Authors: Hongru Li, Wentao Yu, Hengtao He, Jiawei Shao, Shenghui Song, Jun Zhang, Khaled B. Letaief

Published on: May 21, 2023

Impact Score: 8.22

Arxiv code: Arxiv:2305.12423

Summary

• What is new: A new approach based on the information bottleneck framework to effectively handle unknown out-of-distribution data in task-oriented communication systems.
• Why this is important: Existing task-oriented communication systems struggle with open-world scenarios where they encounter data distributions different from what they were trained on.
• What the research proposes: A class conditional information bottleneck (CCIB) approach that focuses on extracting distinguishable, compact, and informative features from in-distribution data, improving the system’s ability to detect out-of-distribution data.
• Results: Simulation results show that the proposed CCIB approach detects out-of-distribution data more efficiently than existing baselines and methods, without harming the rate-distortion tradeoff.

Technical Details

Technological frameworks used: Information Bottleneck (IB)

Models used: Deep Learning (DL)-based systems

Data used: In-distribution and out-of-distribution datasets

Potential Impact

Communication network providers, IoT platform developers, and companies leveraging DL for data processing and analytics could benefit or need to adapt.

Want to implement this idea in a business?

We have generated a startup concept here: DataGuardAI.