Machine learning allows computers to mimic human behaviour by teaching them historical knowledge. This article describes many machine learning techniques. It includes the Label propagation algorithm, the Dehaene–Changeux model, and the LogitBoost algorithm.

Label propagation algorithm

Label propagation is a machine learning algorithm that gives labels to data points that didn't have labels before. At the beginning of the algorithm, labels are only on a small number of the data points (or classifications). 

Real networks tend to have a community structure even though they are complex. Label propagation is a way to find groups of people. Label propagation is better than other algorithms regarding how long it takes to run and how much information about the network structure. The problem is that it doesn't give a single answer but rather a collection of many answers.

Furthermore, Label propagation is different from other algorithms because it can lead to different community structures from the same starting point. Some nodes are preliminary labels, while others are left unlabeled. This approach can help narrow the number of possible solutions. So, unlabeled nodes will more likely adapt to the labelled ones. Jaccard's index gathers all the essential information from different community structures to find more accurately.

Dehaene–Changeux model

The Dehaene–Changeux model (DCM), also called the global neuronal workspace model or the global cognitive workspace model is a part of Bernard Baars's "global workspace model" for consciousness. It is a neural network-based computer model of the neural correlates of consciousness. It tries to mimic how the brain's higher cognitive functions, like consciousness, making decisions, and the central executive functions, act like a swarm. Both cognitive neuroscientists Stanislas Dehaene and Jean-Pierre Changeux started working on it in 1986. It figures out how to solve the Tower of London test and study "inattentional blindness."

Furthermore, the Dehaene–Changeux model is a meta neural network, which is a network of neural networks. The neurons are set in complex thalamo-cortical columns with long-range connections. The interaction between von Economo's areas is crucial to the brain's work. Each thalamo-cortical column comprises pyramidal cells and inhibitory interneurons. That receive long-distance excitatory neuromodulation, which could be noradrenergic input.

LogitBoost

LogitBoost is a boosting algorithm by Jerome Friedman, Trevor Hastie, and Robert Tibshirani. It is in machine learning and computational learning theory. 

The AdaBoost algorithm is a statistical framework in the original paper. In particular, the LogitBoost algorithm can be made by taking AdaBoost as a generalized additive model and then applying the cost function of logistic regression to it. In addition, both LogitBoost and AdaBoost perform an additive logistic regression, so they are similar. The difference is that AdaBoost tries to reduce the loss as much as possible.

Furthermore, LogitBoost is a popular type of Boosting. We can use that for binary classification and with more than two classes. From a statistical point of view, LogitBoost can be used as additive tree regression where the Logistic loss is as low as possible. However, using this setting, it is still harder to make an excellent multiclass LogitBoost than to make its binary equivalent. In multiclass Logistic loss, two essential things cause problems. 

1. The first is that the Logistic loss implies that the optimal classifier output is not unique. It means that adding a constant to each part of the output vector won't change the value of the loss. 
2. The second is the density of the Hessian matrices created when computing tree node split gain and node value fittings.

In math, the Hessian matrix is a square matrix of partial second-order derivatives of a scalar field or function. It talks about how a function with many variables curves locally. Ludwig Otto Hesse, a German mathematician from the 1800s, came up with the Hessian matrix. Hesse used the phrase "functional determinants" at first.