|The intelligent software of Intellix has a theoretical foundation in the fields of artificial intelligence (AI) and machine learning. These disciplines are investigating how to incorporate intelligence into computer systems and how to make computer algorithms learn by experience. The products are based on a core of patented technology (see more on Intellix patents here).|
This is a relatively new and very research-intensive scientific field and Intellix is among the first companies in the world, to put the methods and results of these research areas to common practical use. The result is a line of products being able to map and distribute knowledge in a unique way, enabling our customers to provide their customers, partners and employees with advice, decision support, and self-service solutions on even the most complex and complicated problems.
The vision of Intellix is to provide technology for supporting intellectual business processes. In order to achieve this goal, we have developed the SOUL kernel (Self-Optimising Universal Learner) that is able to map human knowledge and business policies into a form applicable for computation. The SOUL kernel is a complex inference engine build using a number of patented algorithms, e.g. neural networks, originating within the fields of machine learning and AI. The principles underlying the SOUL kernel are a result of research performed at Risø National Laboratory in the mid-1980'ies and the technologies have been improved ever since. Our aim is to constantly improve the SOUL kernel through research and collaboration with academia like Risø National Laboratory.
In order to map the knowledge required for a given business process within an organisation, it is important that the system can acquire both explicit and tacit knowledge. Explicit knowledge is knowledge that can be articulated or communicated in writing. An example of explicit knowledge is a business policy, stating clearly how a given situation should be tackled. Explicit knowledge can be mapped directly in the SOUL kernel by specifying the corresponding rule-base in a so-called knowledge domain. Besides explicit knowledge, business processes can also include the intangible form of knowledge present in the minds of employees - tacit knowledge. Due to the tacit nature, such knowledge cannot be expressed explicit, but it can be transferred between individuals in a learning process. An example of tacit knowledge is the mental model that provides a stock-dealer with intuition. Although tacit knowledge cannot be expressed explicit, it can participate in the process of generating new, possible explicit, knowledge.
Figure 1 Knowledge mapping using the Intellix Designer. During use of the system, the knowledge flow is reversed, and the end-user can learn from the system
Figure 1 illustrates the technologies used by, and in combination with, the Intellix Designer and its SOUL kernel when mapping knowledge. The knowledge engineering method used by Intellix facilitates the conversion of tacit knowledge into explicit knowledge that easily can be mapped by the Intellix Designer. The SOUL kernel is also able to learn tacit knowledge from a set of examples. As a human trainee can acquire an intuitive feeling for a specific problem by studying examples of the decisions made by a skilled member of staff, the SOUL kernel is able to acquire a similar quality by generalising from the knowledge contained in a set of examples upon which it is trained. An important property of the SOUL kernel is that it can acquire and learn new knowledge in real time and that it is possible to create mixed domains, containing both explicit and tacit knowledge. This implies that the process of knowledge mapping can be made interactive through the "active knowledge acquisition" technique, dramatically reducing the time needed to map a given domain as the system by selective queries can acquire knowledge from experts. This query process is guided by a unique ability of the SOUL kernel to provide sound online quality measures of its internal mapped knowledge.
By use of the Intellix Knowledge Server the knowledge present in a knowledge domain can be distributed internally in an organisation and/or to end-users obtaining self-service on a company's homepage. It is furthermore possible to make the system query the end-user though an "intelligent" dialog, where the system automatic rank the information needed in order to make a decision. This effectively shortens the users dialog with the system without compromising the precision of the decision made.
The SOUL kernel
The architecture of the n-tuple classifier has resemblance with decision trees, neural networks as well as architectures for case-based reasoning. In fact, the n-tuple classifier can be seen as a constrained neural network having a number of neurons highly exceeding what normally is seen for neural networks. Furthermore, a trained decision tree can be seen as a special instance of an n-tuple classifier. Most algorithms used for building data-driven models are so-called strong models, trained with the purpose of having as high accuracy on the available training examples as possible. The n-tuple classifier is taking a radically different approach: Instead of building a single strong model, a high number of "weak" sub-models are combined into an ensemble by a voting process. An ensemble is known to outperform single models - even if the single model is "strong" and the ensemble is combined from weak models. The sub-models of the n-tuple classifier are characterised by a very simple design operating in a sub-space of the full input space.
From the native n-tuple classifier, Intellix has developed and patented extended methods that are used as corner stones in the SOUL kernel. By keeping the attractive properties of the native n-tuple classifier, it is still possible to access quality measures that allow optimisation of the overall model with little or no user-intervention. Important quality measures are the leave-one-out cross-validation and extensions hereof based on the number of examples that supports a given decision. Similar quality measures are computational expensive to obtain for algorithms like decision trees or neural networks.
|© 2012 Intellix. All rights reserved.