Empiricism without magic: transformational abstraction in deep convolutional neural networks

Abstract

In artificial intelligence, recent research has demonstrated the remarkable potential of Deep Convolutional Neural Networks (DCNNs), which seem to exceed state-of-the-art performance in new domains weekly, especially on the sorts of very difficult perceptual discrimination tasks that skeptics thought would remain beyond the reach of artificial intelligence. However, it has proven difficult to explain why DCNNs perform so well. In philosophy of mind, empiricists have long suggested that complex cognition is based on information derived from sensory experience, often appealing to a faculty of abstraction. Rationalists have frequently complained, however, that empiricists never adequately explained how this faculty of abstraction actually works. In this paper, I tie these two questions together, to the mutual benefit of both disciplines. I argue that the architectural features that distinguish DCNNs from earlier neural networks allow them to implement a form of hierarchical processing that I call “transformational abstraction”. Transformational abstraction iteratively converts sensory-based representations of category exemplars into new formats that are increasingly tolerant to “nuisance variation” in input. Reflecting upon the way that DCNNs leverage a combination of linear and non-linear processing to efficiently accomplish this feat allows us to understand how the brain is capable of bi-directional travel between exemplars and abstractions, addressing longstanding problems in empiricist philosophy of mind. I end by considering the prospects for future research on DCNNs, arguing that rather than simply implementing 80s connectionism with more brute-force computation, transformational abstraction counts as a qualitatively distinct form of processing ripe with philosophical and psychological significance, because it is significantly better suited to depict the generic mechanism responsible for this important kind of psychological processing in the brain.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4

(a is public domain from Gray (1918), b is available from Selket under a Creative Commons 3.0 Share-Alike License, c from Vidyasagar (2013) on a Creative Commons Attribution License, and d from Perry and Fallah (2014) on a Creative Commons Attribution License)

Fig. 5
Fig. 6

(reproduced from Grósz and Nagy 2014)

Fig. 7
Fig. 8
Fig. 9
Fig. 10

(reproduced from Singhal 2017)

Fig. 11
Fig. 12
Fig. 13

(Images reproduced from Dosovitskiy et al. 2015)

Fig. 14

(figure reproduced from Odena et al. 2016)

Fig. 15
Fig. 16

Notes

  1. 1.

    This question has been raised as the “interpretation problem”; however, this label has been used too broadly and inconsistently to admit of a single solution. Some commentators use it to broach the question addressed here—why do DCNNs succeed where other neural network architectures struggle—while others use it to raise other questions, such as semantic interpretability or decision justification.

  2. 2.

    Some residual problems may be extracted from the critiques, however, especially regarding the biological plausibility of the procedures used to train DCNNs. I address these residual concerns in the final section.

  3. 3.

    For prominent empiricist nodes in the current debate, see (Botvinick et al. 2017; McClelland et al. 2010).

  4. 4.

    Even three-layer perceptrons have been trained to categorize triangle exemplars with a high degree of accuracy (Spasojević et al. 2012).

  5. 5.

    This is but the barest gloss on a rich research area in the foundations of logic and math going back to Hilbert—for a recent overview, see Antonelli (2010).

  6. 6.

    Achille and Soatto (2017) have recently argued that implicit or explicit regularization is a fourth crucially important feature in generalizing DCNN performance (to prevent them from simply memorizing the mapping for every exemplar in the training set), but since there is significant diversity in regularization procedures and this idea is more preliminary, I do not discuss it further here.

  7. 7.

    In the interests of space, we move quickly over the history here; for more background and discussion, see (Buckner and Garson 2018; Schmidhuber 2015).

  8. 8.

    Note that some functionalists (i.e. Weiskopf 2011a, b) defend the explanatory power of idealized models and so may not think much is gained by restricting DCNNs to more biologically-plausible values (though for rebuttal from a more mechanistic perspective, see Buckner 2015).

  9. 9.

    Note that when DCNNs are deployed for categorization or other forms of decision-making, the final layer of the network will typically be a fully-connected classifier that takes input from all late-stage nodes (i.e. a fully connected layer of nodes or set of category-specific support-vector machines). These are used to draw the boundaries between the different category manifolds in the transformed similarity space. Since these components are deployed in many other machine learning methods that do not model transformational abstraction, I do not discuss them further here.

  10. 10.

    An important current point of controversy is whether specifically max-pooling is required to reduce the search space and avoid overfitting, or whether other downsampling methods might be as effective. For two poles in this debate, see (Patel et al. 2016; Springenberg et al. 2014). The present paper holds that even if alternative solutions are also practically effective, biologically-relevant networks must somehow implement the aggregative role of complex cells—though max-pooling is perhaps only one possible technique in a family of downsampling operations that could accomplish this (DiCarlo and Cox 2007).

  11. 11.

    For some early empirical support for this view, see Achille and Soatto (2017).

  12. 12.

    For a worked example, see Goodfellow et al. (2016, p. 334), who show that edge detection alone can be roughly 60,000 times more computationally efficient when performed by a DCNN, compared to a traditional 3-layer perceptron.

  13. 13.

    One could also worry here that AlphaGo did not learn the rules of Go from experience, but this does not impugn the point. What is claimed is rather that once these rules were provided, a DCNN can learn strategies without any domain-specific strategy heuristics (which knowledge of the rules do not provide). This is especially driven home by AlphaGo Zero, which acquired strategies entirely through self-play (Silver et al. 2017).

  14. 14.

    Interestingly, the DeepArt team found that average-pooling was more effective than max-pooling when the network was in generation mode.

  15. 15.

    This general characterization washes across differences in various canonical accounts of mechanism; see (Bechtel and Abrahamsen 2005; Glennan 2002; Machamer et al. 2000).

  16. 16.

    A likelier critical outcome is that both DCNNs and mammalian neocortex are members of the LN generic mechanism family, but there are other members in this family besides DCNNs that provide a tighter fit in performance and structure to humans. For example, while a more recent study by DiCarlo and co-authors confirmed that DCNNs predict many low-resolution patterns in human perceptual similarity judgments and do so using the same sorts of features that are found in late-stage ventral stream processing in V4/5 and IT, they found that these models were not as predictive of high-resolution, image-by-image comparisons in humans as were rhesus monkeys (Rajalingham et al. 2018). They speculate that an alternative but nearby subfamily of models that tweaks one or more typical features of DCNNs—i.e. their diet of training on static images, or lack of recurrent connections between layers—might provide an even better mechanistic model of human perceptual similarity and categorization judgments without unduly complicating the model. However, whether this prospect will pay off—and do so without inhibiting the ability of DCNNs to generalize to non-primate species—remains an open empirical question, and DCNNs remain the most successful mechanistic model of primate visual perception that we have to date.

References

  1. Achille, A., & Soatto, S. (2017). Emergence of invariance and disentangling in deep representations. arXiv Preprint arXiv:1706.01350.

  2. Antonelli, G. A. (2010). Notions of invariance for abstraction principles. Philosophia Mathematica, 18(3), 276–292.

    Article  Google Scholar 

  3. Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577–660.

    Google Scholar 

  4. Bechtel, W., & Abrahamsen, A. (2005). Explanation: A mechanist alternative. Studies in History and Philosophy of Science Part C: Studies in History and Philosophy of Biological and Biomedical Sciences, 36(2), 421–441.

    Article  Google Scholar 

  5. Berkeley, G. (1710/1982). A treatise concerning the principles of human knowledge. Indianapolis: Hackett. (Original work published in 1710).

  6. Beth, E. W. (1957). Uber lockes “Allgemeines Dreieck”. Kant-Studien, 1(48), 361–380.

    Google Scholar 

  7. Blundell, C., Uria, B., Pritzel, A., Li, Y., Ruderman, A., Leibo, J. Z., et al. (2016). Model-free episodic control. arXiv Preprint arXiv:1606.04460.

  8. Boone, W., & Piccinini, G. (2016). Mechanistic abstraction. Philosophy of Science, 83(5), 686–697.

    Article  Google Scholar 

  9. Botvinick, M., Barrett, D. G., Battaglia, P., de Freitas, N., Kumaran, D., Leibo, J. Z., et al. (2017). Building machines that learn and think for themselves. Behavioral and Brain Sciences, 40, 26–28.

    Article  Google Scholar 

  10. Boyd, R. (1999). Kinds, complexity and multiple realization. Philosophical Studies, 95(1–2), 67–98.

    Article  Google Scholar 

  11. Brooks, R. A. (1991). Intelligence without representation. Artificial Intelligence, 47(1–3), 139–159.

    Article  Google Scholar 

  12. Buckner, C. (2011). Two approaches to the distinction between cognition and “mere association”. International Journal of Comparative Psychology, 24(4), 314–348.

    Google Scholar 

  13. Buckner, C. (2015). Functional kinds: A skeptical look. Synthese, 192(12), 3915–3942.

    Article  Google Scholar 

  14. Buckner, C., & Garson, J. (2018). Connectionism: Roots, revolution, and radiation. In M. Sprevak & M. Columbo (Eds.), The Routledge handbook of the computational mind. New York: Routledge.

    Google Scholar 

  15. Camp, E. (2015). Logical concepts and associative characterizations. In E. Margolis & S. Laurence (Eds.), The conceptual mind: New directions in the study of concepts (pp. 591–621). Cambridge: MIT Press.

    Google Scholar 

  16. Chatterjee, A. (2010). Disembodying cognition. Language and Cognition, 2(1), 79–116.

    Article  Google Scholar 

  17. Churchland, P. M. (1989). A neurocomputational perspective: The nature of mind and the structure of science. Cambridge: MIT press.

    Google Scholar 

  18. Clark, A. (1989). Microcognition: Philosophy, cognitive science, and parallel distributed processing (Vol. 6). Cambridge: MIT Press.

    Google Scholar 

  19. Craver, C., & Kaplan, D. M. (2018). Are more details better? On the norms of completeness for mechanistic explanations. The British Journal for the Philosophy of Science, axy015. https://doi-org.ezproxy.lib.uh.edu/10.1093/bjps/axy015.

  20. Cybenko, G. (1989). Approximation by superpositions of a sigmoidal function. Mathematics of Control, Signals, and Systems, 2(4), 303–314.

    Article  Google Scholar 

  21. DeMers, D., & Cottrell, G. W. (1993). Non-linear dimensionality reduction. In S. J. Hanson, J. D. Cowan & C. L. Giles (Eds.), Advances in neural information processing systems (NIPS) 5 (pp. 580–587). San Mateo: Morgan Kaufmann.

    Google Scholar 

  22. DiCarlo, J. J., & Cox, D. D. (2007). Untangling invariant object recognition. Trends in Cognitive Sciences, 11(8), 333–341. https://doi.org/10.1016/j.tics.2007.06.010.

    Article  Google Scholar 

  23. DiCarlo, J. J., Zoccolan, D., & Rust, N. C. (2012). How does the brain solve visual object recognition? Neuron, 73(3), 415–434.

    Article  Google Scholar 

  24. Dosovitskiy, A., Springenberg, J. T., & Brox, T. (2015). Learning to generate chairs with convolutional neural networks. In 2015 IEEE conference on computer vision and pattern recognition (CVPR) (pp. 1538–1546). https://doi.org/10.1109/CVPR.2015.7298761.

  25. Elsayed, G. F., Shankar, S., Cheung, B., Papernot, N., Kurakin, A., Goodfellow, I., & Sohl-Dickstein, J. (2018). Adversarial examples that fool both human and computer vision. arXiv Preprint arXiv:1802.08195.

  26. Fukushima, K. (1979). Neural network model for a mechanism of pattern recognition unaffected by shift in position-Neocognitron. IEICE Technical Report, A, 62(10), 658–665.

    Google Scholar 

  27. Fukushima, K. (2003). Neocognitron for handwritten digit recognition. Neurocomputing, 51, 161–180.

    Article  Google Scholar 

  28. Gärdenfors, P. (2004). Conceptual spaces: The geometry of thought. Cambridge: MIT press.

    Google Scholar 

  29. Gatys, L. A., Ecker, A. S., & Bethge, M. (2016). Image style transfer using convolutional neural networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 2414–2423).

  30. Gauker, C. (2011). Words and images: An essay on the origin of ideas. Oxford: OUP.

    Google Scholar 

  31. Glennan, S. (2002). Rethinking mechanistic explanation. Philosophy of Science, 69(S3), S342–S353.

    Article  Google Scholar 

  32. Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. Book in preparation for MIT Press. http://www.deeplearningbook.org.

  33. Goodfellow, I. J., Shlens, J., & Szegedy, C. (2014). Explaining and harnessing adversarial examples. arXiv Preprint arXiv:1412.6572.

  34. Gray, H. (1918). Anatomy of the human body, rev. and re-edited by Warren H. Lewis. Philadelphia: Lea & Febiger.

    Google Scholar 

  35. Grósz, T., & Nagy, I. (2014). Document classification with deep rectifier neural networks and probabilistic sampling. In Proceedings of the international conference on text, speech, and dialogue (pp. 108–115). Cham: Springer.

  36. Hahnloser, R. H., Sarpeshkar, R., Mahowald, M. A., Douglas, R. J., & Seung, H. S. (2000). Digital selection and analogue amplification coexist in a cortex-inspired silicon circuit. Nature, 405(6789), 947.

    Article  Google Scholar 

  37. Hassabis, D., Kumaran, D., Summerfield, C., & Botvinick, M. (2017). Neuroscience-inspired artificial intelligence. Neuron, 95(2), 245–258.

    Article  Google Scholar 

  38. Hinton, G. E., & Salakhutdinov, R. R. (2006). Reducing the dimensionality of data with neural networks. Science, 313(5786), 504–507.

    Article  Google Scholar 

  39. Hochreiter, S. (1991). Untersuchungen zu dynamischen neuronalen Netzen. Diploma thesis for Technische Universität München, München.

  40. Hong, H., Yamins, D. L., Majaj, N. J., & DiCarlo, J. J. (2016). Explicit information for category-orthogonal object properties increases along the ventral stream. Nature neuroscience, 19(4), 613.

    Article  Google Scholar 

  41. Hornik, K. (1991). Approximation capabilities of multilayer feedforward networks. Neural Networks, 4(2), 251–257.

    Article  Google Scholar 

  42. Hubel, D. H., & Wiesel, T. N. (1962). Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. The Journal of Physiology, 160(1), 106–154.

    Article  Google Scholar 

  43. Hume, D. (1739). A treatise on human nature. Oxford: Oxford University Press.

    Google Scholar 

  44. Kaplan, D. M., & Craver, C. F. (2011). The explanatory force of dynamical and mathematical models in neuroscience: A mechanistic perspective. Philosophy of Science, 78(4), 601–627.

    Article  Google Scholar 

  45. Khaligh-Razavi, S.-M., & Kriegeskorte, N. (2014). Deep supervised, but not unsupervised, models may explain IT cortical representation. PLoS Computational Biology, 10(11), e1003915. https://doi.org/10.1371/journal.pcbi.1003915.

    Article  Google Scholar 

  46. Kumaran, D., Hassabis, D., & McClelland, J. L. (2016). What learning systems do intelligent agents need? Complementary learning systems theory updated. Trends in Cognitive Sciences, 20(7), 512–534.

    Article  Google Scholar 

  47. Lake, B. M., Ullman, T. D., Tenenbaum, J. B., & Gershman, S. J. (2017). Building machines that learn and think like people. Behavioral and Brain Sciences, 40, E253.

    Article  Google Scholar 

  48. Laurence, S., & Margolis, E. (2012). Abstraction and the origin of general ideas. Philosopher’s Imprint, 12(19), 1–22.

    Google Scholar 

  49. Laurence, S., & Margolis, E. (2015). Concept nativism and neural plasticity. In E. Margolis & S. Laurence (Eds.), The conceptual mind: New directions in the study of concepts (pp. 117–147). Cambridge: MIT Press.

    Google Scholar 

  50. LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444.

    Article  Google Scholar 

  51. LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., et al. (1989). Backpropagation applied to handwritten zip code recognition. Neural Computation, 1(4), 541–551.

    Article  Google Scholar 

  52. LeCun, Y., Boser, B. E., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W. E., & Jackel, L. D. (1990). Handwritten digit recognition with a back-propagation network. In Advances in neural information processing systems (pp. 396–404).

  53. Levy, A., & Bechtel, W. (2013). Abstraction and the organization of mechanisms. Philosophy of Science, 80(2), 241–261.

    Article  Google Scholar 

  54. Lillicrap, T. P., Cownden, D., Tweed, D. B., & Akerman, C. J. (2016). Random synaptic feedback weights support error backpropagation for deep learning. Nature Communications. https://doi.org/10.1038/ncomms13276.

    Google Scholar 

  55. Luc, P., Neverova, N., Couprie, C., Verbeek, J., & LeCun, Y. (2017). Predicting deeper into the future of semantic segmentation. In IEEE international conference on computer vision (ICCV) (Vol. 1).

  56. Machamer, P., Darden, L., & Craver, C. F. (2000). Thinking about mechanisms. Philosophy of Science, 67(1), 1–25.

    Article  Google Scholar 

  57. Machery, E. (2009). Doing without concepts. Oxford: Oxford University Press.

    Google Scholar 

  58. Marcus, G. (2018). Deep learning: A critical appraisal. arXiv:1801.00631 [cs, Stat].

  59. McClelland, J. L. (1988). Connectionist models and psychological evidence. Journal of Memory and Language, 27(2), 107–123.

    Article  Google Scholar 

  60. McClelland, J. L., Botvinick, M. M., Noelle, D. C., Plaut, D. C., Rogers, T. T., Seidenberg, M. S., et al. (2010). Letting structure emerge: Connectionist and dynamical systems approaches to cognition. Trends in cognitive sciences, 14(8), 348–356.

    Article  Google Scholar 

  61. Mnih, V., Kavukcuoglu, K., Silver, D., Graves, A., Antonoglou, I., Wierstra, D., & Riedmiller, M. (2013). Playing atari with deep reinforcement learning. arXiv Preprint arXiv:1312.5602.

  62. Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., et al. (2015). Human-level control through deep reinforcement learning. Nature, 518(7540), 529.

    Article  Google Scholar 

  63. Montúfar, G. F., Pascanu, R., Cho, K., & Bengio, Y. (2014). On the number of linear regions of deep neural networks. In Advances in neural information processing systems (pp. 2924–2932).

  64. Odena, A., Dumoulin, V., & Olah, C. (2016). Deconvolution and checkerboard artifacts. Distill, 1(10), e3.

    Article  Google Scholar 

  65. Patel, A. B., Nguyen, M. T., & Baraniuk, R. (2016). A probabilistic framework for deep learning. In Advances in Neural Information Processing Systems (pp. 2558–2566).

  66. Perry, C. J., & Fallah, M. (2014). Feature integration and object representations along the dorsal stream visual hierarchy. Frontiers in Computational Neuroscience, 8, 84. https://doi.org/10.3389/fncom.2014.00084.

    Article  Google Scholar 

  67. Piccinini, G., & Craver, C. (2011). Integrating psychology and neuroscience: Functional analyses as mechanism sketches. Synthese, 183(3), 283–311.

    Article  Google Scholar 

  68. Priebe, N. J., Mechler, F., Carandini, M., & Ferster, D. (2004). The contribution of spike threshold to the dichotomy of cortical simple and complex cells. Nature Neuroscience, 7(10), 1113.

    Article  Google Scholar 

  69. Quine, W. V. (1971). Epistemology naturalized. Akten Des XIV. Internationalen Kongresses Für Philosophie, 6, 87–103.

    Google Scholar 

  70. Rajalingham, R., Issa, E. B., Bashivan, P., Kar, K., Schmidt, K., & DiCarlo, J. J. (2018). Large-scale, high-resolution comparison of the core visual object recognition behavior of humans, monkeys, and state-of-the-art deep artificial neural networks. bioRxiv, 240614.

  71. Ren, S., He, K., Girshick, R., & Sun, J. (2017). Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(6), 1137–1149. https://doi.org/10.1109/TPAMI.2016.2577031.

    Article  Google Scholar 

  72. Ritter, S., Barrett, D. G., Santoro, A., & Botvinick, M. M. (2017). Cognitive psychology for deep neural networks: A shape bias case study. arXiv Preprint arXiv:1706.08606.

  73. Rogers, T. T., & McClelland, J. L. (2014). Parallel distributed processing at 25: Further explorations in the microstructure of cognition. Cognitive Science, 38(6), 1024–1077. https://doi.org/10.1111/cogs.12148.

    Article  Google Scholar 

  74. Rosch, E. (1978). Principles of categorization. In E. Rosch & B. Lloyd (Eds.), Cognition and categorization (pp. 27–48). Hillsdale, NJ: Erlbaum.

    Google Scholar 

  75. Scellier, B., & Bengio, Y. (2017). Equilibrium propagation: Bridging the gap between energy-based models and backpropagation. Frontiers in Computational Neuroscience, 11, 24. https://doi.org/10.3389/fncom.2017.00024.

    Article  Google Scholar 

  76. Schmidhuber, J. (2015). Deep learning in neural networks: An overview. Neural Networks, 61, 85–117.

    Article  Google Scholar 

  77. Sejnowski, T. J., Koch, C., & Churchland, P. S. (1988). Computational neuroscience. Science, 241(4871), 1299–1306.

    Article  Google Scholar 

  78. Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., van den Driessche, G., et al. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484–489. https://doi.org/10.1038/nature16961.

    Article  Google Scholar 

  79. Silver, D., Schrittwieser, J., Simonyan, K., Antonoglou, I., Huang, A., Guez, A., et al. (2017). Mastering the game of go without human knowledge. Nature, 550(7676), 354.

    Article  Google Scholar 

  80. Singhal, H. (2017). Convolutional neural network with TensorFlow implementation. Retrieved September 7, 2018, from https://medium.com/data-science-group-iitr/building-a-convolutional-neural-network-in-python-with-tensorflow-d251c3ca8117.

  81. Spasojević, S. S., Šušić, M. Z., & DJurović, Ž. M. (2012). Recognition and classification of geometric shapes using neural networks. In 2012 11th symposium on neural network applications in electrical engineering (NEUREL) (pp. 71–76). IEEE.

  82. Springenberg, J. T., Dosovitskiy, A., Brox, T., & Riedmiller, M. (2014). Striving for simplicity: The all convolutional net. arXiv Preprint arXiv:1412.6806. Retrieved from https://arxiv.org/abs/1412.6806

  83. Stinson, C. (2016). Mechanisms in psychology: ripping nature at its seams. Synthese, 193(5), 1585–1614.

    Article  Google Scholar 

  84. Stinson, C. (2017). Back to the cradle: Mechanism schemata from piaget to DNA. In M. Adams, Z. Biener, U. Feest, & J. Sullivan (Eds.), Eppur si muove: Doing history and philosophy of science with Peter Machamer (pp. 183–194). Cham: Springer.

    Google Scholar 

  85. Stinson, C. (2018). Explanation and connectionist models. In M. Colombo & M. Sprevak (Eds.), The Routledge handbook of the computational mind. New York, NY: Routledge.

    Google Scholar 

  86. Vidyasagar, T. R. (2013). Reading into neuronal oscillations in the visual system: implications for developmental dyslexia. Frontiers in Human Neuroscience. https://doi.org/10.3389/fnhum.2013.00811.

    Google Scholar 

  87. Weiskopf, D. A. (2011a). Models and mechanisms in psychological explanation. Synthese, 183(3), 313.

    Article  Google Scholar 

  88. Weiskopf, D. A. (2011b). The functional unity of special science kinds. The British Journal for the Philosophy of Science, 62(2), 233–258.

    Article  Google Scholar 

  89. Yamins, D. L., & DiCarlo, J. J. (2016). Using goal-driven deep learning models to understand sensory cortex. Nature Neuroscience, 19(3), 356.

    Article  Google Scholar 

  90. Ylikoski, P., & Kuorikoski, J. (2010). Dissecting explanatory power. Philosophical Studies, 148(2), 201–219.

    Article  Google Scholar 

  91. Yu, C., & Smith, L. B. (2011). What you learn is what you see: using eye movements to study infant cross-situational word learning. Developmental Science, 14(2), 165–180.

    Article  Google Scholar 

Download references

Acknowledgements

This paper benefitted from an extraordinary amount of feedback from others, far too many to mention individually here. Particular thanks are due to Colin Allen, David Barack, Hayley Clatterbuck, Christopher Gauker, Bob Kentridge, Marcin Miłkowski, Mathias Niepert, Gualtiero Piccinini, Brendan Ritchie, Bruce Rushing, Whit Schonbein, Susan Sterrett, Evan Westra, Jessey Wright, two anonymous reviewers for this journal, and audiences at the University of Evansville, the Society for Philosophy and Psychology, the Southern Society for Philosophy and Psychology, Rice University’s CogTea, and the UH Department of Philosophy’s “works in progress” colloquium.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Cameron Buckner.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Buckner, C. Empiricism without magic: transformational abstraction in deep convolutional neural networks. Synthese 195, 5339–5372 (2018). https://doi.org/10.1007/s11229-018-01949-1

Download citation

Keywords

  • Abstraction
  • Connectionism
  • Convolution
  • Deep learning
  • Empiricism
  • Mechanism
  • Nuisance variation