Geometry Network size

The finished network automatically reflects the characteristics of the data domain Not only do the network weights evolve so that they describe the data as fully as possible, but so also does the network geometry. The size of the network is not chosen in advance and as topology is determined by the algorithm and the dataset in combination, it is more likely to be appropriate than the geometry used for a SOM, especially in the hands of an inexperienced user, who might find it difficult to choose an appropriate size of network or suitable values for the adjustable parameters in the SOM. 

The pore geometry described in the above section plays a dominant role in the fluid transport through the media. For example, Katz and Thompson [64] reported a strong correlation between permeability and the size of the pore throat determined from Hg intrusion experiments. This is often understood in terms of a capillary model for porous media in which the main contribution to the single phase flow is the smallest restriction in the pore network, i.e., the pore throat. On the other hand, understanding multiphase flow in porous media requires a more complete picture of the pore network, including pore body and pore throat. For example, in a capillary model, complete displacement of both phases can be achieved. However, in real porous media, one finds that displacement of one or both phases can be hindered, giving rise to the concept of residue saturation. In the production of crude oil, this often dictates the fraction of oil that will not flow. 

Overfitting is a potentially serious problem in neural networks. It is tackled in two ways (1) by continually monitoring the quality of training as it occurs using a test set, and (2) by ensuring that the geometry of the network (its size and the way the nodes are connected) is appropriate for the size of the dataset. 

The growing cell structure algorithm is a variant of a Kohonen network, so the GCS displays several similarities with the SOM. The most distinctive feature of the GCS is that the topology is self-adaptive, adjusting as the algorithm learns about classes in the data. So, unlike the SOM, in which the layout of nodes is regular and predefined, the GCS is not constrained in advance to a particular size of network or a certain lattice geometry. 

The SANS experiment is applicable to polymeric networks containing some deuterium labeled chains. The chain geometry can be probed not only in the unperturbed network, but changes in chain shape and size can be measured as a function of strain or swelling. This enhances the applicability of SANS experiments for elastomeric systems. 

As discussed above, hysteresis loops can appear in sorption isotherms as result of different adsorption and desorption mechanisms arising in single pores. A porous material is usually built up of interconnected pores of irregular size and geometry. Even if the adsorption mechanism is reversible, hysteresis can still occur because of network effects which are now widely accepted as being a percolation problem [21, 81] associated with specific pore connectivities. Percolation theory for the description of connectivity-related phenomena was first introduced by Broad-bent et al. [88]. Following this approach, Seaton [89] has proposed a method for the determination of connectivity parameters from nitrogen sorption measurements. 

Several studies have considered the influence of filler type, size, concentration and geometry on shear yielding in highly loaded polymer melts. For example, the dynamic viscosity of polyethylene containing glass spheres, barium sulfate and calcium carbonate of various particle sizes was reported by Kambe and Takano [46]. Viscosity at very low frequencies was found to be sensitive to the network structure formed by the particles, and increased with filler concentration and decreasing particle size. However, the effects observed were dependent on the nature of the filler and its interaction with the polymer melt. 

At the end of the 1970 s and the early 1980 s this programme showed the highly non linear elastoviscous behaviour of salt rock. From the 1980 s, 2D software allowed the sizing of isolated or network caverns and from the 1990 s, 3D software permitted the sizing of any cavity as well as the prediction of changes in cavity geometry during operation. 

The three-dimensional network crystallises in the cubic space group P2i3, Z = 4. The general features of the crystal structure have been discussed earlier. As depicted in Figure 26, the structure is such that the three-dimensional oxalate backbone provides cavities which in terms of size and geometry provide a perfect fit for [Mn(bpy)3]2+ ions. 

Die mathematical significance of rh has been established for a number of other pore geometries (Everett, 1958), but with most real systems it is not possible to arrive at an unambiguous evaluation of f p or ivp or a usefiil interpretation of rK. For example, with an assemblage of packed spheres, as already noted, the porosity is dependent on the packing density as well as the particle size. Similarly, in the case of a network of intersecting pores, the value of rh is dependent on both the pore radius and the lattice spacing of the intersections. 

The percolation model suggests that it may not be necessary to have a rigid geometry and definite pathway for conduction, as implied by the proton-wire model of membrane transport (Nagle and Mille, 1981). For proton pumps the fluctuating random percolation networks would serve for diffusion of the ion across the water-poor protein surface, to where the active site would apply a vectorial kick. In this view the special nonrandom structure of the active site would be limited in size to a dimension commensurate with that found for active sites of proteins such as enzymes. Control is possible conduction could be switched on or off by the addition or subtraction of a few elements, shifting the fractional occupancy up or down across the percolation threshold. Statistical assemblies of conducting elements need only partially fill a surface or volume to obtain conduction. For a surface the percolation threshold is at half-saturation of the sites. For a three-dimensional pore only one-sixth of the sites need be filled. 

---