Classification : technique for identifying to which of a set of possible categories an observation belongs

Deep learning (DL) : subset of the machine learning that involves deep or complex techniques, usually neural networks with complex architectures

Gradient descent : optimization algorithm with iterative function used to find minima of differentiable loss functions

Hyper-parameters : technical parameters of an AI model. As various architectures or techniques might be used, the list of hyper-parameters is very broad. They are often opposed to the ‘parameters’ that are non-AI-model specific

Machine learning (ML) : is a subset of AI that focuses on the ability of machines or computers to learn complex cognitive functions

Neural network (NN or ANN) : is an AI set of algorithmic techniques vaguely mimicking the humain brain functioning that 

Overfitting : problem encountered in machine learning when the model is very efficient with the training set but performs poorly with the test set. “High variance” is often used to describe overfitting problem

Regression : technique for estimating the relationship between a dependent variable ‘y’ and independent variable(s) Xi

Reinforcement learning : technique of machine learning dedicated to guide an ‘intelligent agent’ to maximize cumulative rewards for its actions/decisions. Reinforcement learning is the third main domain of ML together with supervised learning and unsupervised learning

Replicability : ability to obtain similar results with similar but not identical parameters and similar data

Reproducibility : ability to obtain exactly the same results when the model is tested by a different person and / or on a different computer

Supervised learning :  machine learning task of learning a function that maps an input to an output based on examples of input-output pairs. It infers a function from labeled training set that is then applied to non-labeled test dataset

Underfitting : problem encountered in machine learning when the model is not efficient with the training set and does not reach a satisfactory result. This often occurs when the model is not deep or complex enough to apprehend the embedded complexity of the data. “High bias” is often used to refer to underfitting problems

Unsupervised learning : machine learning task of learning a function that maps an input to an output based on examples of input-output pairs. It infers a function from labeled training set that is then applied to non-labeled test dataset