Abstract
In this article we present a new modelling framework for structured concepts using a category-theoretic generalisation of conceptual spaces, and show how the conceptual representations can be learned automatically from data, using two very different instantiations: one classical and one quantum. A contribution of the work is a thorough category-theoretic formalisation of our framework. We claim that the use of category theory, and in particular the use of string diagrams to describe quantum processes, helps elucidate some of the most important features of our approach. We build upon Gärdenfors’ classical framework of conceptual spaces, in which cognition is modelled geometrically through the use of convex spaces, which in turn factorise in terms of simpler spaces called domains. We show how concepts from the domains of shape, colour, size and position can be learned from images of simple shapes, where concepts are represented as Gaussians in the classical implementation, and quantum effects in the quantum one. In the classical case we develop a new model which is inspired by the \(\beta \)-VAE model of concepts, but is designed to be more closely connected with language, so that the names of concepts form part of the graphical model. In the quantum case, concepts are learned by a hybrid classical-quantum network trained to perform concept classification, where the classical image processing is carried out by a convolutional neural network and the quantum representations are produced by a parameterised quantum circuit. Finally, we consider the question of whether our quantum models of concepts can be considered conceptual spaces in the Gärdenfors sense.
Similar content being viewed by others
Notes
Note that we are not making any claims of “quantum supremacy” (Preskill 2012) for the particular set of quantum models that we implement in this article. However, we do anticipate the possibility of quantum models of concepts satisfying our framework which require quantum hardware for their efficient training and deployment, especially as we scale to more realistic datasets and larger quantum circuits.
Section 2.6 describes entanglement; we leave the use of partial orders in experiments for future work.
For example if \(f \le g\) then \(h \circ f \le h \circ g\), \(f \circ h \le g \circ h\) and \(f \otimes h \le g \otimes h\) where h is any morphism h of an appropriate type for each case.
Later we will define instances as special cases of points. Instances and points differ in quantum models, because of entanglement, but coincide classically.
Henceforth we use the generic term “model” rather than “space” since a conceptual model can be defined in a category without any spatial character.
Here we use the standard definition of integration on a measurable space, which exists since \(g(-,A)\) is measurable and bounded in [0, 1] by assumption.
Here we use the standard “bra-ket” notation whereby vectors and linear functionals on \(\mathcal {H}\) are written in the form \(|{\psi }\rangle \), \(\langle {\phi }|\) respectively. Then for a unit vector \(\psi \in \mathcal {H}\), \(|{\psi }\rangle \langle {\psi }|\) is the density operator of the corresponding pure state on \(\mathcal {H}\).
In this article “combination” of concepts is always meant in this sense. However there are many distinct meaningful operations on concepts which could also be called their combination, such as the more conjunction-like notion of combining “pet” and “fish” into “pet fish” (Aerts and Gabora 2005).
In this section we use bold font for variables, e.g. the conceptual space \(\textbf{Z}\), to be consistent with the machine learning literature.
The same patterns were observed on the development data. We used the training data since this gives denser plots.
The idea of plotting transitions along a dimension is taken from Higgins et al. (2017).
The entangling layer is self-inverse, so that two layers allow us to implement a rotation on any qubit. A swap operation on any pair of qubits can be implemented using three layers, and from this any CX gate. Hence we may implement the universal gate set given by single-qubit phase and Clifford gates; see, for example, Van de Wetering (2021).
In Section 4.2.2 below we investigate how the addition of a decoder can affect the instance and concept representations.
Of course there is nothing to prevent us from using more than one qubit per domain, in order to provide a larger Hilbert space in which to represent the additional colours, but the visualisation is harder with more qubits.
One possibility for future work is to develop and implement a “quantum VAE” (Khoshaman et al. 2018) for concept modelling, and have a generative model in which all parts of the model are quantum.
References
Abramson J, Ahuja A, Barr I, Brussee A, Carnevale F, Cassin M (2020) DeepMind-interactive-agents-group. Imitating Interactive Intelligence arXiv:2012.05672
Aerts D (2009) Quantum structure in cognition. J Math Psychol 53(5):314–348
Aerts D, Gabora L (2005) A state-context-property model of concepts and their combinations I: the structure of the sets of contexts and properties. Kybernetes 34:151–175
Aisbett J, Gibbon G (2001) A general formulation of conceptual spaces as a meso level representation. Artif Intell 133(1–2):189–232
Bechberger L, Kühnberger K-U (2017) A thorough formalization of conceptual spaces. Joint German/Austrian conference on artificial intelligence (künstliche intelligenz) (pp 58–71)
Benedetti M, Lloyd E, Sack S, Fiorentini M (2019) Parameterized quantum circuits as machine learning models. Quantum Sci Technol 4(043001)
Bengio Y, Courville A, Vincent P (2013) Representation learning: a review and new perspectives. IEEE Trans Pattern Anal Mach Intell
Birkhoff G, von Neumann J (1936) The logic of quantum mechanics. Annal Math 37(4):823–843
Bogacz R (2017) A tutorial on the free-energy framework for modelling perception and learning. J Math Psychol 76:198–211
Bolt J, Coecke B, Genovese F, Lewis M, Marsden D, Piedeleu R (2019) Interacting conceptual spaces I: grammatical composition of concepts. Conceptual spaces: elaborations and applications, Springer (pp 151–181)
Bražinskas A, Havrylov S, Titov I (2018) Embedding words as distributions with a Bayesian skip-gram model. Proceedings of the 27th international conference on computational linguistics (pp 1775–1789). Santa Fe, New Mexico, USA: Association for Computational Linguistics. Retrieved from https://aclanthology.org/C18-1151
Cho K, Jacobs B (2019) Disintegration and Bayesian inversion via string diagrams. Math Struc Comput Sci 29(7):938–971
Cho K, Jacobs B, Westerbaan B, Westerbaan A (2015) An introduction to effectus theory. arXiv:1512.05813
Clark S, Lerchner A, von Glehn T, Tieleman O, Tanburn R, Dashevskiy M, Bosnjak M (2021) Formalising concepts as grounded abstractions (Tech. Rep.). https://arxiv.org/pdf/2101.05125.pdf: DeepMind, London
Coecke B (2006) Introducing categories to the practicing physicist. What is Category Theory 30:45–74
Coecke B, Kissinger A (2017) Picturing quantum processes: a first course in quantum theory and diagrammatic reasoning. Cambridge University Press
Coecke B, Sadrzadeh M, Clark S (2010) Mathematical foundations for a compositional distributional model of meaning. arXiv:1003.4394
Doersch C (2016) Tutorial on variational autoencoders (Tech. Rep.), UC Berkeley. arXiv:1606.05908
Epping GP, Busemeyer JR (2022) Using diverging predictions from classical and quantum models to dissociate between categorization systems. https://doi.org/10.31234/osf.io/fq2k5
Epping GP, Fisher EL, Zeleznikow-Johnston A, Pothos E, Tsuchiya N (2021) A quantum geometric framework for modeling color similarity judgements. https://doi.org/10.31234/osf.io/vtzrq
Fong B (2019) An invitation to applied category theory - seven sketches in compositionality. Cambridge University Press
Friston K, Kiebel S (2009) Predictive coding under the free-energy principle. Philosophical transactions of the Royal Society B: Biological sciences 364(1521):1211–1221
Ganter B, Obiedkov S (2016) Conceptual exploration. Springer
Ganter B, Wille R (1999) Formal concept analysis: mathematical foundations. Springer Science & Business Media
Gärdenfors P (2004) Conceptual spaces: the geometry of thought. MIT press
Gärdenfors P (2014) The geometry of meaning. The MIT Press
Goodfellow I, Bengio Y, Courville A (2016) Deep learning. The MIT Press
Goodwin GP, Johnson-Laird PN (2013) The acquisition of Boolean concepts. Trends Cognit Sci 17. https://doi.org/10.1016/j.tics.2013.01.007
Gopnik A, Meltzoff A (1997) Words, thoughts, and theories. MIT Press
Harnad S (1990) The symbol grounding problem. Physica D: Nonlinear Phenomona 42:335–346
Havlicek V, Corcoles AD, Temme K, Harrow AW, Kandala A, Chow JM, Gambetta JM (2019) Supervised learning with quantum-enhanced feature spaces. Nature 567:209–212
Higgins I, Matthey L, Pal A, Burgess CP, Glorot X, Botvinick M, Lerchner A (2017) ß-VAE: learning basic visual concepts with a constrained variational framework. Proceedings of ICLR 2017
Higgins I, Sonnerat N, Matthey L, Pal A, Burgess CP, Bošnjak M, Lerchner A (2018) SCAN: learning hierarchical compositional visual concepts. Proceedings of ICLR 2018
Huang Q, Smolensky P, He X, Deng L, Wu D (2018) Tensor product generation networks for deep NLP modeling. Proceedings of the 2018 conference of the north American chapter of the association for computational linguistics: Human language technologies, vol 1 (long papers) (pp 1263–1273). New Orleans, Louisiana: Association for Computational Linguistics. Retrieved from https://aclanthology.org/N18-1114
Khoshaman A, Vinci W, Denis B, Andriyash E, Sadeghi H, Amin MH (2018) Quantum variational autoencoder. Quantum. Sci Technol 4(1):014001
Kingma DP, Welling M (2014) Auto-encoding variational Bayes. Proceedings of the international conference on learning representations (ICLR 2014)
Lake BM, Ullman TD, Tenenbaum JB, Gershman SJ (2017) Building machines that learn and think like people. Behav Brain Sci 40
Lewis M, Lawry J (2016) Hierarchical conceptual spaces for concept combination. Artif Intell 237:204–227
Locatello F, Bauer S, Lucic M, Rätsch G, Gelly S, Schölkopf B, Bachem O (2019) Challenging common assumptions in the unsupervised learning of disentangled representations. Proceedings of the 36th international conference on machine learning. Long Beach, California
Lorenz R, Pearson A, Meichanetzidis K, Kartsaklis D, Coecke B (2023) QNLP in practice: running compositional models of meaning on a quantum computer. J Artif Intell Res 76. https://doi.org/10.1613/jair.1.14329
Margolis E, Laurence S (Eds.) (2015) The conceptual mind: new directions in the study of concepts. The MIT Press
Margolis E, Laurence S (2022) Concepts. https://plato.stanford.edu/archives/fall2022/entries/concepts/. (The Stanford Encyclopedia of Philosophy)
Murphy GL (2002) The big book of concepts. The MIT Press
Panangaden P (1998) Probabilistic relations. School of Computer Science Research Reports-University of Birmingham CSR 59–74
Pothos EM, Busemeyer JR (2013) Can quantum probability provide a new direction for cognitive modeling? Behav Brain Sci 36(3)
Preskill J (2012) Quantum computing and the entanglement frontier. (Rapporteur talk at the 25th Solvay Conference on Physics - The Theory of the Quantum World). arXiv:1203.5813
Rezende DJ, Mohamed S, Wierstra D (2014) Stochastic backpropagation and approximate inference in deep generative models. Proceedings of the 31st international conference on machine learning (pp 1278–1286)
Rickard JT, Aisbett J, Gibbon G (2007) Reformulation of the theory of conceptual spaces. Inf Sci 177(21):4539–4565
Rodatz B, Shaikh RA, Yeh L (2021) Conversational negation using worldly context in compositional distributional semantics. arXiv:2105.05748
Rosch EH (1973) Natural categories. Cognit Psychol 4(3):328–350
Schlangen D, Zarrieß S, Kennington C (2016) Resolving references to objects in photographs using the words-as-classifiers model. Proceedings of the 54th annual meeting of the association for computational linguistics (vol. 1: Long papers) (pp 1213–1223). Berlin, Germany: Association for Computational Linguistics. Retrieved from https://aclanthology.org/P16-1115
Schuld M, Killoran N (2019) Quantum machine learning in feature Hilbert spaces. Phys Rev Lett 122:040504. https://doi.org/10.1103/PhysRevLett.122.040504
Selinger P (2010) A survey of graphical languages for monoidal categories. New structures for physics, Springer (pp 289–355)
Shaikh RA, Yeh L, Rodatz B, Coecke B (2021) Composing conversational negation. arXiv:2107.06820
Shiebler D, Gavranovic B, Wilson P (2021) Category theory in machine learning. The 4th international conference on applied category theory. Cambridge, UK
Smolensky P, Legendre G (2006) The harmonic mind. The MIT Press
Tomas V, Sylvie D (2015) Unitary transformations in the quantum model for conceptual conjunctions and its application to data representation. Front Psychol 6
Trueblood JS, Busemeyer JR (2011) A quantum probability account of order effects in inference. Cognit Sci 35:1518–1552
Tull S (2019) Categorical operational physics. arXiv:1902.00343
Tull S (2021) A categorical semantics of fuzzy concepts in conceptual spaces. Proceedings of Applied Category Theory 2021
Van de Wetering J (2021) Constructing quantum circuits with global gates. New J Phys 23(4):043015
Watters N, Matthey L, Borgeaud S, Kabra R, Lerchner A (2019) Spriteworld: a flexible, configurable reinforcement learning environment. https://github.com/deepmind/spriteworld/. Retrieved from https://github.com/deepmind/spriteworld/
Yan F, Li N, Hirota K (2021) Qhsl: a quantum hue, saturation, and lightness color model. Inf Sci 577:196–213
Funding
N/A
Author information
Authors and Affiliations
Contributions
Sean Tull developed the mathematical formalisation and wrote the theory sections. Razin A. Shaikh wrote the code, ran the experiments, and prepared some of the figures. Sara Sabrina Zemljic created the data and helped run the experiments. Stephen Clark oversaw the project, ran some of the experiments, and wrote the remainder of the manuscript. All authors took part equally in setting the general direction of the project.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Ethics approval
N/A
Consent to participate
N/A
Consent for publication
Yes
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
A The shapes dataset
The parameters used in the Spriteworld software to generate the Shapes dataset:
Additional parameters for the colour domain:
A.1 The extended colour dataset
The parameters used in the Spriteworld software to generate the Shapes dataset with more (rainbow) colours:
B Neural architectures and hyper-parameters
image width | 64 |
image height | 64 |
image channels | 3 |
CNN kernel size | \(4\times 4\) |
CNN stride | \(2\times 2\) |
CNN layers | 4 |
CNN filters | 64 |
CNN dense layers | 2 |
CNN dense layer size | 256 |
dimensions of latent space | 6 |
initialization interval | \([-1.0, 1.0]\) |
for means of priors | |
initialization interval | \([-7.0, 0.0]\) |
for log-variances of priors | |
batch size | 32 |
Adam learning rate | \(10^{-3}\) |
Adam \(\beta _1\) | 0.9 |
Adam \(\beta _2\) | 0.999 |
Adam \(\epsilon \) | \(10^{-7}\) |
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Tull, S., Shaikh, R.A., Zemljič, S.S. et al. From conceptual spaces to quantum concepts: formalising and learning structured conceptual models. Quantum Mach. Intell. 6, 21 (2024). https://doi.org/10.1007/s42484-023-00134-z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s42484-023-00134-z