Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks: Work-in-Progress

Shreyas Ramakrishna; Zahra Rahiminasab; Arvind Easwaran; Abhishek Dubey

doi:10.1109/EMSOFT51651.2020.9244027

Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks: Work-in-Progress

Shreyas Ramakrishna, Zahra Rahiminasab, Arvind Easwaran, Abhishek Dubey

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

4 Citations (Scopus)

Abstract

Perception-based deep neural networks used in Cyber Physical Systems are known to fail when faced with inputs that are out-of-distribution (ODD). ODD detection is a complex problem as we need to first identify the shift in the test data from the training distribution and then we need to isolate the responsible generative factor(s) (weather, lighting levels, traffic density, etc.), Unlike the state of the art that uses multi-chained one-class classifiers, we propose an efficient single monitor that uses the principle of disentanglement to train the latent space of a variational autoencoder to be sensitive to distribution shifts in different generative factors. We demonstrate our approach using an end-to-end driving controller in the CARLA simulator.

Original language	English
Title of host publication	Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020
Editors	Tulika Mitra, Andreas Gerstlauer
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	40-42
Number of pages	3
ISBN (Electronic)	9781728191959
DOIs	https://doi.org/10.1109/EMSOFT51651.2020.9244027
Publication status	Published - Sept 20 2020
Externally published	Yes
Event	14th Turkish National Software Engineering Symposium, UYMS 2020 - Istanbul, Turkey Duration: Oct 7 2020 → Oct 9 2020

Publication series

Name	Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020

Conference

Conference	14th Turkish National Software Engineering Symposium, UYMS 2020
Country/Territory	Turkey
City	Istanbul
Period	10/7/20 → 10/9/20

Bibliographical note

Publisher Copyright:
© 2020 IEEE.

ASJC Scopus Subject Areas

Software

Keywords

Disentanglement
Inductive Conformal Prediction
Mutual Information Gap
ß-VAE

Access to Document

10.1109/EMSOFT51651.2020.9244027

Cite this

Ramakrishna, S., Rahiminasab, Z., Easwaran, A., & Dubey, A. (2020). Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks: Work-in-Progress. In T. Mitra, & A. Gerstlauer (Eds.), Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020 (pp. 40-42). Article 9244027 (Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/EMSOFT51651.2020.9244027

Ramakrishna, Shreyas ; Rahiminasab, Zahra ; Easwaran, Arvind et al. / Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks : Work-in-Progress. Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020. editor / Tulika Mitra ; Andreas Gerstlauer. Institute of Electrical and Electronics Engineers Inc., 2020. pp. 40-42 (Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020).

@inproceedings{a0f4d2b46dc94ab98ee441b899b1b7e7,

title = "Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks: Work-in-Progress",

abstract = "Perception-based deep neural networks used in Cyber Physical Systems are known to fail when faced with inputs that are out-of-distribution (ODD). ODD detection is a complex problem as we need to first identify the shift in the test data from the training distribution and then we need to isolate the responsible generative factor(s) (weather, lighting levels, traffic density, etc.), Unlike the state of the art that uses multi-chained one-class classifiers, we propose an efficient single monitor that uses the principle of disentanglement to train the latent space of a variational autoencoder to be sensitive to distribution shifts in different generative factors. We demonstrate our approach using an end-to-end driving controller in the CARLA simulator.",

keywords = "Disentanglement, Inductive Conformal Prediction, Mutual Information Gap, {\ss}-VAE",

author = "Shreyas Ramakrishna and Zahra Rahiminasab and Arvind Easwaran and Abhishek Dubey",

note = "Publisher Copyright: {\textcopyright} 2020 IEEE.; 14th Turkish National Software Engineering Symposium, UYMS 2020 ; Conference date: 07-10-2020 Through 09-10-2020",

year = "2020",

month = sep,

day = "20",

doi = "10.1109/EMSOFT51651.2020.9244027",

language = "English",

series = "Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "40--42",

editor = "Tulika Mitra and Andreas Gerstlauer",

booktitle = "Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020",

address = "United States",

}

Ramakrishna, S, Rahiminasab, Z, Easwaran, A & Dubey, A 2020, Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks: Work-in-Progress. in T Mitra & A Gerstlauer (eds), Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020., 9244027, Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020, Institute of Electrical and Electronics Engineers Inc., pp. 40-42, 14th Turkish National Software Engineering Symposium, UYMS 2020, Istanbul, Turkey, 10/7/20. https://doi.org/10.1109/EMSOFT51651.2020.9244027

Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks: Work-in-Progress. / Ramakrishna, Shreyas; Rahiminasab, Zahra; Easwaran, Arvind et al.
Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020. ed. / Tulika Mitra; Andreas Gerstlauer. Institute of Electrical and Electronics Engineers Inc., 2020. p. 40-42 9244027 (Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks

T2 - 14th Turkish National Software Engineering Symposium, UYMS 2020

AU - Ramakrishna, Shreyas

AU - Rahiminasab, Zahra

AU - Easwaran, Arvind

AU - Dubey, Abhishek

PY - 2020/9/20

Y1 - 2020/9/20

N2 - Perception-based deep neural networks used in Cyber Physical Systems are known to fail when faced with inputs that are out-of-distribution (ODD). ODD detection is a complex problem as we need to first identify the shift in the test data from the training distribution and then we need to isolate the responsible generative factor(s) (weather, lighting levels, traffic density, etc.), Unlike the state of the art that uses multi-chained one-class classifiers, we propose an efficient single monitor that uses the principle of disentanglement to train the latent space of a variational autoencoder to be sensitive to distribution shifts in different generative factors. We demonstrate our approach using an end-to-end driving controller in the CARLA simulator.

AB - Perception-based deep neural networks used in Cyber Physical Systems are known to fail when faced with inputs that are out-of-distribution (ODD). ODD detection is a complex problem as we need to first identify the shift in the test data from the training distribution and then we need to isolate the responsible generative factor(s) (weather, lighting levels, traffic density, etc.), Unlike the state of the art that uses multi-chained one-class classifiers, we propose an efficient single monitor that uses the principle of disentanglement to train the latent space of a variational autoencoder to be sensitive to distribution shifts in different generative factors. We demonstrate our approach using an end-to-end driving controller in the CARLA simulator.

KW - Disentanglement

KW - Inductive Conformal Prediction

KW - Mutual Information Gap

KW - ß-VAE

UR - http://www.scopus.com/inward/record.url?scp=85097642048&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85097642048&partnerID=8YFLogxK

U2 - 10.1109/EMSOFT51651.2020.9244027

DO - 10.1109/EMSOFT51651.2020.9244027

M3 - Conference contribution

AN - SCOPUS:85097642048

T3 - Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020

SP - 40

EP - 42

BT - Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020

A2 - Mitra, Tulika

A2 - Gerstlauer, Andreas

PB - Institute of Electrical and Electronics Engineers Inc.

Y2 - 7 October 2020 through 9 October 2020

ER -

Ramakrishna S, Rahiminasab Z, Easwaran A, Dubey A. Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks: Work-in-Progress. In Mitra T, Gerstlauer A, editors, Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020. Institute of Electrical and Electronics Engineers Inc. 2020. p. 40-42. 9244027. (Proceedings of the 2020 International Conference on Embedded Software, EMSOFT 2020). doi: 10.1109/EMSOFT51651.2020.9244027

Efficient Multi-Class Out-of-Distribution Reasoning for Perception Based Networks: Work-in-Progress

Abstract

Publication series

Conference

Bibliographical note

ASJC Scopus Subject Areas

Keywords

Access to Document

Other files and links

Fingerprint

Cite this