Artificial immune recognition system
The artificial immune recognition system is inspired by the theory of clonal selection of acquired immunity. The theory of clonal selection attributed to Burnet was proposed to take into account the behavior and capacities of antibodies in the acquired immune system. Drawing on the principles of the theory of evolution of Darwinian natural selection, the theory proposes that antigens select lymphocytes (both B cells and T cells).
When a lymphocyte is selected and binds to an antigenic determinant, the cell proliferates making several thousand more copies of itself and differentiates into different types of cells (plasma and memory cells). Plasma cells have a short lifespan and produce large amounts of antibody molecules, while memory cells live for an extended period of time in the host, anticipating future recognition of the same determinant.
The important feature of the theory is that when a cell is selected and proliferated, it is subjected to small copying errors (changes in the genome called somatic hypermutation) that change the shape of the expressed receptors. It also affects the recognition abilities of subsequent determinants of both antibodies bound to the cell surface of lymphocytes and antibodies produced by plasma cells.
The theory suggests that from an initial repertoire of general immune cells, the system is able to change itself (the compositions and densities of cells and their receptors) in response to experience gained from the environment. Through a blind process of selection and accumulated variation on a large scale of several billion cells, the acquired immune system is able to acquire the information necessary to protect the host organism against specific pathogenic dangers of the environment. It also suggests that the system must anticipate (guess) the pathogen to which it will be exposed and requires exposure to a pathogen that can harm the host before it can acquire the information necessary to provide a defense.
The objective of the technique is to prepare a set of real-valued vectors for classifying the models. The artificial immune recognition system maintains a pool of memory cells that are prepared by exposing the system to a single iteration of training data. Candidate memory cells are prepared for the event that the memory cells are insufficiently stimulated for a given input model. A process of cell cloning and mutation occurs for the most stimulated memory cell.
Clones compete with each other for the memory pool based on stimulation and the amount of resources used by each cell. This concept of resources comes from previous work on artificial immune networks, where a single cell (an artificial recognition ball, or ARB) represents a set of similar cells. Here, a cell's resources are a function of its stimulation to a given input pattern and the number of clones it can create.
The following algorithm provides a pseudocode for preparing memory cell vectors using the artificial immune recognition system, in particular the canonical AIRS2 (artificial immune recognition system 2 in French). An affinity measure (distance) between the input models must be defined. For real-valued vectors, this is usually the Euclidean distance:
where n is the number of attributes, x is the input vector, and c is a given cell vector. The variation of cells during cloning (somatic hypermutation) occurs inversely proportional to the stimulation of a given cell to an input pattern.
the artificial immune recognition system was designed as a supervised algorithm for classification problem areas. the artificial immune recognition system is not parametric, which means that it does not rely on assumptions about the structure of the function which is approximate.
The actual values in the input vectors must be normalized such that x is in [0; 1]. Euclidean distance is commonly used to measure the distance between real-valued vectors (an affinity calculation), although other distance measures can be used (such as the dot product), and data-specific distance measures. may be required for non-scalar attributes. Cells can be initialized with small random values or more commonly with values of instances in the training set. The affinity of a cell is typically minimizing, whereas as a cell the stimulation is maximizing and typically in [0; 1].