InfoNet: Missing Information Retrieval in Multi-Stream Sensing Systems
Keywords: Early attention, Feature Recovery, Information theory, Entropy, Multi-stream sensing system
TL;DR: InfoNet recovers missing input stream information from the feature domain by estimating information entropy at every element of the input feature, interpolating the missing information, and performing the guided replacement.
Abstract: Faulty sensors in a multiple input stream setup are more prone to corrupted input data streams, hindering the performance of Deep Neural Networks (DNN), which focus on deducing information from data. However, the relevant information among multiple input streams has correlations and contains mutual information. This paper utilizes this opportunity to retrieve perturbed information caused by corrupted input streams. We propose InfoNet, which estimates the information entropy at every element of the input feature to the network and retrieves the missing information in the input feature matrix. Finally, using the estimated information entropy and retrieved data, we introduce a novel guided replacement procedure to recover the complete information that is the input to the downstream DNN task. We evaluate the proposed algorithm for sound localization where audio streams from the microphone array are corrupted. We have recovered the performance drop due to the corrupted input stream and reduced the localization error with non-corrupted input streams. Finally, we assess the potential of using the proposed algorithm for retrieving information in other sensing modalities, e.g., wireless signal-based source localization.
Primary Area: representation learning for computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 5962
Loading