Geometry problem solving (GPS) is a high-level mathematical reasoning requiring the capacities of multi-modal fusion and geometric knowledge application. Recently, neural solvers have shown great potential in GPS but still be short in diagram presentation and modal fusion. In this work, we convert diagrams into basic textual clauses to describe diagram features effectively, and propose a new neural solver called PGPSNet to fuse multi-modal information efficiently. Combining structural and semantic pre-training, data augmentation and self-limited decoding, PGPSNet is endowed with rich knowledge of geometry theorems and geometric representation, and therefore promotes geometric understanding and reasoning. In addition, to facilitate the research of GPS, we build a new large-scale and fine-annotated GPS dataset named PGPS9K, labeled with both fine-grained diagram annotation and interpretable solution program. Experiments on PGPS9K and an existing dataset Geometry3K validate the superiority of our method over the state-of-the-art neural solvers. Our code, dataset and appendix material are available at \url{https://github.com/mingliangzhang2018/PGPS}.

With the global promotion and application of 5G technology, the data transmitted on the network show explosive growth, and the secure sharing of its important data is still one of the research hotspots. Steganography embeds the data that needs to be shared into digital carrier files and transmits it through open channels, which has important applications in protecting data sharing and realizing covert communication. However, the encrypted files generated by traditional steganography are susceptible to compression, cropping, geometric attacks, and man-made destruction in the process of open channel transmission, resulting in data loss, making it difficult for the receiver to correctly extract secret message. Currently, covert communication on blockchain can solve the above problems, but it also brings some new problems such as high computational complexity, low transmission efficiency, and nondetection resistance. Therefore, in this article, a covert communication method based on Bitcoin transactions is proposed. The proposed method first designs the index matrix of the transaction address. Then, the address interaction relationship that carries the secret message through the transaction index matrix is constructed. Finally, the address interaction relationship that carries the secret message is combined with the transaction amount to complete the covert transmission of the secret message on the blockchain environment. The proposed method improves the security and embedding efficiency of covert communication, reduces the number of transactions, and ensures the integrity of extracting secret message. A series of experimental results show that under the condition of ensuring the necessary security, the proposed method retains the strong robustness of the existing blockchain steganography and has strong resistance to detection.

Zhong-Zhi Li, Ming-Liang Zhang, Fei Yin, Cheng-Lin Liu. Findings of the Association for Computational Linguistics ACL 2024. 2024.

Covert communication is a crucial technology that hides information in the redundant structure of the file and transmission through public channel to achieve the secure delivery of information. The blockchain network, with the characteristics of anonymity, decentralization and tamper-proofing, can make up for the shortcomings of multimedia-based covert communication, which include the easy exposure of the identity for both parties, the vulnerability to destruction during communication and the weak robustness of the channel. Therefore, the blockchain network is an ideal channel for covert communication. Nevertheless, the existing covert communication methods face certain challenges based on blockchain, such as the lack of a secure channel for transferring the master key, low embedding capacity, and weak detection resistance. In view of this, this paper proposes a covert communication method based on Shamir threshold and STC mapping, which is suitable for public chain networks. The proposed method first decomposes the master key into sub-keys by introducing Shamir scheme, and the sub-keys are shared with the help of transaction amounts on a blockchain. Then, a mapping relation is established to ensure that the transaction amounts carrying the secret are evenly distributed. Finally, secret information is hidden in the mapping relationship and the transaction amount is interwoven, which is published to the blockchain through transactions to complete covert communication. The introduction of Shamir threshold breaks the limitation that master key cannot be safely transmitted due to the lack of a secure channel in the research of covert communication based on blockchain, thereby enhances the security of the method. Meanwhile, Shamir threshold scheme based on public chain, can solve the issue that the master key cannot be reconstructed due to dishonest participants providing invalid subkeys on traditional network. In addition, the proposed STC mapping can not only improve the detection resistance but also increase the embedding capacity. A series of experimental results illustrate that the proposed method is more resistant to detection, and the embedding efficiency is enhanced by 27.56 times compared with existing public chain-based covert communication methods, effectively reducing the number of transactions and saving resource consumption.

Consumers often have the need to use social media to transmit secret information, but such communication often faces distortion due to lossy compression and limited sending frequency. In order to solve these issues, this paper proposes a covert communication method based on the time modulation of online video bullet comments. Bullet comment is a kind of time-anchored on-screen comment, which is a popular feature for online video platforms. The proposed method converts the absolute time and relative time of the bullet comments to carry the index value of the loaded bullet comments and encrypted characters. The receiver converts the time information of the obtained bullet comments into index values and uses them to extract the secret message. The experimental results show that the proposed method is suitable for multiple online video platforms. Compared with the existing covert communication methods based on social media, the secure operation frequency of the proposed method is at least 9.06 times higher when sending a secret message of the same length. The security of the proposed method is evaluated by the metrics designed according to the characteristics of the large-scale bullet comments.