Abstract
Autonomous knowledge transfer from a known task to a new one requires discovering task similarities and knowledge generalization without the help of a designer or teacher. How transfer mechanisms in such learning may work is still an open question. Transfer of knowledge makes most sense for learners for whom novelty is regular (other things being equal), as in the physical world. When new information must be unified with existing knowledge over time, a cumulative learning mechanism is required, increasing the breadth, depth, and accuracy of an agent’s knowledge over time, as experience accumulates. Here we address the requirements for what we refer to as autonomous cumulative transfer learning (ACTL) in novel task-environments, including implementation and evaluation criteria, and how it relies on the process of similarity and ampliative reasoning. While the analysis here is theoretical, the fundamental principles of the cumulative learning mechanism in our theory have been implemented and evaluated in a running system described priorly. We present arguments for the theory from an empirical as well as analytical viewpoint.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
Notes
- 1.
We define a teacher as a process outside the learner whose interaction helps reduce the search space for a solution to a goal or task.
- 2.
Unless otherwise noted, we use the term induction in the reasoning sense, not in Solomonoff’s “universal induction” sense [11].
- 3.
Peirce’s use of the concept of ‘ampliative reasoning’ included abduction, induction and analogy [10]; ours adds (corrigible) deduction to that list.
- 4.
We use ‘aspects’ as shorthand for ‘sub-divisions of a phenomenon that are of pragmatic importance to an agent’s goals and tasks’.
- 5.
Since phenomena in the physical world contain an infinite set of subdivisions such a claim would always be limited by pragmatic considerations (see prior footnote). Time and energy will also present hard limits for any such consideration. Thus, there is no literal sense in which complete familiarity may be reached.
- 6.
The term ‘percept’ as used here references sets of variables in the preceding sense, whether generated by sensors here-and-now, retrieved from memory, or imaginatively constructed.
- 7.
This may be done by backward-chaining from the goal state to the present state using various assumptions about the task-environment [15]. Other options exist; an adequate explanation and demonstration of these would require a separate paper.
References
Brockman, G., et al.: Openai gym. arXiv preprint arXiv:1606.01540 (2016)
Castro, P.S., Precup, D.: Automatic construction of temporally extended actions for MDPs using bisimulation metrics. In: Sanner, S., Hutter, M. (eds.) EWRL 2011. LNCS (LNAI), vol. 7188, pp. 140–152. Springer, Heidelberg (2012). https://doi.org/10.1007/978-3-642-29946-9_16
Eberding, L.M., Sheikhlar, A., Thórisson, K.R.: SAGE: task-environment platform for evaluating a broad range of AI learners. In: International Conference on Artificial General Intelligence. Springer, submitted in 2020
Kaptelinin, V., Nardi, B.A.: Acting with Technology: Activity Theory and Interaction Design. MIT Press, Cambridge (2006)
Ng, K.H., Du, Z., Ng, G.W.: DSO cognitive architecture: unified reasoning with integrative memory using global workspace theory. In: Everitt, T., Goertzel, B., Potapov, A. (eds.) AGI 2017. LNCS (LNAI), vol. 10414, pp. 44–53. Springer, Cham (2017). https://doi.org/10.1007/978-3-319-63703-7_5
Nivel, E., et al.: Bounded recursive self-improvement. Tech report number: RUTR-SCS13006, Reykjavik University - School of Computer Science (2013)
Özkural, E.: Zeta distribution and transfer learning problem. In: Iklé, M., Franz, A., Rzepka, R., Goertzel, B. (eds.) AGI 2018. LNCS (LNAI), vol. 10999, pp. 174–184. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-97676-1_17
Pearl, J.: Bayesianism and causality, or, why i am only a Half-Bayesian. In: Corfield, D., Williamson, J. (eds.) Foundations of Bayesianism, pp. 19–36. Springer, Dordrecht (2001). https://doi.org/10.1007/978-94-017-1586-7_2
Perkins, D.N., Salomon, G., et al.: Transfer of learning. Int. Encyclopedia Educ. 2, 6452–6457 (1992)
Psillos, S.: An explorer upon untrodden ground: Peirce on abduction. In: Handbook of the History of Logic, vol. 10, pp. 117–151. Elsevier (2011)
Solomonoff, R.J.: A formal theory of inductive inference. Part I. Inf. Control 7(1), 1–22 (1964)
Sorg, J., Singh, S.: Transfer via soft homomorphisms. In: Proceedings of The 8th International Conference on Autonomous Agents and Multiagent Systems-Volume 2, pp. 741–748 (2009)
Taylor, M.E., Stone, P.: Transfer learning for reinforcement learning domains: a survey. J. Mach. Learn. Res. 10, 1633–1685 (2009)
Thórisson, K.R., Bieger, J., Li, X., Wang, P.: Cumulative learning. In: Hammer, P., Agrawal, P., Goertzel, B., Iklé, M. (eds.) AGI 2019. LNCS (LNAI), vol. 11654, pp. 198–208. Springer, Cham (2019). https://doi.org/10.1007/978-3-030-27005-6_20
Thórisson, K.R., Talbot, A.: Cumulative learning with causal-relational models. In: Iklé, M., Franz, A., Rzepka, R., Goertzel, B. (eds.) AGI 2018. LNCS (LNAI), vol. 10999, pp. 227–237. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-97676-1_22
Wang, P.: Non-axiomatic reasoning system: exploring the essence of intelligence. Citeseer (1995)
Wang, P.: Non-Axiomatic Logic: A Model of Intelligent Reasoning. World Scientific, Singapore (2013)
Yosinski, J., Clune, J., Bengio, Y., Lipson, H.: How transferable are features in deep neural networks? In: Advances in Neural Information Processing Systems, pp. 3320–3328 (2014)
Acknowledgements
The authors would like to thank Hjörleifur Henriksson at IIIM for help with computer setup and data collection. This work was supported in part by grants from the Icelandic Institute for Intelligent Machines, Reykjavik University and Cisco Systems.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Sheikhlar, A., Thórisson, K.R., Eberding, L.M. (2020). Autonomous Cumulative Transfer Learning. In: Goertzel, B., Panov, A., Potapov, A., Yampolskiy, R. (eds) Artificial General Intelligence. AGI 2020. Lecture Notes in Computer Science(), vol 12177. Springer, Cham. https://doi.org/10.1007/978-3-030-52152-3_32
Download citation
DOI: https://doi.org/10.1007/978-3-030-52152-3_32
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-52151-6
Online ISBN: 978-3-030-52152-3
eBook Packages: Computer ScienceComputer Science (R0)