In this paper, we introduce the Reinforced Mnemonic Reader for machine reading comprehension tasks, which enhances previous attentive readers in two aspects. First, a reattention mechanism is proposed to refine current attentions by directly accessing to past attentions that are temporally memorized in a multi-round alignment architecture, so as to avoid the problems of attention redundancy and attention deficiency. Second, a new optimization approach, called dynamic-critical reinforcement learning, is introduced to extend the standard supervised method. It always encourages to predict a more acceptable answer so as to address the convergence suppression problem occurred in traditional reinforcement learning algorithms. Extensive experiments on the Stanford Question Answering Dataset (SQuAD) show that our model achieves state-of-the-art results. Meanwhile, our model outperforms previous systems by over 6% in terms of both Exact Match and F1 metrics on two adversarial SQuAD datasets.

A reactive molecular dynamics simulation employing the multistate empirical valence bond (MS-EVB) methodology is reported for the hydration structure of an excess proton in a (6,6) carbon nanotube as well as for the mechanism of proton transport (PT) within the nanoconfined environment. The proton is found to be hydrated in a distorted Zundel cation (H(5)O(2)(+)) form within the one-dimensional, confined water chain. Proton transfer events occur via a "Zundel-Zundel" mechanism through a transient H(7)O(3)(+) intermediate that differs significantly from the "Eigen-Zundel-Eigen" mechanism found in bulk water.

We study the problem of extracting accurate correspondences for point cloud registration. Recent keypoint-free methods have shown great potential through bypassing the detection of repeatable keypoints which is difficult to do especially in low-overlap scenarios. They seek correspondences over downsampled superpoints, which are then propagated to dense points. Superpoints are matched based on whether their neighboring patches overlap. Such sparse and loose matching requires contextual features capturing the geometric structure of the point clouds. We propose Geometric Transformer, or GeoTransformer for short, to learn geometric feature for robust superpoint matching. It encodes pair-wise distances and triplet-wise angles, making it invariant to rigid transformation and robust in low-overlap cases. The simplistic design attains surprisingly high matching accuracy such that no RANSAC is required in the estimation of alignment transformation, leading to 100 times acceleration. Extensive experiments on rich benchmarks encompassing indoor, outdoor, synthetic, multiway and non-rigid demonstrate the efficacy of GeoTransformer. Notably, our method improves the inlier ratio by 18∼31 percentage points and the registration recall by over 7 points on the challenging 3DLoMatch benchmark. Our code and models are available at https://github.com/qinzheng93/GeoTransformer.

Grotthuss shuttling of an excess proton charge defect through hydrogen bonded water networks has long been the focus of theoretical and experimental studies. In this work we show that there is a related process in which water molecules move (“shuttle”) through a hydrated excess proton charge defect in order to wet the path ahead for subsequent proton charge migration. This process is illustrated through reactive molecular dynamics simulations of proton transport through a hydrophobic nanotube, which penetrates through a hydrophobic region. Surprisingly, before the proton enters the nanotube, it starts “shooting” water molecules into the otherwise dry space via Grotthuss shuttling, effectively creating its own water wire where none existed before. As the proton enters the nanotube (by 2–3 Å), it completes the solvation process, transitioning the nanotube to the fully wet state. By contrast, other monatomic cations (e.g., K+) have just the opposite effect, by blocking the wetting process and making the nanotube even drier. As the dry nanotube gradually becomes wet when the proton charge defect enters it, the free energy barrier of proton permeation through the tube via Grotthuss shuttling drops significantly. This finding suggests that an important wetting mechanism may influence proton translocation in biological systems, i.e., one in which protons “create” their own water structures (water “wires”) in hydrophobic spaces (e.g., protein pores) before migrating through them. An existing water wire, e.g., one seen in an X-ray crystal structure or MD simulations without an explicit excess proton, is therefore not a requirement for protons to transport through hydrophobic spaces.