The Boundaries of Verifiable Accuracy, Robustness, and Generalisation in Deep Learning

Alexander Bastounis; Alexander N. Gorban; Anders C. Hansen; Desmond J. Higham; Danil Prokhorov; Oliver Sutton; Ivan Y. Tyukin; Qinghua Zhou

doi:10.1007/978-3-031-44207-0_44

The Boundaries of Verifiable Accuracy, Robustness, and Generalisation in Deep Learning

Alexander Bastounis, Alexander N. Gorban, Anders C. Hansen, Desmond J. Higham, Danil Prokhorov, Oliver Sutton, Ivan Y. Tyukin^*, Qinghua Zhou

^*Corresponding author for this work

Mathematics

Research output: Contribution to conference types › Paper › peer-review

4 Citations (Scopus)

Abstract

In this work, we assess the theoretical limitations of determining guaranteed stability and accuracy of neural networks in classification tasks. We consider classical distribution-agnostic framework and algorithms minimising empirical risks and potentially subjected to some weights regularisation. We show that there is a large family of tasks for which computing and verifying ideal stable and accurate neural networks in the above settings is extremely challenging, if at all possible, even when such ideal solutions exist within the given class of neural architectures.

Original language	English
Pages	530-541
Number of pages	12
DOIs	https://doi.org/10.1007/978-3-031-44207-0_44
Publication status	E-pub ahead of print - 22 Sept 2023
Event	32nd International Conference on Artificial Neural Networks, ICANN 2023 - Heraklion, Greece Duration: 26 Sept 2023 → 29 Sept 2023

Conference

Conference	32nd International Conference on Artificial Neural Networks, ICANN 2023
Country/Territory	Greece
City	Heraklion
Period	26/09/2023 → 29/09/2023

Keywords

AI robustness
AI stability
AI verifiability
deep learning

Access to Document

10.1007/978-3-031-44207-0_44

Cite this

Bastounis, A., Gorban, A. N., Hansen, A. C., Higham, D. J., Prokhorov, D., Sutton, O., Tyukin, I. Y., & Zhou, Q. (2023). The Boundaries of Verifiable Accuracy, Robustness, and Generalisation in Deep Learning. 530-541. Paper presented at 32nd International Conference on Artificial Neural Networks, ICANN 2023, Heraklion, Greece. Advance online publication. https://doi.org/10.1007/978-3-031-44207-0_44

@conference{35395077e4c94cd580d49402964e05ef,

title = "The Boundaries of Verifiable Accuracy, Robustness, and Generalisation in Deep Learning",

abstract = "In this work, we assess the theoretical limitations of determining guaranteed stability and accuracy of neural networks in classification tasks. We consider classical distribution-agnostic framework and algorithms minimising empirical risks and potentially subjected to some weights regularisation. We show that there is a large family of tasks for which computing and verifying ideal stable and accurate neural networks in the above settings is extremely challenging, if at all possible, even when such ideal solutions exist within the given class of neural architectures.",

keywords = "AI robustness, AI stability, AI verifiability, deep learning",

author = "Alexander Bastounis and Gorban, {Alexander N.} and Hansen, {Anders C.} and Higham, {Desmond J.} and Danil Prokhorov and Oliver Sutton and Tyukin, {Ivan Y.} and Qinghua Zhou",

note = "Funding Information: Acknowledgements. This work is supported by the UKRI, EPSRC [UKRI Turing AI Fellowship ARaISE EP/V025295/2 and UKRI Trustworthy Autonomous Systems Node in Verifiability EP/V026801/2 to I.Y.T., EP/V025295/2 to O.S., A.N.G., and Q.Z., EP/V046527/1 and EP/P020720/1 to D.J.H, EP/V046527/1 to A.B.]. Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 32nd International Conference on Artificial Neural Networks, ICANN 2023 ; Conference date: 26-09-2023 Through 29-09-2023",

year = "2023",

month = sep,

day = "22",

doi = "10.1007/978-3-031-44207-0_44",

language = "English",

pages = "530--541",

}

Bastounis, A, Gorban, AN, Hansen, AC, Higham, DJ, Prokhorov, D, Sutton, O, Tyukin, IY & Zhou, Q 2023, 'The Boundaries of Verifiable Accuracy, Robustness, and Generalisation in Deep Learning', Paper presented at 32nd International Conference on Artificial Neural Networks, ICANN 2023, Heraklion, Greece, 26/09/2023 - 29/09/2023 pp. 530-541. https://doi.org/10.1007/978-3-031-44207-0_44

TY - CONF

T1 - The Boundaries of Verifiable Accuracy, Robustness, and Generalisation in Deep Learning

AU - Bastounis, Alexander

AU - Gorban, Alexander N.

AU - Hansen, Anders C.

AU - Higham, Desmond J.

AU - Prokhorov, Danil

AU - Sutton, Oliver

AU - Tyukin, Ivan Y.

AU - Zhou, Qinghua

N1 - Funding Information: Acknowledgements. This work is supported by the UKRI, EPSRC [UKRI Turing AI Fellowship ARaISE EP/V025295/2 and UKRI Trustworthy Autonomous Systems Node in Verifiability EP/V026801/2 to I.Y.T., EP/V025295/2 to O.S., A.N.G., and Q.Z., EP/V046527/1 and EP/P020720/1 to D.J.H, EP/V046527/1 to A.B.]. Publisher Copyright: © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

PY - 2023/9/22

Y1 - 2023/9/22

N2 - In this work, we assess the theoretical limitations of determining guaranteed stability and accuracy of neural networks in classification tasks. We consider classical distribution-agnostic framework and algorithms minimising empirical risks and potentially subjected to some weights regularisation. We show that there is a large family of tasks for which computing and verifying ideal stable and accurate neural networks in the above settings is extremely challenging, if at all possible, even when such ideal solutions exist within the given class of neural architectures.

AB - In this work, we assess the theoretical limitations of determining guaranteed stability and accuracy of neural networks in classification tasks. We consider classical distribution-agnostic framework and algorithms minimising empirical risks and potentially subjected to some weights regularisation. We show that there is a large family of tasks for which computing and verifying ideal stable and accurate neural networks in the above settings is extremely challenging, if at all possible, even when such ideal solutions exist within the given class of neural architectures.

KW - AI robustness

KW - AI stability

KW - AI verifiability

KW - deep learning

UR - http://www.scopus.com/inward/record.url?scp=85174637708&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-44207-0_44

DO - 10.1007/978-3-031-44207-0_44

M3 - Paper

AN - SCOPUS:85174637708

SP - 530

EP - 541

T2 - 32nd International Conference on Artificial Neural Networks, ICANN 2023

Y2 - 26 September 2023 through 29 September 2023

ER -