Prior to the SIENA project it was already recognised that, although Europe is strong in research into this area of technology, it may lag behind the USA and Japan in its industrialisation. The objectives of the project were therefore set in order to examine the current state of commercial ANN usage across Europe and to understand the information dissemination routes which could excite business interest in this topic. The project acronym, SIENA, is derived from Stimulation Initiative for European Neural Applications.
The project has received support from the European Commission under its framework III programme. However, unlike most of the previous ESPRIT projects in this field (of which there have been just over 30) it has not involved issues of technology development. Instead, SIENA has focused on the way in which 'known' technology can be deployed commercially. In some ways, the state of many of the previous ESPRIT projects reflects the concerns which originally prompted SIENA: high quality research but limited follow-through into real-world application.
The project was conducted over a period of 18 months by a consortium of six organisations - all Small and Medium Enterprises (SMEs) - drawn from five countries. The total project budget was 600 kECU. Half of the funding was provided by the European Commission, the remainder by the consortium members. The geographic spread of the partners allowed a reasonably broad coverage, but this was necessarily limited by the available budget (for example, the SIENA budget for the whole of Europe is only a small fraction of the amount required for a comprehensive project in just one country). Within this over-riding constraint, we believe that we can discern a fairly consistent picture of the current European situation.
This document is a summary of one of the major deliverables from the SIENA project. The full 'white paper' gives an expanded description of the current European position in the commercial use of ANNs.
(Note for non-European readers: financial figures are given in ECUs - European Currency Units. One ECU is equal to approximately 1.27 US Dollars.)
The following table summarises the organisations in the project. Each acted as a nucleus for a particular region of Europe, as shown in the table.
| Organisation name | Country | Geographical Region |
|---|---|---|
| Augusta Technology Ltd (ATL) | UK | UK, Eire, Scandinavia |
| Instituto de Ingeniería del Conocimiento (IIC) | E | Spain, Portugal |
| Neural Computing Applications Forum (NCAF) | UK | UK, Eire, Scandinavia |
| Neuroptics Consulting | F | France, Italy |
| Stichting Neurale Netwerken (SNN) | NL | Benelux |
| Zentrum für Neuroinformatik GmbH (ZN) | D | Germany |
The major effort in the project has been directed to surveys of the existing ANN market and dissemination of information on the application of ANNs. The survey work was split into two elements to give two different perspectives on the market: suppliers and users. Contacts were established with supply-side companies around Europe (by the appropriate regional partner) to collect information on their products and services and to obtain an insight into the market from a supply perspective. Similarly, contacts were established with representative users to obtain information on how and why ANNs are being deployed.
Generally speaking, the coverage of the supply side represents a good sample of the industry. We estimate that we have probably identified at least 40% of all European suppliers active at the time of the survey, and a much higher percentage - probably over 80% - in some regions such as the UK, Spain and Benelux. The number of users is obviously much greater than the number of suppliers, probably by a factor of 10 to 20 at present, and so the survey covers only a relatively small sample from the total range of users around Europe.
Dissemination activities included the collection of materials to support the spread of information on ANNs, the presentation of material at events such as conferences and workshops, and publication of information. A particularly valuable result is the availability of a set of high-quality case studies.
The precise mix of the above varies from one supplier to another, as does the form in which the product or service is packaged for sale. The figure below shows in diagrammatic form a model of the supplier relationships which are relevant to SIENA. We are most concerned with the primary supply of ANN technology, shown by the solid lines in the figure.
Most of the supplier companies (excluding academic organisations) can be classified into one of a small number of categories:
The following graphs give an overview of the breakdown of suppliers by country, by industry sectors and by application types. The full white paper gives a more extensive analysis of supplier data.
An alternative view is presented, if we split the data into industry sectors.
One of the most important activities related to dissemination has been the collection of a large number of case studies of successful ANN applications. Case studies are a particularly valuable tool to encourage businesses to take up ANN technology. Cases have been collected covering a variety of ANN applications across a broad range of sectors. Each has been arranged in a standard presentation format to make it easy for potential users to locate relevant case studies and to assess the potential value of ANNs within a particular business context. We have taken care to ensure that only 'real' cases are included: many have been rejected on the grounds that they were only experimental prototypes or based on marketing 'hype'.
In addition to general information about the application and contacts, each case study tries to answer the following questions.
In order to provide direct interaction with the potential market a number of seminars, workshops and other kinds of briefings were organised. A number of publications were also produced, including electronically over the Internet.
The closest comparison to these figures is given by a 1992 report by Frost and Sullivan. At that time, the 1996 world-wide market for neural networks was predicted to be around 1000 MECU, with an annual growth rate of 45%.
Across the EU countries, the market breakdown is not quite in the same proportions as the ratio of each country's GDP to the EU total. The market is more developed in some areas than others, for reasons such as national programmes designed to stimulate interest in ANN technology. Taking these factors into account we have arrived at the following estimate of the proportion of the total EU ANN market in each country.
| Country | Percentage of EU GDP* | Percentage of EU ANN market |
|---|---|---|
| Austria | 2.6% | 2.1% |
| Belgium | 3.0% | 2.5% |
| Denmark | 2.0% | 1.9% |
| Finland | 1.2% | 2.0% |
| France | 18.2% | 14.8% |
| Germany | 27.4% | 31.1% |
| Greece | 1.1% | 0.2% |
| Ireland | 0.7% | 0.1% |
| Italy | 14.6% | 9.4% |
| Netherlands | 4.5% | 5.2% |
| Luxembourg | 0.2% | 0.1% |
| Portugal | 1.1% | 0.2% |
| Spain | 7.0% | 4.5% |
| Sweden | 2.7% | 2.6% |
| UK | 13.8% | 22.3% |
| Total | 100% | 100% |
percentage values are rounded (* based on 1993 figures)
Remarkably, the level of neural activity in Belgium is much lower than in The Netherlands. This is confirmed by several Belgian experts whom we have contacted. One can only speculate upon the reason. One more or less cultural explanation which has been given is the technological stimulus caused by the traditionally large R&D laboratories of Dutch multinationals.
It is interesting to consider the relationship between the general characteristics of the Dutch economy, and the results of the survey and case studies. The Dutch market could be characterised as follows. From the sector view point, there are relatively few large high-tech manufacturing companies (car, aircraft, consumer electronics, defence and so on). On the other hand, there is a significant chemical industry and a large transportation sector. In addition, there are many service-provider businesses, such as financial institutions and management consultants.
This picture is well reflected in the survey. For instance, the case studies from AKZO (and in a sense SHELL) represent the large chemical industry. In the services sector, management and marketing consultants are clearly present in the survey. In addition it is known that neural technology is heavily used in the Dutch (and Belgian) financial sector. However, organisations in this sector are reluctant to provide information for obvious reasons, and as a result this sector is not well covered by the survey. Automated handwriting recognition is another example of an application used by service-providing businesses. Finally, typical Dutch examples are the applications in water management and in the high-tech agricultural sector.
With regard to the size of companies, the Dutch economy comprises a small number of very large multinationals and a large number of small companies. Medium size companies are under-represented compared with other countries. The effect on the neural network market is two-fold. On the one hand, most multi-nationals (such as AKZO and SHELL) have developed their applications in-house, using their own expertise or in collaboration with universities or large high-tech research institutions. These activities have already existed for a long time and are well developed, since it is important for these industries to maintain a technological lead. These companies are well aware of the existing technology. For small companies, on the other hand, the technology is still too expensive.
The principal market for the application of existing technology may be found in medium sized companies (e.g. SMIT, Telegraaf), where the survey shows a moderately growing market. This situation is clearly reflected in the supply side. This consists on the one hand of some well-established large high-tech contract research organisations and the usual large software houses. For these organisations neural networks are only a very small part of their total activities. On the other hand, specialised neural network application providers are actively present, but they are limited in number and small in size.
There seems to be a correlation (not necessarily a causal relation) between neural activities in a country and national stimulation programs. For instance, in Belgium such programs are hardly existent. In the Netherlands stimulation programs are on a moderate level: In addition to existing technology transfer programs the following specific neural networks initiatives were taken by the Dutch government. In 1990 a national research program on neural networks called SPIN (1.7 MECU in 4 years) was initiated by the Ministry of Economic Affairs. Since 1994, STW (the Technology Foundation) has organised a bi-annual call for proposals to fund neural networks research projects in which universities and industries collaborate.
Some clear conclusions emerge from the regional survey. Multinationals and other large companies should clearly be encouraged to keep their lead in neural network technology. This could be done by providing national or European support of their long term R&D projects. In this context, the current Esprit initiative may be too much oriented towards applications in order to maintain the European competitive edge on a global level. For smaller companies, on the other hand, these novel R&D developments are probably out of reach for over 10 years. For them, only proven neural network technology is of interest. The lower-end neural network technology is sufficiently mature to be used by all companies. The best place for a technology to prove itself is in the market. To stimulate the market, the governments could support end-users in the take-up of such technology, or offer in some way to compensate for the risks. Care should be taken against unfair competition or other mechanisms which could distort the internal market. Other mechanisms to stimulate the market include awareness programs. In these, care should be taken against over-optimistic promises, which could have disastrous and irreversible effects.
Finally, an important point to note is that for successful application of the current standard neural network technology - in contrast to standard statistics - still a lot of expertise and testing is needed. Application of neural networks is still often considered as an 'art'. As a consequence many SMEs hesitate over starting neural network projects. The solution to this problem is to automate the application of neural networks themselves as far as possible. The problems herein are often at a fundamental and deep scientific level. In the Netherlands some scientific-technological projects to solve these problems have been started. We recommend that these research projects be organised on a European level.
This difference is also reflected in the survey of the ANN applications market. A first point to consider is the absence in both countries of any official support for ANN research and development, let alone for the introduction of these technologies into the market. There are certainly neural product suppliers and neural applications in use in this area, but the market awareness of the concrete or potential benefits of ANN technology is low at present.
The market survey performed within the SIENA framework tends to support this view. To begin with, 22 end users have been identified in Spain, and 2 in Portugal. These numbers are certainly low in comparison with those of other European countries (of course, there are cases, especially in Spain, of companies that actually use neural products although they fail to acknowledge it). Taking into account these matters, the questionnaire mailing to end users was carried out in two phases. In the first phase, a large scale mailing was done to the person in charge of technology issues in more than 130 companies, in the covered area, followed by a detailed questionnaire to the companies that showed an interest. Some of the companies that requested the second questionnaire did not answer it, and they could be considered as potential rather than actual end users, but in any case their number was in the above range.
Concerning suppliers, the above considerations are reflected in the small number of companies identified. Taking university departments out of the analysis (as was agreed by the SIENA consortium), 14 suppliers have been identified in Spain and just one in Portugal. Moreover, some of these were actually system integrators or value-added resellers of neural products developed by other companies.
An interesting fact obtained from the analysis of the data base is that nearly all companies contacted began their ANN activities around 1990, although the bulk of products did not start selling until 1994. This implies a period of several years of development for ANN technology based applications or products. A parallel between annual ANN turnover and number of ANN experts can also be highlighted. An analysis has led to the conclusion that those companies with an annual NN turnover of about 150 kECUs have less than 5 NN experts and those companies with a turnover in the range 150-300 kECUs have between 5 and 20 NN experts.
The main area of neural technology utilisation by far was OCR, in which products of neural development companies were part of larger developments done by regional companies. It can be certainly said that the majority of OCR undertakings in Spain have at their core a neural recognition engine. Another area of relative ANN penetration was industrial modelling and process forecasting and control. A third area of interest is the banking and finance industry, either as tailor-made developments, or as data segmentation and mining tools. Although some success stories do exist, it is clear that the neural market both in Spain and in Portugal is far from either being mature or having more than a moderate revenue size.
From the example of the OCR market we can conclude that ANNs may have a business niche in the Iberian peninsula, but further efforts will be needed to create a larger commercial awareness of the potential advantages of the available technology.
In this direction, SIENA has certainly represented a first important step to capture the market picture in both countries. More efforts should, however, be made both in more precise market analysis and in dissemination activities to increase the awareness of working ANN applications and benefits, not only in big corporations, but also among SMEs. The materials and results of SIENA will certainly be of great help to make sure that companies realise the benefits of more readily disclosing their neural activities, and take further interest in these technologies.
On the user side, it is clear that German industry is increasingly taking up neural networks. This is done even more readily if it can be demonstrated that there is added value to existing products or production procedures. Here, especially larger companies have their own internal labs that keep their work confidential. As far as the competitive situation in high technology segments is concerned, this results in a severe shortfall of Small and Medium Size Enterprises (SMEs) that have the capability to evaluate and introduce a new technology such as neural networks. This is reinforced by the fact that German companies tend to be quite conservative when introducing new technologies especially when compared to Japan or the US. Clearly, additional work in demonstrating the usefulness of neural networks would be beneficial.
Nevertheless, the companies working with neural networks have experienced a considerable growth in the last three years which also reflects the strong interest on the industry side for mature and well proven ANN technology.
Activity is therefore displaced from commercial ANN suppliers to these quasi-governmental research bodies, which may partly explain why the French market appears smaller than would otherwise be expected.
A very similar position exists in Italy. The National Research Council funds bodies such as IESI-CNR (Image and Signal Processing Institute) and the Istituto per la Ricerca sui Sistemi Informatici Paralleli in Naples. Agencies such as IRST (Istituto per la Ricerca Scientifica e Tecnologica) and ENEA (Ente per le Nuove tecnologie, l'Energia e l'Ambiente) have significant activities related to ANNs. Amongst the large companies we find CSELT (Centro Studi e Laboratori Telecomunicazioni S.p.A.) and ENEL S.p.A., the Italian National Electricity Company, engaged in development work related to their particular industrial concerns. We also know that some major manufacturers in fields such as domestic appliances are interested in adding neural components to their products, but we have the usual problem of a lack of material which can be quoted publicly.
The supply side in Italy appears to be less developed than in France at present, with only a few independent companies being identified. We suspect that the reasons for this correspond closely to the equivalent situation in France.
The UK suppliers represent a significant proportion of the data base, accounting for about 30% of the total. The number of suppliers is partly a reflection of the greater level of awareness, but also the general climate of activity in the UK which has a bias towards commercial activity rather than publicly funded research agencies.
As with other regions, it has not been practical to attempt anything more than a sampling of users, but the total clearly exceeds several hundred in number. For example, the Neural Computing Applications Forum (NCAF) has over 150 members, and this represents only a fraction of the UK user base. The number of companies who registered interest with the DTI NCTTP was around 5000, but we have applied rather strict definitions of 'user' in this study.
In contrast to the situation in the UK, the ANN market in Ireland appears to be very low key at present. Although good quality research is being conducted at Universities and elsewhere, which provides an appropriate scientific base, there appears to be little of the demand from business which would be needed to sustain an identifiable ANN market. At the time of the survey we were unable to identify any specialist commercial ANN suppliers based in Ireland, and most of the application effort seems to be limited to the laboratories of large multinational companies. The response to a large-scale mailing to companies in Ireland carried out in conjunction with a SIENA seminar held in Dublin indicated a low level of awareness of the relevance of ANN technology to business applications. Potential suppliers therefore face not only the usual difficulties involved in marketing any product or service, but also need to take on a 'missionary' role in promoting the technology itself.
The general Scandinavian market is intermediate between the UK and Ireland in terms of activity. Several specialist suppliers have been identified, and there are a number of commercial applications of the technology. Significant industry sectors include marine engineering, and agriculture. Again, this reflects to a large extent the focus of major national industries such as wood pulp and paper in Finland.
The research base in Scandinavia is of high quality, and in some academic sites can justifiably claim world leadership in their field. However, we gained an impression that the gulf between academics (and other researchers) and indigenous commercially-oriented companies is somewhat wider in Scandinavia than in some other parts of Europe. Insufficient resources were available to attempt an investigation of the apparent discrepancy between the relatively high level of research activity and low level of commercialisation. This could be due to a variety of reasons (some of which are discussed in the full White Paper) and it may be useful to look at the paths to commercial exploitation in depth in some future study.
Finland occupies a somewhat different position to the other Scandinavian countries because it has embarked on a government-sponsored initiative entitled Adaptive and Intelligent Systems Applications. This is aimed at improving and developing production methods and products incorporating neural computing and fuzzy logic. The initiative is being channelled through TEKES, an agency established by the Ministry of Trade and Industry to coordinate and finance applied technical and industrial R&D.;
Applications are envisaged in industry, commerce and banking, particularly in sectors with high importance to the Finnish economy such as pulp and paper, communications and chemicals. The programme is collaborative, involving over 80 companies and 14 research organisations. The budget for the programme is 18 MECU, and it is scheduled to run from 1994 to 1998.
Some of the project work has been hampered by the difficulties of obtaining information. For example, we originally expected that all suppliers would be keen to participate. In practice we encountered a range of attitudes ranging from complete frankness to secrecy and information quality ranging from precise to vague. This applied even more strongly to some classes of user. Inevitably we have had to supplement the bare facts with our own knowledge to assemble a reasonably comprehensive picture.
At various stages we encountered the need for careful definition in order to produce sensible classifications for industries, sectors, hardware, suppliers and various other things. Understanding these definitions is crucial to understanding the results of the study and comparing information from this project with other surveys and reviews of the market.
The European commercialisation of ANN technology is now clearly under way. However, it is also clear that both the absolute size of the market and the rate of take-up lag behind the USA which is the leading area in this field for application, although not necessarily in research.
European companies appear to face greater obstacles than their opposite numbers in the USA. For supplier companies, the largest single problem is easy access to the capital needed to fund innovation. User companies appear in general to be rather more conservative than their transatlantic counterparts, and less aware of the ability of technology to make a significant contribution to the success of their business.
These problems are not insoluble, and the impact made by some national programmes indicate ways in which the potential could be unlocked.
We are aware that in many areas we have only scratched the surface of the European ANN market. In order to provide a more comprehensive view of the market we recommend that the impetus gained through this work should be continued in an appropriate follow-on project.