Automatically recovering the original structure of tables from unstructured images is a challenging task, combining techniques from computer vision (CV) and natural language processing (NLP). Unfortunately, common feature extraction methods, naive fusion strategies, and rigid inductive biases have become roadblocks to the effective improvement of previous approaches. Distinguished from other modes of data representation, tables consist of many dispersed cells that are interdependent. Therefore, in this paper, we aim to propose a novel approach for recognizing table structures by mining the special properties of tables. The method begins by utilizing the adaptive fusion method to fuse visual and textual features acquired through a two-stream network. In the second stage, the layout features will be seamlessly integrated using a Kronecker-based strategy. The table elements with multi-modality features are then modeled based on spatial relationships. Interactions among them are established by a hybrid contextual aggregator that allows message passing at both local and global levels. Finally, table structure recognition is achieved by predicting the relationship between elements. We meticulously evaluate the proposed approach on various public datasets, including ICDAR2013, UNLV, WTW, SciTSR, and SciTSR-COMP, as well as a more complicated private dataset. The proposed method performs excellently on these datasets.

Origami has been proved as a valuable tool to develop deployable, multifunctional, and tunable devices for diverse engineering applications. Inspired by the Kresling origami, this article presents a soft pneumatic actuator with vacuum control to realize the compound motion of twist and contraction. The twist direction varies as the crease angle changes and it offers pure contraction after combining two actuators with reverse creases. Our actuator can lift 180 folds of its weight and achieve the contraction ratio of 47%, while the response time is below 0.3 s. The origami-based actuator can realize an adjustable bistable property, leading to the possibility for external stimulus detection and mechanical memory devices, where the sensitivity can be regulated by the initial vacuum pressure. Through several actuators, we demonstrate the reconfigurable crawling robots capable of worm-like creep, alternating-step locomotion as well as obstacle detection. After activating all the actuators, the four-actuator robot with two working modes is validated to perform agilely for confined space navigation, showing our actuator has the potential in soft robotics applications.

Return-Address Stack (RAS) has been widely used in modern high-performance microprocessor to predict the return address of a routine. But speculative execution frequently corrupts the RAS, making repair mechanisms necessary. In this paper, based on analysis of RAS activities after misprediction and current RAS repair mechanisms, an effective return address prediction structure called Self-Aligning Return-Address Stack (SARAS) is proposed. SARAS is composed of a classical RAS, a self-aligning queue and a top-of-stack counter. The self-aligning queue records the contents popped from RAS for recovery. The top-of-stack counter tracks the latest correct position of the top-of-stack pointer. The experiment results show that SARAS can reduces hardware-cost and enhances prediction accuracy, with a 32-entry stack, the average prediction accuracy of SARAS is nearly 100%, improving performance by 3.91% up to 11.27% over the conventional checkpointing repair mechanisms for RAS. At the same hardware budget, this speedup is even higher.