Discover the world's scientific knowledge

Hello, I am working on (two-class) image classification problem. One observation is that the complexity/variability of the data for each class has a major role to play in the classification accuracy. What complexity measures can be used to decide the parameters used in supervised classification?

Can the Bacterial Foraging Otimisation Algorithm be classified as Np-Hard problem ? and do it belongs to Evolutionary Algorithms (EAs)? 

Algorithms

I am comparing the computational complexity of my algorithm with ISTA (Iterative Shrinkage-Thresholding Algorithm) and OMP (Orthogonal Matching Pursuit) algorithm for a classification task. But now I want to find out complexities for these two compared algorithms in terms of sparsity and dimension of vector. Could anybody suggest some corresponding solutions or literature for this problem?

Thanks in advance.

A segmentation process gives phonetic segments of variable length, then this segments are labeled one out of many phonemes(classification). Just i would want to know how this (segmentation+classification) task is  different from recognition process.

Decision Trees are extremely fast in classification. However, they are slow in construction during learning phase. Is there any paper on complexity analysis of Multiway Split, Multi-Class Decision tree?

Suppose we conveniently extended the standard concept of cellular automaton to include

graphs and state-spaces of any cardinality and that the transition function F belonged to a certain adequate notion of "(hyper)computable function". We call this a hyper-cellular automaton HCA.

Consider the postulate: the universe can be described by a HCA with transition function F.

We cannot escape the problem of the initial condition Q_0. In the Wolfram Classification random initial condition are considered. Hence expediency for some topology of measure on Q.

It is an empirical fact that this HCA must be WC4 "complex patterns of localised structures" in the Wolfram Classification.

A major problem in the goal of reverse engineering F is that we do not have evolutions for other initial conditions at our disposal neither for the universe nor for subsystems of the universe. For physics at least a lot of locality and invariance hypothesis come in to play to justify the universality of experimental conclusions. The chemistry we observe on earth must also be that of the most distant star.

For biology the situation is drastically different. My question is: how can biology go beyond being a merely descriptive or descriptive-hypothetical science ?

How can the natural language based, often highly metaphorical and anthropomorphic, descriptions given in evolutionary biology, be translated into rigorous formal concepts involving a HCA ?

We need a general qualitative theory of emergent complex behaviour in cellular automata.

How can I merge covariance matrices in one matrice to reduce the computing complexity?

There are many nested (partially or fully) communities in a complex network. Taking a country for example, cities are communities within the country, neighborhoods are communities within the cities, and families are communities within the neighborhoods. Eventually, there are far more small communities than large ones, or the community sizes demonstrate power laws or heavy tailed distributions, which we have empirically verified.

Jiang B. and Ma D. (2015), Defining least community as a homogeneous group in complex networks, Physica A, 428, 154-160

Jiang B., Duan Y., Lu F., Yang T. and Zhao J. (2014), Topological structure of urban street networks from the perspective of degree correlations, Environment and Planning B: Planning and Design, 41(5), 813-828.

However, communities detected by our algorithm hardly match to those by previous community detection methods. My question is, should communities be nested? 

If we have multiple classifiers and we need to know which one is under-fitting, and which one is overfitting based on performance factors (classification accuracy, and model complexity)

Are there any method to select the dominate classifier (optimal fitting) that balance between the above-mentioned two factors?