## Emergence and Complex Systems — Towards a New Science of Industrial Automation (2003)

Venue: | Proceedings of the 4th International Conference on Intelligent Processing and Manufacturing of Materials (IPMM’03), May 18–23, 2003 |

Citations: | 4 - 1 self |

### BibTeX

@INPROCEEDINGS{Hyötyniemi03emergenceand,

author = {Heikki Hyötyniemi},

title = {Emergence and Complex Systems — Towards a New Science of Industrial Automation},

booktitle = {Proceedings of the 4th International Conference on Intelligent Processing and Manufacturing of Materials (IPMM’03), May 18–23, 2003},

year = {2003}

}

### OpenURL

### Abstract

Complex systems theory promises to solve all problems, big and small – but, concretely, what would such a solution mean in the first place? Specially, what would the automation environment of the future look like? The following presentation introduces these issues, trying to reach a “systemic view ” of complexity.

### Citations

4252 |
Artificial Intelligence: A Modern Approach
- Russell, Norvig
- 1995
(Show Context)
Citation Context ...odelled became larger, the decentralisation of control within the formalisms continued, and different kinds of object formalisms were developed (C++, Java). Today, it seems that in the agent paradigm =-=[19]-=- this decentralisation is being taken to its extreme: No centralised control any more exists. In the specialpurpose formalisms, for example in simulation languages the same tendencies can be seen: Fir... |

1715 | Independent Component Analysis
- Hyvärinen, Karhunen, et al.
- 2001
(Show Context)
Citation Context ...some kind of average model that is really good in none of the modes. As motivated by the independent component analysis (ICA) research, the physically meaningful features typically are not orthogonal =-=[7]-=-. It can be claimed that sparse components are best suited for modelling of natural, multimodal data representing a Gaussian mixture model. To conclude: A natural data distribution is multimodal, cons... |

459 |
Linked: The New Science of Networks
- Barabasi
- 2002
(Show Context)
Citation Context ...urther based on Aristotelian taxonomies with strict organisation among the entities. The theories beyond agents and agent networks are supported, for example, by the studies by Albert-Laszlo Barabasi =-=[1]-=-, and also by John Doyle and collaborators [5], where they show that new approaches are needed to master the robustness properties in decentralised but mutually interconnected devices. As an extreme c... |

135 |
A New Kind of Science (Wolfram
- Wolfram
- 2002
(Show Context)
Citation Context .... The “New Science” During the last year the ever so turbulent arena of complex systems has been in a total turmoil. Stephen Wolfram, a rascal in the academic circles, published his long-awaited book =-=[21]-=-, claiming that now it is there – the complete theory of everything, the New Science, explaining all systems in terms of cellular automata. However, the research of complex systems has a longer histor... |

127 |
Statistical Factor Analysis and Related Methods
- Basilevsky
- 1994
(Show Context)
Citation Context ... basic approach is principal component analysis (PCA): The features are the most significant eigenvectors (as measured by the corresponding eigenvalue) of the data covariance matrix (for example, see =-=[2]-=-). This method is mathematically efficient and there are various technical benefits – for example, the feature basis determined by the principal components is automatically orthonormal, making it easy... |

93 |
Fuzzy Thinking: The New Science of Fuzzy Logic
- Kosko
- 1993
(Show Context)
Citation Context ...other constructs are continuous-valued. The links between units are quantitative rather than qualitative – connections are not “yes” or “no”, but, indeed, the fuzzy subclass structure is spanned (see =-=[17]-=-). This means that it is not only the subclass that inherits properties from its superclass, but the superclass properties are also affected, to a lesser extent, when new subclasses are introduced. Mu... |

31 |
The End of Science. Facing the Limits of Knowledge
- Horgan
- 1996
(Show Context)
Citation Context ...t another word for logical thinking – the risk is that without consistent conceptual tools everything is just handwaving. Is the New Science just a form of “ironic science”, as John Horgan has put it =-=[6]-=-: The field of chaoplexity is full of promises, with much less concrete results. Scientific salesmen have been selling miracle cure for all problems. All possible breakthroughs have already been distr... |

7 |
On mental images and “computational semantics
- Hyötyniemi
(Show Context)
Citation Context ... can be identified with data clusters (centers of relevant data) and the attributes with the degrees of freedom within the cluster (typical variations around the nominal prototype) – for example, see =-=[8]-=-. Structure cannot be deduced when only data is seen. However, because the clusters/categories determine the structural entities, within the selected framework structure can be learned from data, at l... |

3 |
From Intelligent Models to Smart Ones
- Hyötyniemi
(Show Context)
Citation Context ...em is not so bad: One is not totally lost in the jungle of structureless data. There exist some structural properties that apply, and special assumptions about the distribution properties can be made =-=[9]-=-. 1 m (7)Feature extraction and pattern recognition have been extensively studied in the field of computational intelligence. What makes the current task somewhat different, is that the assumed under... |

3 |
Life-like control
- Hyötyniemi
- 2002
(Show Context)
Citation Context ...es the partial data, and reconstructs the unknown variables (outputs) by mapping this internal representation back to external representation χ, the unknown entries being automatically filled in (see =-=[14]-=-). Because parameter estimation is now based on PCA type subspace-oriented schemes, one does not have to be so careful about explicitly minimising the data dimension. For example, if there is some add... |

3 |
third edition). Sciences of the Artificial
- Simon
- 1996
(Show Context)
Citation Context ...model construction. There are also philosophically deeper motivations for the practical programming languages. The current object-oriented formalisms are an implementation of the Simonian hierarchies =-=[20]-=-, being further based on Aristotelian taxonomies with strict organisation among the entities. The theories beyond agents and agent networks are supported, for example, by the studies by Albert-Laszlo ... |

2 |
revised edition). General System Theory: Foundations, Development, Applications. George Braziller
- Bertalanffy
- 1976
(Show Context)
Citation Context ... how systemic thinking may help in understanding complex systems, and, on the other hand, it is visualised what kind of new challenges the new, data-centered world is issuing to general system theory =-=[3]-=-. Levels of Abstraction It seems that when there are high numbers of lower-level entities together, the appropriate way of looking at the whole system may change altogether – the relevant variables ar... |

2 |
HOT: A mechanism for power laws in designed systems
- Carlson, Doyle
- 1999
(Show Context)
Citation Context ...trict organisation among the entities. The theories beyond agents and agent networks are supported, for example, by the studies by Albert-Laszlo Barabasi [1], and also by John Doyle and collaborators =-=[5]-=-, where they show that new approaches are needed to master the robustness properties in decentralised but mutually interconnected devices. As an extreme case of decentralisation and local interactions... |

2 |
Multivariate regression – From static models to subspace identification
- Hyötyniemi
- 2001
(Show Context)
Citation Context ...vidual signals can be abstracted away. o However, new problems also emerge: Because of the high dimensionality of the data, new kinds of tools for multivariate statistics are needed (for example, see =-=[11]-=-). In [12], it is illustrated how the model or controller parameters can be optimised applying multivariate regression techniques in the above framework. In what follows, however, the homogeneity of t... |

2 |
Towards new languages for systems modelling, and On emergent models and optimization of parameters
- Hyötyniemi
- 2002
(Show Context)
Citation Context ...nals can be abstracted away. o However, new problems also emerge: Because of the high dimensionality of the data, new kinds of tools for multivariate statistics are needed (for example, see [11]). In =-=[12]-=-, it is illustrated how the model or controller parameters can be optimised applying multivariate regression techniques in the above framework. In what follows, however, the homogeneity of the data wi... |

2 |
Studies on emergence and cognition – Parts 1 and 2: Low-level functions, and High-level functionalities
- Hyötyniemi
- 2002
(Show Context)
Citation Context ...ng the steepest descent algorithm, taking successive steps of length µ along the negative gradient direction: T ( ) x( κ + 1) = I −µ H −µ C H C ⋅ x( κ )+µ C H ⋅χ −µ h . (17) T 1 2 2 As illustrated in =-=[13]-=-, it may be motivated to introduce non-positive definite weighting matrices. This means that there does not exist a finite solution for the optimisation problem. This problem can be avoided, 3if the ... |

2 |
Towards Perception Hierarchies
- Hyötyniemi
- 2002
(Show Context)
Citation Context ...ated by using some of the xi’s as control variables – this control mechanism is the key to managing larger models and flexible hierarchies, consisting of various mutually exclusive substructures (see =-=[15]-=-). It is evident that efficient conceptual tools are needed to master the complicated control structures in such a simplistic framework – this question is elaborated on in next section. Increasing Fun... |

2 |
On the universality and undecidability in dynamic systems
- Hyötyniemi
- 2002
(Show Context)
Citation Context ... and learning by examples can be implemented; the highest valued element of a vector can also be determined, making it possible to distinguish between mutually exclusive clusters. Indeed, as shown in =-=[16]-=-, all algorithms can also be implemented in the proposed framework; using appropriate tools, the matrix A can be written so that truly anything can be realised in this extremely simple framework (see ... |

1 |
Calibration of an X-ray fluorescence analyzer using clustered components
- Hyötyniemi, Koskinen, et al.
- 2000
(Show Context)
Citation Context ...eoretical limitations of linear regression, the robustness of the model is not very good; not all available data can be included in the model, even if it might be potentially informative. As shown in =-=[10]-=-, multivariate methods can offer considerable benefits in calibration. What is nice, the latent variable approach of the more sophisticated multivariate methods is completely compatible with the propo... |

1 |
Image analysis and vision systems for processing plants
- Niemi, Hyötyniemi, et al.
- 1999
(Show Context)
Citation Context ... x. For example, in the case of unimodal data, the features can represent the directions of the most significant principal components. Some practical examples of “featuring a system” are presented in =-=[18]-=-. In mathematical terms, the columns of C determine the (not necessarily orthonormal) basis vectors, or coordinate axes in the lower-dimensional subspace, whereas the elements in x determine the corre... |