Business Clusters

Размещено на http://www.allbest.ru/

Business Clusters

A number of regulatory and strategic concepts designed to increase the efficiency of innovative entrepreneurship, transform the results of innovative ideas and new developments into a sought-after global economy and society into an intellectual product, put on the agenda the realization of the task of a cluster approach to the cooperation of key business entities. In modern science, the practical implementation of this approach is characterized by the controversy of scientific validity and methodological imperfection of the author's positions. The available research results do not provide an unambiguous answer to the question of what the tools for forming “model” clusters should be, based on increased partnership synergistic interaction, especially in the unfavorable external and internal environment of the innovation system. From this point of view, scientists formed models of scientific-innovation clusters which are still independent disparate network associations, which sometimes appear in an uncertain position with respect to each other and, in addition, often do not take into account the life cycle trends of an innovative enterprise that has an impact on the formation of general cluster management logic.

One type of adaptation of innovative enterprises to changing conditions is the administrative and market support of their activities by means of cluster partnerships. Administrative support is to inform innovative enterprises, to create and simplify the conditions for activating their activity, to provide outsourcing services for individual business processes, to get acquainted with and communicate with other market participants (sectoral or territorial), etc. Market support is more effective and motivating, namely the co-operation of small and medium-sized enterprises in a certain socio-economic location. This will allow them to fulfill their primary purpose - to produce competitive products and sell them to large market players whose interest in such interaction is associated with the launch of new products in the markets, the implementation of subcontracting, the activation of innovative processes.

Studies of enterprise activation by means of cluster partnerships and activation of innovative activity are devoted to scientific works of many scientists. From a systematic approach, I.Ablaev (2018) explores the issues of comprehensive analysis of the processes of economic interaction that occur in clusters and identifies the main problems of regional cluster development. Slaper, Ortuzar (2015) identify the issues of forming industry clusters and achieving economic development from network partnerships. Foley, Freihaut, Hallacher, and Knapp (2011) focus on new foundations for the formation of public-private research partnerships. Sheffi, Saenz, Rivera, Gligor (2019) deeply define the forms of cluster partnership, the potential of market conglomerates and determine the mechanisms of horizontal cooperation within the cluster. Beckie, Kennedy, Wittman (2012) investigate the nature and importance of clustering of farmers' markets in the western Canadian provinces, focusing on the possible link between clustering and the "expansion" of alternative food networks. Palyvoda (2015) examines the conditions of formation and sustainable functioning of territorial-production clusters on the example of the leather industry, determines the levels of economic confidence and peculiarities of its impact on the economic environment. The feasibility of using “hub-and-spoke” clusters in low economic confidence is substantiated. Pucci, Brumana, TommasoMinola, Zanni (2020) investigate the impact of different typologies of firm relationships on innovation performance and identify tripartite interaction to produce the effect of producing innovation in territorial relations, relationships and family involvement. Bergman, Feser (2020) identify that industry clusters refer to the close links that bind certain firms and industries together in various aspects of shared behavior, such as geographical location, sources of innovation, joint suppliers, and factors of production, etc. Kutsenko, Meissner (2013) determine that cluster policy is recognized as one of the key elements of modern innovation policy. Scientists are looking at collaborative models for science, technology and innovation with clear goals for value and progress in these areas. The authors determine that an important special form of cooperation is public-private partnership, which also exists in various forms. Laur, Klofsten, Bienkowska, Wincent, Ylinenpдд (2013) deeply explore how cluster initiative participants contribute to cluster initiatives, and determine what dependencies exist between the level of preparedness and engagement of cluster initiative participants. Hnatenko (2018, 2019) defines the methodological foundations of institutional analysis that is necessary to implement a cluster initiative and transform a market economy into a market type of development.

Undoubtedly, the above authors' conclusions are important in the context of the development of the idea of ??innovative clusters formation, but in the mentioned scientific works the problems of innovative entrepreneurship development with life cycle stages are not sufficiently covered, the problem of expedient clustering based on the ideas of self-organization clusters and minimal state impact on these processes is not considered.

The object of the study is innovative clusters as a source of competitiveness, the subject of the study is to manage the life cycle management of the clusters and to use self-organizing neural networks.

In the process of research, we used the following methods of knowledge of economic processes and social phenomena to solve the problems posed in the article: abstract-logical method - to formulate the life cycle of innovative enterprises, which is appropriate to use in the processes of clustering; graphical method - for a visual representation of the life cycle and logic of the neuronperformance.

According to the results of the scientific research, the logic of the foundation of cluster formation can be used by entrepreneurs and statesmen for strategic planning of a cluster in a certain territory, as well as for determining the necessary resources for an innovative enterprise according to the life cycle.

The variety of socio-economic processes occurring in the innovation system is realized through active partnerships between business entities. Among them, innovative enterprises of various directions and spheres of economic activity play a special role. The competitiveness of such an innovative enterprise is one of the most important goals of a market economy and emphasizes the issue of its effective adaptation to the changing conditions of the market environment. Modern adaptation to the conditions of unstable and changing external environment of innovative enterprises should be realized in the process of thorough and continuous analysis of competitors' activities, their own strategic potential and evaluation of their efficiency, as well as determining the position of the enterprise in the market of goods and services with respect to competitors, that is, the evaluation of the enterprise's potential. One of the means to ensure effective enterprise innovation is to create clusters that help businesses survive in an unstable environment.

The basic foundation of our research is the following depiction of the need for the process of enterprise clustering:

study of stages of enterprise life cycle, both within the formed cluster and outside the cluster (stage 1);

formation of cluster grouping by means of neural networks (stage 2);

creation of a competitive innovation cluster (stage 3).

Implementation of the 1^st stage. The logic of engaging enterprises in clusters should take into account the stages of the enterprise life cycle, which we understand as an activity that is expressed in 7 stages (Table 1).

Table 1 An extensive characteristic of the life cycle stage of an innovative enterprise

This characteristic of the stages of the innovative enterpriselifecycle allows to determine how it is necessary to manage the activity of the enterprise in order to effectively involve it in the cluster and to maintain its activity in the long term within the cluster formation at the micro level. The main purpose of planning the formation of an innovation cluster is to systematically monitor the life cycle stage of an innovation enterprise and to select ways for such a cluster self-organization that would allow the clusters to realize the effect of self-coordination with minimal state influence.

Implementation of the 2^nd stage. At present, there are problems of implementation of expedient methods and selection of indicators for the formation of national clusters based on self-coordination, which is complicated by the existence of circumstances of various actions of formal and informal institutions in certain territories, the branching and uniqueness of the factors that determine the basic innovative development of the entities of entrepreneurship, the evaluation of which is difficult due to the uncertainty, complexity and openness of the environment, as well as to asymmetry of information and lack of available quantitative and qualitative data. In addition, the results of the algorithms are highly dependent on many arbitrary choices, such as initial conditions and threshold values.

Currently, there are a number of methods that are applied in the economy to solve the problem of modeling innovative clusters based on self-coordination.

1. Probabilistic approach: K-means, K-medians, EM-algorithm, FOREL family algorithms, discriminant analysis.

2. Approaches based on artificial intelligence systems: C-means fuzzy clustering method, Kohonen neural network, Genetic algorithm

3. Logical approach: decision trees, etc.

4. Graph Theoretical Approach: Graph clustering algorithms

5. Hierarchical approach

6. Other methods:

a. Statistical clustering algorithms

b. ensemble of clusters

c. KRAB family algorithms

d. Algorithm based on screening method, etc.

The choice of a clustering method based on self-coordination depends on the amount of data and whether it is necessary to work with several types of data simultaneously. Each method has its strengths and weaknesses, and none of them claims to be versatile. However, there is an opportunity to provide solutions to the problems posed by the procedure of cluster formation by intellectual means.

It should be borne in mind that the management of the innovation system is focused on finding and choosing the optimal path of evolutionary development and maintaining stability in the future. According to the synergistic approach, the methodology of cluster management consists in the synthesis and analysis of many development paths according to management goals, identification of regions of gravity (attractors) of data of cluster trajectories, barriers on optimal trajectories, and a set of measures to overcome them. Effectively managing the cluster from a synergistic perspective is to create the environment with the best conditions for self-organization and to track these conditions. In the presence of the algorithm of access to the desired attractor there is a cost reduction and time saving. Management should be feedback and learning process. This approach allows to take into account the dynamic aspect of the development of innovative entrepreneurship system through research of such categories of economic activity of participants as complexity, openness, uncertainty, self-organization, etc. Identification of attractors in the field of spatial development allows to interpret the change of the system in the context of building interactions of innovative agents in a branched innovation environment.

Determination of order parameters and influence factors that determine the behavior of system components allows us to synthesize models and methods of predictive modeling of system dynamics in order to select strategies for its evolutionary cluster development.

Commonly used methods of modeling nonlinear dependencies between variables are the fit of curves (in particular, quadratic, logarithmic, cubic, hyperbolic, static, exponential, logistic, etc.), the use of dummy variables, and others. However, it should be recognized as a more promising method of modeling nonlinear dependencies in order to solve modeling problems using the neural network approach. Modeling the creation of an innovation cluster based on a neural network approach using machine learning technology will provide an opportunity to generalize the accumulated knowledge, produce their training, to use to determine the current level of innovative development and predict the future state of the system.

Neural networks are by far one of the most well-known and effective tools for intelligent grouping of data based on self-coordination, in our case the clustering of enterprise clusters. The logic of cluster grouping by means of neural network will be described in details. Any artificial neural network consists of relatively simple, in most cases the same, functional elements that imitate the work of brain neurons. Each neuron has a group of synapses - unidirectional input connections connected to the outputs of other neurons, and also has an axon - the output connection of a given neuron, from which a signal (excitation or inhibition) arrives at the synapses of subsequent neurons (Fig. 1).

Each synapse is characterized by the magnitude of the synaptic connection or its weight - (the weight of the connection of a neuron iwith a neuron j, which in physical content is equivalent to electrical conductivity. The current state of the neuron is defined as the sum of n inputs weighted with its synaptic weights:

(1)

where the sum to the right is interpreted as the state of the neuron or the neuroactivity of the neuron.

Fig. 1. Schematic representation of the neuron (left) and the type of activation functions (right: a - single jump function; b - linear threshold (hysteresis); c - pygmoid - hyperbolic tangent; g - sigmoid

The output of a neuron is a function of its state: У = f (S).The nonlinear function f is called the neuron activation function. One of the most commonly used activation functions is a non-linear saturation function: the so-called logistic function, or sigmoid (S-shaped function):

 (2)

As parameter б decreases, the sigmoid becomes flatter, in the limit when б = 0 degenerates into a horizontal line at the level of 0.5, as б increases, the sigmoid approaches in appearance to the single jump function with a threshold equal to 1 at the point X = 0. From the expression for the sigmoid it is obvious that the output value of the neuron lies in the range [0, 1]. It should be noted that the sigmoid function is differentiated across the entire abscissa, which is used in some learning algorithms. In addition, it has the ability to amplify weak signals better than strong ones, and prevents saturation from large signals.

The neurons collected in a system of a particular architecture of this system are called neural networks, the type of architecture of this system is determined by the types of tasks that are planned to be solved through the network. Anyway, the architecture has an input and an output layer (data is fed to the input layer and after processing the network is obtained from the source layer). Learning a neural network is called changing the synapses (weights of each neuron) of the network. In this case, training can be "with a teacher" or "without a teacher." In the first case, a "reference" output pattern is recommended for each input pattern of the network, and the difference of the response of the network with the "reference" pattern is adjusting its outputs, adapting to the inputs to its input signals. Only the data itself can serve as a "teacher" of the network, that is, they have information, regularities that distinguish the input from random noise. In this case, unlike teaching with a teacher, the problem is solved not the minimization of the target functional (network error function), but the optimal encoding of information into prototypes. Of course, as with any training, it is necessary to formulate the rules of learning. Self-organizing networks are a type of neural networks that are taught without a teacher and can be used to solve our problem (Fig. 2).

Fig. 2 Schematic representation of a self-organizing network for clustering

The principles of functioning and learning of a self-organizing networkwill be detailed. It should be noted at once that the dimensions of the input vector and the vector, composed of the weights of the output layer neurons, coincide. This is obvious that each neuron of the source layer is connected to all neurons of the input layer. The vector of the weights of the i-th neuron will be denoted by .

The algorithm of the so-called training competition is used for training the network. This algorithm will be briefly described. With this training, the outputs of the network are as highly correlated as possible: for any input value, the activity (or vector length ,) of all neurons except the so-called winner neuron is equal to zero.

This mode of operation of the network is called "the winner takes everything". The winner neuron (with index weight and weight *) is for each input vector X. Therefore, the winner is chosen so that its vector of the weight*, defined in d - dimensional space of the input data, is closer to the given input vector X than all other neurons:

| X |, ? і *, де \ а \ =

- the usual Euclidean norm of d-dimensional space). In case, as it is usually done, of application the training rules for neurons that provide the same normalization of all weights, such as | | = 1, then the winner will be the neuron, which gives the largest response to this input "stimulus". The output of such a neuron is amplified to a single one, while others are suppressed to zero.

The number of neurons in the competition layer (in this case, the competition layer coincides with the output) determines the maximum variety of outputs and is selected according to the required degree of detail of the input information. The trained network can then classify the inputs: the winner neuron determines which class the given input vector belongs to.

Unlike teaching with a teacher, self-study did not imply the a priori task of class structure. The input vectors should be broken down into categories (clusters), consistent with the internal patterns of the data themselves. This is the task of self-organized learning of the competitive layer of neurons.

According to the Oja rule, =?) where is the change in the weight of the i-th neuron when presented with the r-th example, is the input vector, is the response of the i-th neuron to the r-th example, and ? is the learning rate. This is the basic algorithm for learning the competitive layer. In accordance with the above rule, only the weights of the winner-neuron are adjusted, since it alone has a non-zero (single) output. Then for the winner, the learning rule takes the following form: = ?-).

In 1982, T. Kohonen proposed to includeinformation about the location of neurons in the source layerinto the basic rule of competitive learning. For this purpose, the source layer neurons are ordered to form one- or two-dimensional lattices, i.e. the position of the neurons in such a lattice is indicated by a two-component vector index i. Such ordering naturally introduces the distance between neurons | | in a layer. The Rule of Competitive Learning modified by Kohonen takes into account the distance of neurons from the winner neuron:

=?Л(| |)() (3)

The neighborhood function Л(||) is equal to one for the winner neuron with the index and gradually decreases with distance, for example, according to the law Л(а) = ехр (- / ), where у is called the learning radius. Both the pace of training ? and the radius of learning of neurons у gradually decrease in the learning process, so that at the final stage of training we return to the basic Oja rule of adaptation of the weights only of the winning neurons. With such an individual adaptation of prototypes (neuron weights), Kohonen training is reminiscent of stretching an "elastic grid" of prototypes onto an array of training sample data. As the training progresses, the elasticity of the grid gradually increases, so as not to interfere with the final (fine) adjustment of the neural network weights. A convenient tool for visualizing data on clustering is to paint topographic maps in the same way as it is done on regular graphics maps.

Therefore, it is proposed to use self-organizing (growing or evolving) neural networks to solve the problems of cluster self-coordination. Using self-organization principles allows to synthesize multilayered neural networks on a part-time, non-representative training sample. For neural network synthesis, a minimum number of errors in the training sample is provided, so it is not necessary to estimate in advance the significance of the input variables (order parameters) and to determine synaptic connections.

The algorithm of functioning of the training network is one of the variants of the implementation of expedient multidimensional vectors of cluster creation. An important difference of this algorithm is that all neurons are arranged in some structure inside it. In many cases, the use of self-organizing systems (neural networks) is better than traditional fully connected neural networks. The Kohonen network is formed by two layers: an input and an output (the result of which is a topological map). Data is fed to the input, and output neurons are organized with a simple neighborhood structure. Each neuron is associated with a reference vector (weight vector), and each data point is "mapped" to a neuron with the "closest" (in terms of Euclid distance) reference vector. In the process of performing the algorithm, each data point acts as a training sample that directs the motion of the reference vectors toward the data values ??of that sample. Vectors associated with neuronswhich are called weights, change in the learning process and tend to have characteristic input data distributions.

The use of a neural network will allow the means of self-coordination to group the necessary input parameters, which will determine the appropriate cluster groups.

Implementation of the 3^rd stage. There is a need to improve cluster planning, taking into account not only the tendencies of the scientific and innovative sphere, which has an impact on innovative entrepreneurship, but also changes that occur at the stages of the enterprise life cycle and the functioning of dominant innovative enterprises. In this sense, the organization of clusters involves considering the stages of the life cycle, which will allow to stimulate positive change and promote fair competition within the innovation cluster. Given this, in our study, it is advisable to implement the formation of a cluster based on the French approach "competitive pole" (from the French - Pфles de compйtitivitй), the formation of which in our case is based on the partnership of stake holders of innovative activities with enhancing territorial resources where a cluster functions. We consider it expedient to depict the exposure of creating an innovative cluster based on the "competitiveness policy" approach (Fig. 3).

Fig. 3 Exposition of creation an innovative cluster based on the “competitiveness policy” approach

Summarizing the above, the key idea of ??organizing the creation of innovation clusters should be formed through the prism of the existing life cycle of the enterprise, within which collective productivity, innovative shifts, competitive incentives, efficient models of technology penetration and production of innovative ideas will be enhanced, which will provide competitive advantages at all levels. We propose the organization of a training model of an innovation cluster or an "ideal" cluster by neural networks, whereby the transfer of intellectual knowledge, the creation of innovations, the exchange of experience and the involvement of labor market reserves are fundamental to enhance competitiveness at all levels, rather than trade and business. This idea also takes into account the dense combination of science, business and power (the triple helix) as the most effective form of partnership.

Conclusion

enterprise competitiveness policy cluster

In the course of the conducted research it is suggested to investigate the stages of the enterprise life cycle as the main basis for the effective development of the enterprise both within and outside the cluster. It is determined that the process of formation of innovation clusters should be based on the use of neural networks, which in the best way take into account the idea of ??self-organization of complex innovation systems with limited input parameters. Particular attention is paid to the formation of clusters based on the approach "competitiveness policy", which should be the final logic of a comprehensive cluster planning strategy. Particularly important in the context of our study is the use of a neural network, through which cluster organizations based on self-coordination can be created. The degree of state intervention in such a clustering process should be minimal and based solely on artificial intelligence.

Размещено на Allbest.ru