What question did this study set out to answer?

To develop a compression framework combining quantization, pruning, and knowledge distillation for latent-space variational autoencoder OOD detectors on resource-constrained embedded platforms.

August 21, 2024

Compressing VAE-Based Out-of-Distribution Detectors for Embedded Deployment

View Full Paper

Ask AI

Bookmark

Authors

ABAditya Bansal MYMichael Yuhas AEArvind Easwaran

Discussion

Member takes

Overview

Experimental study demonstrates model compression accelerates VAE-based anomaly detection on embedded hardware, suggesting real-time safety monitoring is feasible under resource constraints.

Key Points

To develop a compression framework combining quantization, pruning, and knowledge distillation for latent-space variational autoencoder OOD detectors on resource-constrained embedded platforms.
Applied a combined pipeline of weight quantization, network pruning, and knowledge distillation to variational autoencoders performing OOD detection in latent space.
Evaluated the compression methodology on two existing OOD detection algorithms deployed on an embedded Jetson Nano CPU and GPU.
Reduced inference time by 20% on the embedded GPU and 28% on the embedded CPU relative to uncompressed baselines.
Maintained OOD detection efficacy with AUROC staying within 5% of the baseline models despite an increase in VAE test loss.

Cite This Study

Bansal et al. (2024) studied this question.

synapsesocial.com/papers/6a0eb314c125403562229dc0 https://doi.org/10.1109/rtcsa62462.2024.00015

View Full Paper

Ask AI

Bookmark

August 21, 2024

Compressing VAE-Based Out-of-Distribution Detectors for Embedded Deployment

View Full Paper

Ask AI

Bookmark

Authors

ABAditya Bansal MYMichael Yuhas AEArvind Easwaran

Discussion

Member takes

Overview

Experimental study demonstrates model compression accelerates VAE-based anomaly detection on embedded hardware, suggesting real-time safety monitoring is feasible under resource constraints.

Key Points

To develop a compression framework combining quantization, pruning, and knowledge distillation for latent-space variational autoencoder OOD detectors on resource-constrained embedded platforms.
Applied a combined pipeline of weight quantization, network pruning, and knowledge distillation to variational autoencoders performing OOD detection in latent space.
Evaluated the compression methodology on two existing OOD detection algorithms deployed on an embedded Jetson Nano CPU and GPU.
Reduced inference time by 20% on the embedded GPU and 28% on the embedded CPU relative to uncompressed baselines.
Maintained OOD detection efficacy with AUROC staying within 5% of the baseline models despite an increase in VAE test loss.

Cite This Study

Bansal et al. (2024) studied this question.

synapsesocial.com/papers/6a0eb314c125403562229dc0 https://doi.org/10.1109/rtcsa62462.2024.00015

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

View Full Paper

Ask AI

Bookmark