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 -