Learning long-term dependences (LTDs) with recurrent neural networks (RNNs) is challenging due to their limited internal memories. In this paper, we propose a new external memory architecture for RNNs called an addressable and working (EALWM)-augmented RNN. This has two distinct advantages over existing architectures, namely the division of into parts-long-term memory-with both capability learn LTDs without suffering from vanishing gradients necessary assumptions. The experimental results on...

Syntactical rule based approaches for aspect extraction, which are free from expensive manual annotation, promising in practice. These extract aspects mainly through the dependency relations surface sentence structures. However, deep and rich semantic information hidden sentences can help improve is difficult them to capture. In order address problem, this paper first proposes employ Logic Programming explore feasibility of representation, then Deep2S, a hybrid rule-based method performance...

In Q-learning, the reduced chance of converging to optimal policy is partly caused by estimated bias action values. The estimation values usually leads biases like overestimation and underestimation thus it hurts current policy. produced maximization operator are overestimated, which well known as bias. For correcting bias, towards double estimators operator. However, according proposed analysis, performances two operators (the operator) rely on undetermined dynamic environment in results...

Aiming at Named Entity Recognition (NER) in the domain of geographical subject, two different methods are proposed to recognize categories entity: core terms and location. In this paper, conditional random field (CRF) model combining commonly used features is employed. We additionally perform an extensive number experiments verify effectiveness achieve 78.51% 83.10% F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> for entity...

Word representation learned from the analysis of natural language does not usually reflect true semantics words. The paper proposes a new method, named Visually Supervised Word2Vec (VS-Word2Vec) model to achieve relation words that are important in knowledge related tasks. Our method first computes visual feature vector based on deep networks, and then similarity matrix for all words, which we think reflects their semantics. VS-Word2Vec combines CBOW builds an optimization problem jointly...

Value function approximation, such as Q-learning, is widely used in the discrete control rather than continuous one because optimal action setting more easily selected. Optimizing a non-convex optimization problem with respect to complex value function. Some notable studies simplify by assuming quadratic actions or discretizing space. However, performance of output policy will decline if these studies' premises do not hold. In order address problem, we propose framework that combines swarm...

Traditional machine learning algorithms rely on the assumption that all features of a given dataset are available for free. However, in disease diagnosis, acquiring requires certain cost because there many concerns, such as monetary data collection costs, patient discomfort medical procedures, and privacy impacts require careful consideration. Moreover, some examination items grouped real world, lab tests return multiple measurements, resulting naturally features. In this work, to consider...

Comparative questions in Chinese, as a special and complex form of question answering (QA), have their own unique sentence structure, existing methods cannot solve them well. Inspired by cognitive studies on how humans problems, we propose hybrid framework which combines Logic Programming attention based Bi-LSTM. This is decomposed into three consecutive components: 1) identify comparative questions, 2) extract elements from the identified 3) answer factoid containing extracted elements....

Abstract In order to improve the haze removal effect of image, a modified Conditional Generative Adversarial Nets (CGAN) based algorithm is proposed. new algorithm, pre-trained visual geometry group (VGG) model adopted, DenseNet instead traditional U-net as network structure generator, Patch-GAN discriminator, and loss function by total variation regularization gradient. The defogged image can be obtained without estimating projection map related defogging feature. experiments indicate that...

Abstract This letter focuses on the problem of lifelong classification in open world, goal which is to achieve an endless process learning. However, incremental data sets (like streaming data) where new classes may be emerging, are unsuited for classical methods. For addressing this problem, existing methods usually retrain whole observed with complex computation and expensive storage cost. attempts improve performance world decomposes into three subproblems: (1) reject unknown instances,...

Memory-augmented recurrent neural networks (M-RNNs) have demonstrated empirically that they are very attractive for many applications, but a good theoretical understanding of their behaviors is unclear yet. In this paper, three analytical indicators named duration, addressability, and capacity general forms the additional memory in M-RNNs formalized. The analysis results interactions among these reveal it hard an M-RNN to simultaneously provide performance on more than two out indicators....

Si and Ding proposed a stream cipher with two keys (the first the second key) an expected security strength. To further measure security, we analyze by considering selective discrete Fourier spectra attack fast attack. The attacks reveal fact that key is more important than key, is, if leaked out, can be obtained lower time complexity of security. In addition, ability to resist guess-and-determine results show attacker able gain exponentially improved polynomial data complexity. It implies...

该文研究了形如f(x,y)的n+1变元bent函数和半bent函数的二阶非线性度，其中xGF(2n), yGF(2)。首先给出了f(x,y)的2n-1个导数非线性度的精确值；然后推导出了函数f(x,y)的其余2n个导数的非线性度紧下界。进而给出了f(x,y)的二阶非线性度的紧下界。通过比较可知所得下界要优于现有的一般结论。结果表明f(x,y)具有较高的二阶非线性度，可以抵抗二次函数逼近和仿射逼近攻击。

There have been a lot of researches about algebric analysis AES. In this paper, we turned to quantum algorithm analyze security AES-128 against the modified HHL algorithm, which is used get classical solutions multivariate equation system. We constructed two types systems AES, and solved them with several variants algorithms respectively. The resulting complexities involved condition number are given. analyzed reasons for different complexity their solution methods, pointed out that...