Search

Encoding a dictionary into another representation means that all the words can be stored in the dictionary in a more efficient way. In this way, we can complete common operations in dictionaries, such as (1) searching for a word in the dictionary, (2) adding some words to the dictionary, and (3) removing some words from the dictionary, in a shorter time. Binary decision diagrams (BDDs) are one of the most famous representations of such encoding and are widely popular due to their excellent properties. Recently, some people have proposed encoding dictionaries into BDDs and some variants of BDDs and showed that it is feasible. Hence, we further investigate the topic of encoding dictionaries into decision diagrams. Tagged sentential decision diagrams (TSDDs), as one of these variants based on structured decomposition, exploit both the standard and zero-suppressed trimming rules. In this paper, we first introduce how to use Boolean functions to represent dictionary files and then design an algorithm that encodes dictionaries into TSDDs with the help of tries and a decoding algorithm that restores TSDDs to dictionaries. We utilize the help of tries in the encoding algorithm, which greatly accelerates the encoding process. Considering that TSDDs integrate two trimming rules, we believe that using TSDDs to represent dictionaries would be more effective, and the experiments also show this. [ABSTRACT FROM AUTHOR]

Adversarial attacks hamper the functionality and accuracy of deep neural networks (DNNs) by meddling with subtle perturbations to their inputs. In this work, we propose a new mask-based adversarial defense scheme (MAD) for DNNs to mitigate the negative effect from adversarial attacks. Our method preprocesses multiple copies of a potential adversarial image by applying random masking, before the outputs of the DNN on all the randomly masked images are combined. As a result, the combined final output becomes more tolerant to minor perturbations on the original input. Compared with existing adversarial defense techniques, our method does not need any additional denoising structure or any change to a DNN's architectural design. We have tested this approach on a collection of DNN models for a variety of datasets, and the experimental results confirm that the proposed method can effectively improve the defense abilities of the DNNs against all of the tested adversarial attack methods. In certain scenarios, the DNN models trained with MAD can improve classification accuracy by as much as 90 % compared to the original models when given adversarial inputs. [ABSTRACT FROM AUTHOR]