Sparse Misinformation Detector

Huan Wang, Ting Han, Tao Zhang, Hanning Zhou, Yun Fu

Keywords: Misinformation detection, fake news detection, sparse training, network pruning

Abstract: We present Sparse Misinformation Detector (SMD), a new efficient misinformation detection network with regular fine-grained sparsity. We propose two technical components to enable SMD. First, CircuSparsity, a new hardware-friendly sparsity pattern, is introduced for improved training and testing efficiency. Second, through dedicated empirical analyses, we discover that document-level misinformation detection is pretty insensitive to a compact model size, which inspires us to make early exit for the document-level misinformation classifier. With these two techniques, we successfully achieve efficient misinformation detection on both document and event levels with one single model. Empirically, our approach significantly outperforms the original dense misinformation detection network while enjoying 50% to 75% sparsity. Extensive experiments and analyses demonstrate the merits of our method compared to other top-performing counterpart approaches. To our best knowledge, this is the first attempt for efficient misinformation detection from the network sparse training perspective.

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.

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: Yes

Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)

TL;DR: We present an efficient sparse misinformation detector based on a special sparsity pattern (CircuSparsity), with very encouraging performance.