Transformer-based pre-trained models, such as BERT, have achieved remarkable results on machine reading comprehension. However, due to the constraint of encoding length (e.g., 512 WordPiece tokens), a long document is usually split into multiple chun
ks that are independently read. It results in the reading field being limited to individual chunks without information collaboration for long document machine reading comprehension. To address this problem, we propose RoR, a read-over-read method, which expands the reading field from chunk to document. Specifically, RoR includes a chunk reader and a document reader. The former first predicts a set of regional answers for each chunk, which are then compacted into a highly-condensed version of the original document, guaranteeing to be encoded once. The latter further predicts the global answers from this condensed document. Eventually, a voting strategy is utilized to aggregate and rerank the regional and global answers for final prediction. Extensive experiments on two benchmarks QuAC and TriviaQA demonstrate the effectiveness of RoR for long document reading. Notably, RoR ranks 1st place on the QuAC leaderboard (https://quac.ai/) at the time of submission (May 17th, 2021).
We study the impact of different interlayers and ferroelectric materials on charge trapping during the endurance fatigue of Si FeFET with TiN/HfxZr1-xO2/interlayer/Si (MFIS) gate stack. We have fabricated FeFET devices with different interlayers (SiO
2 or SiON) and HfxZr1-xO2 materials (x=0.75, 0.6, 0.5), and directly extracted the charge trapping during endurance fatigue. We find that: 1) The introduction of the N element in the interlayer suppresses charge trapping and defect generation, and improves the endurance characteristics. 2) As the spontaneous polarization (Ps) of the HfxZr1-xO2 decreases from 25.9 {mu}C/cm2 (Hf0.5Zr0.5O2) to 20.3 {mu}C/cm2 (Hf0.6Zr0.4O2), the charge trapping behavior decreases, resulting in the slow degradation rate of memory window (MW) during program/erase cycling; in addition, when the Ps further decreases to 8.1 {mu}C/cm2 (Hf0.75Zr0.25O2), the initial MW nearly disappears (only ~0.02 V). Thus, the reduction of Ps could improve endurance characteristics. On the contract, it can also reduce the MW. Our work helps design the MFIS gate stack to improve endurance characteristics.
The online recruitment matching system has been the core technology and service platform in CareerBuilder. One of the major challenges in an online recruitment scenario is to provide good matches between job posts and candidates using a recommender s
ystem on the scale. In this paper, we discussed the techniques for applying an embedding-based recommender system for the large scale of job to candidates matching. To learn the comprehensive and effective embedding for job posts and candidates, we have constructed a fused-embedding via different levels of representation learning from raw text, semantic entities and location information. The clusters of fused-embedding of job and candidates are then used to build and train the Faiss index that supports runtime approximate nearest neighbor search for candidate retrieval. After the first stage of candidate retrieval, a second stage reranking model that utilizes other contextual information was used to generate the final matching result. Both offline and online evaluation results indicate a significant improvement of our proposed two-staged embedding-based system in terms of click-through rate (CTR), quality and normalized discounted accumulated gain (nDCG), compared to those obtained from our baseline system. We further described the deployment of the system that supports the million-scale job and candidate matching process at CareerBuilder. The overall improvement of our job to candidate matching system has demonstrated its feasibility and scalability at a major online recruitment site.
We investigate the charge trapping during endurance fatigue of FeFET with TiN/Hf0.5Zr0.5O2/SiO2/Si (MFIS) gate structure. We propose a method of experimentally extracting the number of trapped charges during the memory operation, by measuring the cha
rges in the metal gate and Si substrate. We verify that the amount of trapped charges increases during the endurance fatigue process. This is the first time that the trapped charges are directly experimentally extracted and verified to increase during endurance fatigue. Moreover, we model the interplay between the trapped charges and ferroelectric polarization switching during endurance fatigue. Through the consistency of experimental results and simulated data, we demonstrate that as the memory window decreases: 1) The ferroelectric characteristic of Hf0.5Zr0.5O2 is not degraded. 2) The trap density in the upper bandgap of the gate stacks increases. 3) The reason for memory window decrease is increased trapped electrons after program operation but not related to hole trapping/de-trapping. Our work is helpful to study the charge trapping behavior of FeFET and the related endurance fatigue process.
Keyphrases, that concisely summarize the high-level topics discussed in a document, can be categorized into present keyphrase which explicitly appears in the source text, and absent keyphrase which does not match any contiguous subsequence but is hig
hly semantically related to the source. Most existing keyphrase generation approaches synchronously generate present and absent keyphrases without explicitly distinguishing these two categories. In this paper, a Select-Guide-Generate (SGG) approach is proposed to deal with present and absent keyphrase generation separately with different mechanisms. Specifically, SGG is a hierarchical neural network which consists of a pointing-based selector at low layer concentrated on present keyphrase generation, a selection-guided generator at high layer dedicated to absent keyphrase generation, and a guider in the middle to transfer information from selector to generator. Experimental results on four keyphrase generation benchmarks demonstrate the effectiveness of our model, which significantly outperforms the strong baselines for both present and absent keyphrases generation. Furthermore, we extend SGG to a title generation task which indicates its extensibility in natural language generation tasks.
We study the average quantum coherence over the pure state decompositions of a mixed quantum state. An upper bound of the average quantum coherence is provided and sufficient conditions for the saturation of the upper bound are shown. These sufficien
t conditions always hold for two and three dimensional systems. This provides a tool to estimate the average coherence experimentally by measuring only the diagonal elements, which remarkably requires less measurements compared with state tomography. We then describe the pure state decompositions of qubit state in Bloch sphere geometrically. For any given qubit state, the optimal pure state decomposition achieving the maximal average quantum coherence as well as three other pure state decompositions are shown in the Bloch sphere. The order relations among their average quantum coherence are invariant for any coherence measure. The results presented in this paper are universal and suitable for all coherence measures.
Coherence and entanglement are fundamental concepts in resource theory. The coherence (entanglement) of assistance is the coherence (entanglement) that can be extracted assisted by another party with local measurement and classical communication. We
introduce and study the general coherence of assistance. First, in terms of real symmetric concave functions on the probability simplex, the coherence of assistance and the entanglement of assistance are shown to be in one-to-one correspondence. We then introduce two classes of quantum states: the assisted maximally coherent states and the assisted maximally entangled states. They can be transformed into maximally coherent or entangled pure states with the help of another party using local measurement and classical communication. We give necessary conditions for states to be assisted maximally coherent or assisted maximally entangled. Based on these, a unified framework between coherence and entanglement including coherence (entanglement) measures, coherence (entanglement) of assistance, coherence (entanglement) resources is proposed. Then we show that the coherence of assistance as well as entanglement of assistance are strictly larger than the coherence of convex roof and entanglement of convex roof for all full rank density matrices. So all full rank quantum states are distillable in the assisted coherence distillation.
This article investigates the energy efficiency issue in non-orthogonal multiple access (NOMA)-enhanced Internet-of-Things (IoT) networks, where a mobile unmanned aerial vehicle (UAV) is exploited as a flying base station to collect data from ground
devices via the NOMA protocol. With the aim of maximizing network energy efficiency, we formulate a joint problem of UAV deployment, device scheduling and resource allocation. First, we formulate the joint device scheduling and spectrum allocation problem as a three-sided matching problem, and propose a novel low-complexity near-optimal algorithm. We also introduce the novel concept of `exploration into the matching game for further performance improvement. By algorithm analysis, we prove the convergence and stability of the final matching state. Second, in an effort to allocate proper transmit power to IoT devices, we adopt the Dinkelbachs algorithm to obtain the optimal power allocation solution. Furthermore, we provide a simple but effective approach based on disk covering problem to determine the optimal number and locations of UAVs stop points to ensure that all IoT devices can be fully covered by the UAV via line-of-sight (LoS) links for the sake of better channel condition. Numerical results unveil that: i) the proposed joint UAV deployment, device scheduling and resource allocation scheme achieves much higher EE compared to predefined stationary UAV deployment case and fixed power allocation scheme, with acceptable complexity; and ii) the UAV-aided IoT networks with NOMA greatly outperforms the OMA case in terms of number of accessed devices.
As one of the most important and famous applications of blockchain technology, cryptocurrency has attracted extensive attention recently. Empowered by blockchain technology, all the transaction records of cryptocurrencies are irreversible and recorde
d in the blocks. These transaction records containing rich information and complete traces of financial activities are publicly accessible, thus providing researchers with unprecedented opportunities for data mining and knowledge discovery in this area. Networks are a general language for describing interacting systems in the real world, and a considerable part of existing work on cryptocurrency transactions is studied from a network perspective. This survey aims to analyze and summarize the existing literature on analyzing and understanding cryptocurrency transactions from a network perspective. Aiming to provide a systematic guideline for researchers and engineers, we present the background information of cryptocurrency transaction network analysis and review existing research in terms of three aspects, i.e., network modeling, network profiling, and network-based detection. For each aspect, we introduce the research issues, summarize the methods, and discuss the results and findings given in the literature. Furthermore, we present the main challenges and several future directions in this area.
Designed for commercial decentralized applications (DApps), EOSIO is a Delegated Proof-of-Stake (DPoS) based blockchain system. It has overcome some shortages of the traditional blockchain systems like Bitcoin and Ethereum with its outstanding featur
es (e.g., free for usage, high throughput and eco-friendly), and thus becomes one of the mainstream blockchain systems. Though there exist billions of transactions in EOSIO, the ecosystem of EOSIO is still relatively unexplored. To fill this gap, we conduct a systematic graph analysis on the early EOSIO by investigating its four major activities, namely account creation, account vote, money transfer and contract authorization. We obtain some novel observations via graph metric analysis, and our results reveal some abnormal phenomenons like voting gangs and sham transactions.
