黑客帝国酬劳(黑客的工资)

图灵之心(TuringKey)目前已实现“机器学习”，”自然语言处理”，”人工智能”，”机器视觉”，”机器人”五个方向的论文收集整理工作，每日更新T-2日的论文信息，包括标题，作者，研究方向，英文摘要，中文摘要（机器翻译）。及时获取论文信息，请关注图灵之心微信公众号：TuringKey。

图灵之心(TuringKey)目前已实现“机器学习”，”自然语言处理”，”人工智能”，”机器视觉”，”机器人”五个方向的论文收集整理工作，每日更新T-2日的论文信息，包括标题，作者，研究方向，英文摘要，中文摘要（机器翻译）。及时获取论文信息，请关注图灵之心微信公众号：TuringKey。

Paper 1 : Stability of Topic Modeling via Matrix Factorization

Authors: Belford Mark, Mac Namee Brian, Greene Derek

Subject: Information Retrieval, Computation and Language, Learning, Machine Learning

Submitted Date: 20170223

Abstract: Topic models can provide us with an insight into the underlying latentstructure of a large corpus of documents. A range of methods have been proposedin the literature, including probabilistic topic models and techniques based onmatrix factorization. However, in both cases, standard implementations rely onstochastic elements in their initialization phase, which can potentially leadto different results being generated on the same corpus when using the sameparameter values. This corresponds to the concept of “instability” which haspreviously been studied in the context of $k$-means clustering. In manyapplications of topic modeling, this problem of instability is not consideredand topic models are treated as being definitive, even though the results maychange considerably if the initialization process is altered. In this paper wedemonstrate the inherent instability of popular topic modeling approaches,using a number of new measures to assess stability. To address this issue inthe context of matrix factorization for topic modeling, we propose the use ofensemble learning strategies. Based on experiments performed on annotated textcorpora, we show that a K-Fold ensemble strategy, combining both ensembles andstructured initialization, can significantly reduce instability, whilesimultaneously yielding more accurate topic models.

展开全文

Abstract_CN: 主题模型可以为我们提供了一个洞察大型语料库的文件的latentstructure。一系列的方法已被提出，文学，包括概率主题模型和基于矩阵分解技术。然而，在这两种情况下，标准的实现依赖随机元素的初始化阶段，这可能导致在使用时sameparameter值相同的主体产生不同的结果。这相当于“不稳定”，此前一直在k均值聚类的语境研究。在主题建模中，这一问题不稳定不考虑主题模型的处理是明确的，尽管结果可能会改变很大，如果初始化过程的改变。在本文中，我们展示的流行主题建模方法存在固有的不稳定性，采用了一些新的措施来评估稳定性。为了解决这一问题的背景下，主题建模矩阵分解，提出了利用集成学习策略。基于注释的textcorpora进行实验，我们证明了k的集成策略，结合系统和结构化的初始化，可以显著减少不稳定，同时产生更精确的主题模型。

View More Information

Paper 2 : Causal Discovery Using Proxy Variables

Authors: Rojas-Carulla Mateo, Baroni Marco, Lopez-Paz David

Subject: Machine Learning, Learning

Submitted Date: 20170223

Abstract: Discovering causal relations is fundamental to reasoning and intelligence. Inparticular, observational causal discovery algorithms estimate the cause-effectrelation between two random entities $X$ and $Y$, given $n$ samples from$P(X,Y)$.

In this paper, we develop a framework to estimate the cause-effect relationbetween two static entities $x$ and $y$: for instance, an art masterpiece $x$and its fraudulent copy $y$. To this end, we introduce the notion of proxyvariables, which allow the construction of a pair of random entities $(A,B)$from the pair of static entities $(x,y)$. Then, estimating the cause-effectrelation between $A$ and $B$ using an observational causal discovery algorithmleads to an estimation of the cause-effect relation between $x$ and $y$. Forexample, our framework detects the causal relation between unprocessedphotographs and their modifications, and orders in time a set of shuffledframes from a video.

As our main case study, we introduce a human-elicited dataset of 10,000 pairsof casually-linked pairs of words from natural language. Our methods discover75% of these causal relations. Finally, we discuss the role of proxy variablesin machine learning, as a general tool to incorporate static knowledge intoprediction tasks.

Abstract_CN: 发现因果关系是推理和智力的基础。特别是，观察因果发现算法估计两随机单位X和Y元美元之间的原因效果关系，鉴于美元从$ P n（x，y）样品美元美元。

在本文中，我们开发了一个框架来估计因果关系两个静态实体X和Y元美元：例如，一个艺术的杰作，其欺诈的副本$ X $ Y $。为此，我们引入proxyvariables的概念，允许建设一对随机单位（A，B）美元美元从对静态实体（x，y）美元美元。然后，估计原因效果关系美元之间，美元和$ B $用观察的因果关系发现algorithmleads在X和Y元的因果关系估计美元。例如，我们的框架检测unprocessedphotographs和修改之间的因果关系，从视频的时间一套shuffledframes订单。

作为主要的案例研究中，我们引入了一个人类引起的数据集的10000对随便链接从自然语言的词。我们的方法discover75 %这些因果关系。最后，我们讨论了机器学习的代理变量的作用，是将静态的知识intoprediction任务的通用工具。

View More Information

Paper 3 : Online Multiclass Boosting

Authors: Jung Young Hun, Tewari Ambuj

Subject: Machine Learning, Learning

Submitted Date: 20170223

Abstract: Recent work has extended the theoretical analysis of boosting algorithms tomulticlass problems and online settings. However, the multiclass extension isin the batch setting and the online extensions only consider binaryclassification. To the best of our knowledge, there exists no framework toanalyze online boosting algorithms for multiclass classification. We fill thisgap in the literature by defining, and justifying, a weak learning conditionfor online multiclass boosting. We also provide an algorithm called onlinemulticlass boost-by-majority to optimally combine weak learners in our setting.

Abstract_CN: 最近的工作已经扩展了理论分析tomulticlass Boosting算法问题和在线设置。然而，多类扩展在批量设置、在线扩展只考虑两类。据我们所知，不存在框架分析的在线Boosting的多类分类算法。我们在文献中填补这一空白的定义，并证明，弱学习条件在线多类Boosting。我们还提供了一个算法称为onlinemulticlass Boost-by-majority最佳结合弱的学习者，在我们的设置。

View More Information

Paper 4 : Rotting Bandits

Authors: Levine Nir, Crammer Koby, Mannor Shie

Subject: Machine Learning, Learning

Submitted Date: 20170223

Abstract: The Multi-Armed Bandits (MAB) framework highlights the tension betweenacquiring new knowledge (Exploration) and leveraging available knowledge(Exploitation). In the classical MAB problem, a decision maker must choose anarm at each time step, upon which she receives a reward. The decision maker’sobjective is to maximize her cumulative expected reward over the time horizon.The MAB problem has been studied extensively, specifically under the assumptionof the arms’ rewards distributions being stationary, or quasi-stationary, overtime. We consider a variant of the MAB framework, which we termed\textit{Rotting Bandits}, where each arm’s expected reward decays as a functionof the number of times it has been pulled. We are motivated by many real-worldscenarios such as online advertising, content recommendation, crowdsourcing,and more. We present algorithms, accompanied by simulations, and derivetheoretical guarantees.

Abstract_CN: 多武装土匪（MAB）框架突出了紧张betweenacquiring新知识（探索），利用现有的知识（开发）。在经典的单克隆抗体的问题，决策者必须选择在每一个时间步也，在她接受的奖励。决策maker’sobjective是最大化自己的累积奖励预期的时间跨度。单克隆抗体的问题已得到了广泛的研究，特别是在考虑了武器的回报分布是固定的，或准静止，加班。我们认为单抗框架的一个变种，我们称之为“系统{土匪}腐烂，每个臂的预期报酬衰变为一个功能的时候已经把数。我们的动机是许多房worldscenarios如在线广告，内容推荐，众包，和更多。我们提出的算法，附有模拟，和derivetheoretical担保。

View More Information

Paper 5 : Sobolev Norm Learning Rates for Regularized Least-Squares Algorithm

Authors: Fischer Simon, Steinwart Ingo

Subject: Machine Learning

Submitted Date: 20170223

Abstract: Learning rates for regularized least-squares algorithms are in most casesexpressed with respect to the excess risk, or equivalently, the $L2$-norm. Forsome applications, however, guarantees with respect to stronger norms such asthe $L\infty$-norm, are desirable. We address this problem by establishinglearning rates for a continuous scale of norms between the $L2$- and the RKHSnorm. As a byproduct we derive $L\infty$-norm learning rates, and in the caseof Sobolev RKHSs we actually obtain Sobolev norm learning rates, which may alsoimply $L_\infty$-norm rates for some derivatives. In all cases, we do not needto assume the target function to be contained in the used RKHS. Finally, weshow that in many cases the derived rates are minimax optimal.

Abstract_CN: 正则化最小二乘算法的学习速率是大多数例可见皮肤相对于过剩的风险，或者说，l2美元美元-规范。对于某些应用，然而，保证相对于美元的更强有力的规范等l\infty $规范，是可取的。我们解决这个问题的establishinglearning率连续尺度的l2美元美元和rkhsnorm规范之间。作为一个副产品，我们得到l美元\infty $规范学习率，和的情况下，我们实际上获得Sobolev范数Sobolev RKHSs学习率，这可能alsoimply $ l_ \infty $一些衍生品规范率。在所有的情况下，我们不需要假设被包含在应用算法的目标函数。最后，我们表明，在许多情况下，派生率极小优化。

View More Information

Paper 6 : A minimax and asymptotically optimal algorithm for stochastic bandits

Authors: Ménard Pierre, Garivier Aurélien

Subject: Machine Learning, Learning, Statistics Theory

Submitted Date: 20170223

Abstract: We propose the kl-UCB ++ algorithm for regret minimization in stochasticbandit models with exponential families of distributions. We prove that it issimultaneously asymptotically optimal (in the sense of Lai and Robbins’ lowerbound) and minimax optimal. This is the first algorithm proved to enjoy thesetwo properties at the same time. This work thus merges two different lines ofresearch, with simple proofs involving no complexity overhead.

Abstract_CN: 我们提出了用分布指数的家庭在stochasticbandit模型遗憾最小化KL UCB C++算法。我们证明，这同时也是渐近最优的（在Lai和罗宾斯下界感）和极大极小优化。这是第一个算法证明享受这两特性的同时。这项工作从而将两条线的研究，以证明不涉及复杂的开销。

View More Information

Paper 7 : Spectral Clustering using PCKID - A Probabilistic Cluster Kernel for Incomplete Data

Authors: Løkse Sigurd, Bianchi Filippo Maria, Salberg Arnt-Børre, Jenssen Robert

Subject: Machine Learning

Submitted Date: 20170223

Abstract: In this paper, we propose PCKID, a novel, robust, kernel function forspectral clustering, specifically designed to handle incomplete data. Bycombining posterior distributions of Gaussian Mixture Models for incompletedata on different scales, we are able to learn a kernel for incomplete datathat does not depend on any critical hyperparameters, unlike the commonly usedRBF kernel. To evaluate our method, we perform experiments on two realdatasets. PCKID outperforms the baseline methods for all fractions of missingvalues and in some cases outperforms the baseline methods with up to 25percentage points.

Abstract_CN: 在本文中，我们提出了一个新的，强大的，pckid，核函数的降维聚类，专门用来处理不完整数据。通过后验分布的高斯混合模型对不同尺度上的不完全的数据像，我们能够学习一种不完全数据，内核不依赖任何的关键参数，不同于一般usedrbf内核。为了评估我们的方法，我们在两个真实数据集上进行实验。pckid优于基线的方法，所有的分数missingvalues和在某些情况下优于基线的方法，以25percentage点。

View More Information

Paper 8 : Automatic Representation for Lifetime Value Recommender Systems

Authors: Hallak Assaf, Mansour Yishay, Yom-Tov Elad

Subject: Machine Learning, Learning

Submitted Date: 20170223

Abstract: Many modern commercial sites employ recommender systems to propose relevantcontent to users. While most systems are focused on maximizing the immediategain (clicks, purchases or ratings), a better notion of success would be thelifetime value (LTV) of the user-system interaction. The LTV approach considersthe future implications of the item recommendation, and seeks to maximize thecumulative gain over time. The Reinforcement Learning (RL) framework is thestandard formulation for optimizing cumulative successes over time. However, RLis rarely used in practice due to its associated representation, optimizationand validation techniques which can be complex. In this paper we propose a newarchitecture for combining RL with recommendation systems which obviates theneed for hand-tuned features, thus automating the state-space representationconstruction process. We analyze the practical difficulties in this formulationand test our solutions on batch off-line real-world recommendation data.

Abstract_CN: 许多现代商业网站使用推荐系统提出相关的内容给用户。虽然大多数系统都集中在最大化的immediategain（点击、购买或评级），一个更好的概念，成功将终身价值（LTV）的用户系统的相互作用。LTV方法考虑未来影响的项目推荐，并旨在最大限度地提高增益随时间累积。强化学习（RL）框架是随着时间累积的成功标准配方的优化。然而，在实践中，由于相关RLIs表示很少使用，优化和验证技术可复杂。在本文中，我们提出了一个RL与推荐系统这就需要手工调整的功能相结合的架构，从而实现自动化的状态空间representationconstruction过程。我们分析这和实际困难考验我们的批量离线真实推荐数据解决方案。

View More Information

Paper 9 : Consistent On-Line Off-Policy Evaluation

Authors: Hallak Assaf, Mannor Shie

Subject: Machine Learning, Learning

Submitted Date: 20170223

Abstract: The problem of on-line off-policy evaluation (OPE) has been actively studiedin the last decade due to its importance both as a stand-alone problem and as amodule in a policy improvement scheme. However, most Temporal Difference (TD)based solutions ignore the discrepancy between the stationary distribution ofthe behavior and target policies and its effect on the convergence limit whenfunction approximation is applied. In this paper we propose the ConsistentOff-Policy Temporal Difference (COP-TD($\lambda$, $\beta$)) algorithm thataddresses this issue and reduces this bias at some computational expense. Weshow that COP-TD($\lambda$, $\beta$) can be designed to converge to the samevalue that would have been obtained by using on-policy TD($\lambda$) with thetarget policy. Subsequently, the proposed scheme leads to a related andpromising heuristic we call log-COP-TD($\lambda$, $\beta$). Both algorithmshave favorable empirical results to the current state of the art on-line OPEalgorithms. Finally, our formulation sheds some new light on the recentlyproposed Emphatic TD learning.

Abstract_CN: 在线的政策评价问题（OPE）一直在积极研究，由于其作为一个独立的问题和策略改进方案作为模块的重要性，过去十年。然而，大多数时间差异（TD）为基础的解决方案忽略的平稳分布的行为和目标的政策和对收敛极限whenfunction逼近效果之间的差异是应用。在本文中，我们提出了consistentoff政策的时间差（cop-td（$ \λ，β））算法彼此这一问题和一些计算费用降低了这种偏见。我们发现cop-td（$ \λ，β）可以收敛到相同的价值将由政策采用TD获得（$ \λ）与目标政策。随后，该方案导致了相关的和启发我们的通话记录警察TD（$ \拉姆达$，$ \测试）。这两种算法都具有良好的实证结果对艺术在线opealgorithms现状。最后，我们制定了新的光对新近提出的强调TD学习。

View More Information

Paper 10 : Scalable Inference for Nested Chinese Restaurant Process Topic Models

Authors: Chen Jianfei, Zhu Jun, Lu Jie, Liu Shixia

Subject: Machine Learning, Distributed, Parallel, and Cluster Computing, Information Retrieval, Learning

Submitted Date: 20170223

Abstract: Nested Chinese Restaurant Process (nCRP) topic models are powerfulnonparametric Bayesian methods to extract a topic hierarchy from a given textcorpus, where the hierarchical structure is automatically determined by thedata. Hierarchical Latent Dirichlet Allocation (hLDA) is a popular instance ofnCRP topic models. However, hLDA has only been evaluated at small scale,because the existing collapsed Gibbs sampling and instantiated weightvariational inference algorithms either are not scalable or sacrifice inferencequality with mean-field assumptions. Moreover, an efficient distributedimplementation of the data structures, such as dynamically growing countmatrices and trees, is challenging.

In this paper, we propose a novel partially collapsed Gibbs sampling (PCGS)algorithm, which combines the advantages of collapsed and instantiated weightalgorithms to achieve good scalability as well as high model quality. Aninitialization strategy is presented to further improve the model quality.Finally, we propose an efficient distributed implementation of PCGS throughvectorization, pre-processing, and a careful design of the concurrent datastructures and communication strategy.

Empirical studies show that our algorithm is 111 times more efficient thanthe previous open-source implementation for hLDA, with comparable or evenbetter model quality. Our distributed implementation can extract 1,722 topicsfrom a 131-million-document corpus with 28 billion tokens, which is 4-5 ordersof magnitude larger than the previous largest corpus, with 50 machines in 7hours.

Abstract_CN: 嵌套的中国餐馆过程（NCRP）主题模型powerfulnonparametric贝叶斯方法提取一个主题层次结构从一个给定的textcorpus，在层次结构的数据自动确定。分层Latent Dirichlet Allocation（HLDA）是一种流行的实例ofncrp主题模型。然而，人类只在小规模的评估，因为现有的崩溃吉布斯抽样和weightvariational推理算法实例化要么是没有可扩展性或牺牲与平均场假设inferencequality。此外，该数据结构的一种有效distributedimplementation，如动态增长countmatrices和树木，是具有挑战性的。

在本文中，我们提出了一个新的部分倒塌吉布斯抽样（PCGS）算法，它结合了倒塌和实例化weightalgorithms实现良好的可扩展性以及高质量模型。aninitialization策略提出了进一步提高模型的质量。最后，我们提出了一个高效的分布式实现完全的throughvectorization，预处理，和仔细的并发数据结构和通信策略设计。

实证研究表明，我们的算法是111倍以上的效率比以前的开源实现HLDA，相当或更好的模型质量。我们的分布式的实现可以提取1722 topicsfrom 280亿个代币1亿3100万文档语料库，即4-5个数量级大于前最大的语料库，50机在7hours。

View More Information

Paper 11 : A Unified Parallel Algorithm for Regularized Group PLS Scalable to Big Data

Authors: de Micheaux Pierre Lafaye, Liquet Benoit, Sutton Matthew

Subject: Machine Learning

Submitted Date: 20170223

Abstract: Partial Least Squares (PLS) methods have been heavily exploited to analysethe association between two blocs of data. These powerful approaches can beapplied to data sets where the number of variables is greater than the numberof observations and in presence of high collinearity between variables.Different sparse versions of PLS have been developed to integrate multiple datasets while simultaneously selecting the contributing variables. Sparsemodelling is a key factor in obtaining better estimators and identifyingassociations between multiple data sets. The cornerstone of the sparsityversion of PLS methods is the link between the SVD of a matrix (constructedfrom deflated versions of the original matrices of data) and least squaresminimisation in linear regression. We present here an accurate deion ofthe most popular PLS methods, alongside their mathematical proofs. A unifiedalgorithm is proposed to perform all four types of PLS including theirregularised versions. Various approaches to decrease the computation time areoffered, and we show how the whole procedure can be scalable to big data sets.

Abstract_CN: 偏最小二乘（PLS）方法已经被过度开采，分析两集团之间的数据关联。这些强大的方法可以应用于数据集，其中的变量的数量是大于数量的观测变量之间的共线性和高存在，请不同的稀疏版本已经开发了整合多个数据集的变量，同时选择。sparsemodelling是获得更好的估计和identifyingassociations多个数据集之间的关键因素。PLS方法的sparsityversion的基石是一个矩阵的奇异值分解之间的联系（从瘪的原始数据矩阵版本）和至少squaresminimisation线性回归。在这里，我们提出了一个准确的描述最流行的PLS方法，以及他们的数学证明。一个unifiedalgorithm提出执行所有四种请包括theirregularised版本。各种方法来减少计算时间的模式，和我们说明整个过程可以扩展到大数据集。

View More Information

Paper 12 : A Converse to Banach’s Fixed Point Theorem and its CLS Completeness

Authors: Daskalakis Constantinos, Tzamos Christos, Zampetakis Manolis

Subject: Computational Complexity, Learning, General Topology, Machine Learning

Submitted Date: 20170223

Abstract: Banach’s fixed point theorem for contraction maps has been widely used toanalyze the convergence of iterative methods in non-convex problems. It is acommon experience, however, that iterative maps fail to be globally contractingunder the natural metric in their domain, making the applicability of Banach’stheorem limited. We explore how generally we can apply Banach’s fixed pointtheorem to establish the convergence of iterative methods when pairing it withcarefully designed metrics.

Our first result is a strong converse of Banach’s theorem, showing that it isa universal analysis tool for establishing uniqueness of fixed points and forbounding the convergence rate of iterative maps to a unique fixed point. Inother words, we show that, whenever an iterative map globally converges to aunique fixed point, there exists a metric under which the iterative map iscontracting and which can be used to bound the number of iterations untilconvergence. We illustrate our approach in the widely used power method,providing a new way of bounding its convergence rate through contractionarguments.

We next consider the computational complexity of Banach’s fixed pointtheorem. Making the proof of our converse theorem constructive, we show thatcomputing a fixed point whose existence is guaranteed by Banach’s fixed pointtheorem is CLS-complete. We thus provide the first natural complete problem forthe class CLS, which was defined in [Daskalakis-Papadimitriou 2011] to capturethe complexity of problems such as P-matrix LCP, computing KKT-points, andfinding mixed Nash equilibria in congestion and network coordination games.

Abstract_CN: 收缩映射的巴拿赫不动点定理已被广泛用于分析非凸问题的迭代法的收敛性。不过是一种经验，迭代图失败是全局contractingunder自然度量在他们的领域，使banach’stheorem有限的适用性。我们将探讨如何可以把巴拿赫的不动点定理建立迭代法的收敛性配对时它审慎设计指标。

我们的第一个结果是一个强大的巴拿赫定理，表明它是一种普遍存在的分析工具，建立和forbounding不动点迭代映射收敛到一个独特的固定点。换句话说，我们表明，每当一个迭代映射全局收敛到独特的固定点，存在一个度量下的迭代映射iscontracting可绑定的迭代untilconvergence数。我们说明我们的方法，在广泛使用的功率的方法，提供了一种新的方式包围其收敛速度通过contractionarguments。

然后我们讨论巴拿赫的不动点定理计算的复杂性。使我们的逆定理构造的证明，我们获得一个固定点的存在是由巴拿赫的不动点定理保证CLS完成。因此我们提供了第一个完整的自然问题类华彩，这是[ 2011 ]拿他定义等问题的计算复杂性矩阵液晶行为，KKT点，并在拥塞和网络协调博弈的混合纳什均衡。

View More Information

Paper 13 : Learning to Draw Dynamic Agent Goals with Generative Adversarial Networks

Authors: Iqbal Shariq, Pearson John

Subject: Neurons and Cognition, Learning, Machine Learning

Submitted Date: 20170223

Abstract: We address the problem of designing artificial agents capable of reproducinghuman behavior in a competitive game involving dynamic control. Given dataconsisting of multiple realizations of inputs generated by pairs of interactingplayers, we model each agent’s actions as governed by a time-varying latentgoal state coupled to a control model. These goals, in turn, are described asstochastic processes evolving according to player-specific value functionsdepending on the current state of the game. We model these value functionsusing generative adversarial networks (GANs) and show that our GAN-basedapproach succeeds in producing sample gameplay that captures the rich dynamicsof human agents. The latent goal dynamics inferred and generated by our modelhas applications to fields like neuroscience and animal behavior, where theunderlying value functions themselves are of theoretical interest.

Abstract_CN: 我们的地址设计能够在竞争激烈的比赛涉及动态控制reproducinghuman行为人工代理问题。鉴于dataconsisting对对interactingplayers生成输入多个实现，我们的模型，每个Agent的行为由一个变latentgoal状态耦合到控制模型。这些目标，反过来，被描述作为过程演变根据球员的特定价值functionsdepending对游戏的当前状态。我们模型中的这些价值功能生成对抗网络（Gans），表明我们的GaN型成功生产样品的游戏，抓住了丰富的人类药物动力学。潜在的目标动力学模型推断的应用产生的神经科学和动物行为学等领域，在其价值函数本身是理论的兴趣。

View More Information

如需转载，请后台联系图灵之心获取授权，未获授权，禁止转载。

如需转载，请后台联系图灵之心获取授权，未获授权，禁止转载。

图灵之心(TuringKey)目前已实现“机器学习”，”自然语言处理”，”人工智能”，”机器视觉”，”机器人”五个方向的论文收集整理工作，每日更新T-2日的论文信息，包括标题，作者，研究方向，英文摘要，中文摘要（机器翻译）。及时获取论文信息，请关注图灵之心微信公众号：TuringKey。

图灵之心(TuringKey)目前已实现“机器学习”，”自然语言处理”，”人工智能”，”机器视觉”，”机器人”五个方向的论文收集整理工作，每日更新T-2日的论文信息，包括标题，作者，研究方向，英文摘要，中文摘要（机器翻译）。及时获取论文信息，请关注图灵之心微信公众号：TuringKey。