Elsevier

Pattern Recognition Letters

Volume 64, 15 October 2015, Pages 3-10
Pattern Recognition Letters

Pattern recognition between science and engineering: A red herring?

https://doi.org/10.1016/j.patrec.2015.06.030Get rights and content

Highlights

  • We aim to reopen a classic debate about the nature of pattern recognition

  • The relation between science and engineering is more problematic than we use to think

  • Pattern recognition is a good example of the interplay between science and technology

  • We offer some novel reflections concerning the notion of progress in our field

Abstract

Pattern recognition has been plagued since its beginnings by the elusiveness of its very nature, permeated as it is by a scientific as well as an engineering outlook, and over the years a debate has taken place among scholars aimed at clarifying its role and function. In this paper, we would like to reopen the discussion around the nature of pattern recognition research in the light of some recent developments both in the philosophy of technology and in the philosophy of science. These suggest that we have to rethink the classical dichotomy between science and engineering as, at the conceptual level, the boundary between the two camps turns out to be more blurred than is commonly thought, and that they stand to each other in a kind of circular, symbiotic relationship. Our analysis will be complemented by some historical examples and by further reflections concerning the notion of “progress” in our field.

Section snippets

Motivations

What is the nature of pattern recognition research? Is it an engineering enterprise or a scientific one? These and similar questions have been asked repeatedly during the (short) history of our field, and the community appears to be split into two parties.

On the one hand, we find those who adhere to a purely engineering perspective, with its emphasis on real-world problems and applications. This view, which is often implied by many a scholar and is revealed by the very structure of the most

Science and engineering: a tale of two myths

In the history of thought we find several misconceptions about science and engineering, two of which can be seen as enduring myths that accompanied the development of scientific and engineering activities for centuries. One views science as a well-demarcated endeavor totally committed to the discovery of (eternal) “truths,” while the other regards engineering simply as a practical activity which is never concerned with knowledge-for-knowledge’s sake [82]. In the past these ideas, which are

Pattern recognition: from science to technology and back

If we were to look for a contemporary research field that best exemplifies the interplay between science and technology, artificial intelligence would naturally suggest itself. But more than in other subfields of artificial intelligence, it is probably in pattern recognition and related areas (machine learning, computer vision, etc.) that one finds the most vivid manifestation of such an intimate relationship.

Perhaps the best example in this regard comes from the (troubled) history of neural

Discussion

What are the implications of these observations? Granted that the demarcation between science and technology appears to be more blurred and problematic than we use to think, what are we to make of this? The debates over the disciplinary status of a field of study are often regarded by practitioners as abstract discussions with little or no practical value. On the contrary, if taken seriously, we believe that the above considerations can offer researchers a fresh perspective that might lead the

Conclusions

With this paper we wanted to reopen a classic debate about the nature of pattern recognition research. Over the past few decades, in fact, two opposite tendencies have emerged: one which considers our field to be a purely engineering or technological discipline, the other which, on the contrary, sees ours as a scientific enterprise. These positions seem irreconcilable, for the simple reason that the two camps are traditionally felt to have different goals, one aiming at “use,” the other aiming

Acknowledgments

We would like to thank Gori and Roli for many stimulating discussions on the notion of progress and performance evaluation in computer vision and pattern recognition, and the anonymous reviewers for their constructive comments.

References (90)

  • C. Bishop

    Pattern Recognition and Machine Learning

    (2006)
  • M. Boon

    In defense of engineering sciences: on the epistemological relations between science and technology

    Techné

    (2011)
  • V. Bruce et al.

    Visual Perception: Physiology, Psychology and Ecology

    (2003)
  • M. Bunge

    Technology as applied science

    Technol. cult.

    (1966)
  • V. Bush

    Science the Endless Frontier

    (1945)
  • E. Constant

    The Origins of the Turbojet Revolution

    (1980)
  • F. Crick

    The recent excitement about neural networks

    Nature

    (1989)
  • H.L. Dreyfus

    What Computers Still Can’t Do: A Critique of Artificial Reason

    (1992)
  • C. Drummond

    Machine learning as an experimental science

    Proceedings of the 1st Workshop on Evaluation Methods for Machine Learning

    (2006)
  • C. Drummond

    Replicability in not reproducibility: nor is it good science

    Proceedings of the 4th Workshop on Evaluation Methods for Machine Learning

    (2009)
  • R.O. Duda et al.

    Pattern Classification

    (2000)
  • R. Duin

    Four scientific approaches to pattern recognition

  • R. Duin, Machine learning and pattern recognition, 2012,...
  • R. Duin et al.

    The science of pattern recognition: achievements and perspectives

  • R.P.W. Duin

    A note on comparing classifiers

    Pattern Recognit. Lett.

    (1996)
  • R.P.W. Duin et al.

    A note on core research issues for statistical pattern recognition

    Pattern Recognit. Lett.

    (2002)
  • J. Ellul

    The Technological Society

    (1964)
  • M. Franssen et al.

    Philosophy of technology

  • A. Freno

    Statistical machine learning and the logic of scientific discovery

    IRIS

    (2009)
  • A. Gehlen

    Man in the Age of Technology

    (1980)
  • J.J. Gibson

    The Perception of the Visual World

    (1950)
  • D. Gillies

    Artificial Intelligence and Scientific Method

    (1996)
  • I.E. Gordon

    Theories of Visual Perception

    (2004)
  • M. Gori et al.

    Learning to see like children: proof of concept

    arXiv:1408.2478v1 [cs.CV]

    (2014)
  • G. Gutting

    Paradigms and Revolutions: Appraisals and Applications of Thomas Kuhn’s Philosophy of Science

    (1980)
  • I. Hacking

    Representing and Intervening. Introductory Topics in the Philosophy of Natural Science

    (1983)
  • D.J. Hand

    Classifier technology and the illusion of progress

    Stat. Sci.

    (2006)
  • R. Haralick

    Computer vision theory: the lack thereof

    Comput. Vis. Gr. and Image Process.

    (1986)
  • G. Harman et al.

    Reliable Reasoning: Induction and Statistical Learning Theory (Jean Nicod Lectures)

    (2007)
  • M. Heidegger

    The Question Concerning Technology and Other Essays

    (1977)
  • D.H. Hubel

    Eye, Brain, and Vision

    (1995)
  • R.C. Jain et al.

    Ignorance, myopia, and naiveté in computer vision systems

    Comput. Vis. Gr. and Image Process.: Image Underst.

    (1991)
  • F. Jäkel et al.

    Does cognitive science need kernels?

    Trends in Cognitive Sci.

    (2009)
  • H. Jonas

    The Imperative of Responsibility: in Search of an Ethics for the Technological Age

    (1984)
  • B. Julesz

    Foundations of Cyclopean Perception

    (1971)
  • Cited by (3)

    • On unifiers, diversifiers, and the nature of pattern recognition

      2015, Pattern Recognition Letters
      Citation Excerpt :

      Dyson made no explicit statements on whether his dichotomy was intended to apply only to science, but we can speculate. The debate over the definition of science versus engineering could form an entire article by itself, and an excellent example is to be found within this special issue [5]. Acknowledging this, but for the purposes of simplicity, I will adopt a distinction as so:

    • Computational Neuroscience: Mathematical and Statistical Perspectives

      2018, Annual Review of Statistics and Its Application

    This paper has been recommended for acceptance by Gabriella Sanniti di Baja.

    View full text