packtpub.com 是一家电子书网站，目前在上面它们每天都会赠送一本电子书，今天送的是这本 Learning Python Application Development。注册账户后就可以免费获取了，网站提供了 ePub、Mobi 和 PDF 三种格式，同时也提供了书中的源码。获取地址：https://www.packtpub.com/packt/offers/free-learning截止时间：2017 年 5 月 21 日 7 时 来自反斗限免您可能也喜欢：免费获取电子书 Learning Robotics Using Python[$35.99→0]免费获取电子书 Tkinter GUI Application Development Blueprints[$35.99→0]免费获取电子书 Learning AWS[$39.99→0]免费获取电子书 MEAN Web Development[$29.99→0]免费获取电子书 Python Machine Learning[$35.99→0]无觅 ...继续阅读 (14)

packtpub.com 是一家电子书网站，目前在上面它们每天都会赠送一本电子书，今天送的是这本Learning Robotics Using Python。注册账户后就可以免费获取了，网站提供了 ePub、Mobi 和 PDF 三种格式，同时也提供了书中的源码。获取地址：https://www.packtpub.com/packt/offers/free-learning截止时间：2017 年 5 月 20 日 7 时 来自反斗限免您可能也喜欢：免费获取电子书 Learning AWS[$39.99→0]免费获取电子书 Python Machine Learning[$35.99→0]免费获取电子书 Mastering BeagleBone Robotics免费获取电子书 Learning Object-Oriented Programming[$20→0]免费获取电子书 Learning Penetration Testing with Python[$39.99→0]无觅 ...继续阅读 (10)

Neural and Evolutionary ComputingLimited-Memory Matrix Adaptation for Large Scale Black-box OptimizationIlya Loshchilov,Tobias Glasmachers,Hans-Georg BeyerSubjects:Neural and Evolutionary Computing (cs.NE); Learning (cs.LG); Optimization and Control (math.OC)The Covariance Matrix Adaptation Evolution Strategy (CMA-ES) is a popularmethod to deal with nonconvex and/or stochastic optimization problems when thegradient information is not available. Being based on the CMA-ES, the recentlyproposed Matrix Adaptation Evolution Strategy (MA-ES) provides a rathersurprising result that the covariance mat ...继续阅读 (33)

Neural and Evolutionary ComputingTransfer Learning for Named-Entity Recognition with Neural NetworksJi Young Lee,Franck Dernoncourt,Peter SzolovitsComments: The first two authors contributed equally to this workSubjects:Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Neural and Evolutionary Computing (cs.NE); Machine Learning (stat.ML)Recent approaches based on artificial neural networks (ANNs) have shownpromising results for named-entity recognition (NER). In order to achieve highperformances, ANNs need to be trained on a large labeled dataset. However,labels might be diffi ...继续阅读 (24)

Neural and Evolutionary ComputingEnsemble of heterogeneous flexible neural trees using multiobjective genetic programmingVarun Kumar Ojha,Ajith Abraham,Václav SnášelJournal-ref: Applied Soft Computing, 2017, Volume 52 Pages 909 to 924Subjects:Neural and Evolutionary Computing (cs.NE)Machine learning algorithms are inherently multiobjective in nature, whereapproximation error minimization and model’s complexity simplification are twoconflicting objectives. We proposed a multiobjective genetic programming (MOGP)for creating a heterogeneous flexible neural tree (HFNT), tree ...继续阅读 (21)

作者：我偏笑_NSNirvana来源：简书浅谈机器学习基础（上）浅谈机器学习基础（下）浅谈深度学习基础（上）浅谈深度学习基础（下）浅谈自然语言处理基础（上）浅谈自然语言处理基础（中）浅谈自然语言处理基础（下）浅谈语音识别基础浅谈推荐系统基础浅谈搜索引擎基础（上）浅谈搜索引擎基础（下）浅谈数据挖掘基础浅谈知识图谱基础 ...继续阅读 (22)

Neural and Evolutionary ComputingDiscovery Radiomics via Evolutionary Deep Radiomic Sequencer Discovery for Pathologically-Proven Lung Cancer DetectionMohammad Javad Shafiee,Audrey G. Chung,Farzad Khalvati,Masoom A. Haider,Alexander WongComments: 8 pagesSubjects:Neural and Evolutionary Computing (cs.NE); Computer Vision and Pattern Recognition (cs.CV); Learning (cs.LG)While lung cancer is the second most diagnosed form of cancer in men andwomen, a sufficiently early diagnosis can be pivotal in patient survival rates.Imaging-based, or radiomics-driven, detection methods have been developed toa ...继续阅读 (59)

...继续阅读 (16)

雷锋网按：本文作者为 Jeffries Consulting 创始人 Daniel Jeffries，他以自己的阅读体验，对当前含金量极高的几本深度学习书籍进行点评；对每本书的内容重点、所适合的读者群进行了总结。非常适合学习者在购书前进行参考，以免白费时间。雷锋网编译。Daniel Jeffries：多年来，由于实验室研究和现实应用效果之间的鸿沟，少有人持续研究人工智能，AI在很多领域停滞不前。但近两年，AI 在一些领域陆续有了重大突破，比如：图像识别；自动驾驶；Alpha Go等。许多八九十年代的算法，由于硬件慢和缺乏数据等原因束之高阁。而现在，有众多大数据集和大规模并行芯片的支持，这些算法终于初见成效。在过去的一年多时间里，研究人员奋笔疾书，竞相出版专著，以满足读者对深度学习知识的渴求。第一本关于深度学习的书已经上架，更多的将会在夏天或者明年年初陆续上架。我有幸提前拜读了若干专著的初稿，这些书的最终出版让人期待万分。接下来给大家推荐一些深度学习书籍，这些书可以引导大家如何学习人工智能，对深度学习的快速理解有很大帮助。理论和实践相互兼顾我父亲常说“凡事均衡最好。”对此我深信不疑，且奉为圭臬。当然，我承认在周末或者维加斯的时候，偶尔也会将其抛诸脑后。我赞成理论与实践要相互均衡。比如针对某个问题，给一个明确的背景信息，然后根据实例让我自己动手实践。书本不要过多阐述理论知识，应该让读者 ...继续阅读 (7)

Neural and Evolutionary ComputingMaking up for the deficit in a marathon runIztok Fister Jr.,Dušan Fister,Suash Deb,Uroš Mlakar,Janez Brest,Iztok FisterComments: ISMSI 2017Subjects:Neural and Evolutionary Computing (cs.NE)To predict the final result of an athlete in a marathon run thoroughly is theeternal desire of each trainer. Usually, the achieved result is weaker than thepredicted one due to the objective (e.g., environmental conditions) as well assubjective factors (e.g., athlete’s malaise). Therefore, making up for thedeficit between predicted and achieved results is th ...继续阅读 (26)

作者：田渊栋DL 的难度取决于看问题的角度。数学系的话应该往理论方向看。DL 用线性和非线性函数堆砌来构造函数，用随机梯度下降就可以训练，而且还有非常好的泛化能力，在我看来这样的东西居然没有原理这是完全不可想像的。但是 DL 很难，难到大家都不知道如何入手，用什么数学工具都不知道，所以说什么样的理论文章都有，凝聚态物理的（spin-glass)，量子物理的（重整化群），ODE 的（动力系统），各种优化理论（这个例子太多），各种复杂度理论，但还没看到真正提炼出本质的。数学这边那么多理论，但现在还没有哪个能搬到 DL 上而且能严丝合缝的，至于能进行有效预测的就更没有了。这就足够说明问题的复杂性，DL 是没那么简单的。要深入 DL，各种基础知识是不可少的。但是基础知识并不是拿来炫耀的本钱，因为学了一大堆，可能一个都没用。比如说 VC Dimension 或者 Rademacher Complexity 对 DL 有没有用？它们本身都很漂亮，但可能假设太强得到的结论和现实差很远，或者假设太宽泛得到的结论太弱，都不在 DL 的工作区间。所以做理论的人，不得不一个个学过来，想过了，试过了，然后放掉，下一个。解决问题是最终的目标，而不是学得比人多些，然后坐在原地不动夸耀自己厉害。耍弄十八般武艺，比不过高手一招制敌。至于这一招怎么找到，或者从理论分析这边入手，或者从实验入手，都是一样有价值，都是平 ...继续阅读 (27)

packtpub.com 是一家电子书网站，目前在上面它们每天都会赠送一本电子书，今天送的是这本Machine Learning with R – Second Edition。注册账户后就可以免费获取了，网站提供了 ePub、Mobi 和 PDF 三种格式，同时也提供了书中的源码。获取地址：https://www.packtpub.com/packt/offers/free-learning截止时间：2017 年 5 月 10 日 7 时 来自反斗限免您可能也喜欢：免费获取电子书 Learning Linux Binary Analysis[$35.99→0]免费获取电子书 Python 3 Object-oriented Programming - Second Edition[$39.99→0]免费获取电子书 Learning Object-Oriented Programming[$20→0]免费获取电子书 Building Machine Learning Systems with Python[$29.99→0]免费获取电子书 Raspberry Pi for Secret Agents - Second Edition[$14.99→0]无觅 ...继续阅读 (20)

Neural and Evolutionary ComputingDeveloping All-Skyrmion Spiking Neural NetworkZhezhi He,Deliang FanSubjects:Neural and Evolutionary Computing (cs.NE)In this work, we have proposed a revolutionary neuromorphic computingmethodology to implement All-Skyrmion Spiking Neural Network (AS-SNN). Suchproposed methodology is based on our finding that skyrmion is a topologicalstable spin texture and its spatiotemporal motion along the magnetic nano-trackintuitively interprets the pulse signal transmission between two interconnectedneurons. In such design, spike train in SNN could be encoded as particle- ...继续阅读 (28)

packtpub.com 是一家电子书网站，目前在上面它们每天都会赠送一本电子书，今天送的是这本Learning Linux Binary Analysis。注册账户后就可以免费获取了，网站提供了 ePub、Mobi 和 PDF 三种格式，同时也提供了书中的源码。获取地址：https://www.packtpub.com/packt/offers/free-learning截止时间：2017 年 5 月 9 日 7 时 来自反斗限免您可能也喜欢：免费获取电子书 Mastering Embedded Linux Programming[$20→0]免费获取电子书 Learning Object-Oriented Programming[$20→0]免费获取电子书 Learning jQuery - Fourth Edition[$23.99→0]免费获取电子书 Learning R for Geospatial Analysis[$29.99→0]免费获取电子书 Learning PHP 7[$39.99→0]无觅 ...继续阅读 (14)

Neural and Evolutionary ComputingDiscrete Modeling of Multi-Transmitter Neural Networks with Neuron CompetitionNikolay Bazenkov,Dmitry Vorontsov,Varvara Dyakonova,Liudmila Zhilyakova,Oleg Kuznetsov,Dmitri SakharovComments: 13 pages, 4 figuresSubjects:Neural and Evolutionary Computing (cs.NE); Neurons and Cognition (q-bio.NC)We propose a novel discrete model of central pattern generators (CPG),neuronal ensembles generating rhythmic activity. The model emphasizes the roleof nonsynaptic interactions and the diversity of electrical properties innervous systems. Neurons in the model release differ ...继续阅读 (39)

How will we reinvent training? These TED Talks discover the newest thinking — from academics, parents, children — on how one can build a better school. Elsewhere in the course of the Middle Ages, Islamic science and arithmetic flourished underneath the Islamic caliphate which was established across the Middle East, extending from the Iberian Peninsula in the west to the Indus in the east and to the Almoravid Dynasty and Mali Empire within the south. Fifteen group-based tasks will obtain funding for the subsequent college yr for projects to assist enhance pupil attendance all through Yukon. In ...继续阅读 (14)

...继续阅读 (17)

Neural and Evolutionary ComputingA New Type of Neurons for Machine LearningFenglei Fan,Wenxiang Cong,Ge WangComments: 5 pages, 8 figures, 11 referencesSubjects:Neural and Evolutionary Computing (cs.NE); Learning (cs.LG)In machine learning, the use of an artificial neural network is themainstream approach. Such a network consists of layers of neurons. Theseneurons are of the same type characterized by the two features: (1) an innerproduct of an input vector and a matching weighting vector of trainableparameters and (2) a nonlinear excitation function. Here we investigate thepossibility of repla ...继续阅读 (35)

Neural and Evolutionary ComputingA Recurrent Neural Model with Attention for the Recognition of Chinese Implicit Discourse RelationsSamuel Rönnqvist,Niko Schenk,Christian ChiarcosComments: To appear at ACL2017, code available atthis https URLSubjects:Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)We introduce an attention-based Bi-LSTM for Chinese implicit discourserelations and demonstrate that modeling argument pairs as a joint sequence canoutperform word order-agnostic approaches. Our model benefits from a ...继续阅读 (28)

雷锋网AI科技评论按：ICLR 2017 于4月24-26日在法国土伦举行，雷锋网AI科技评论的编辑们在法国带来一线报道。在这个深度学习会议举办之际，雷锋网也将围绕会议议程及论文介绍展开一系列的覆盖和专题报道，敬请期待。据雷锋网了解，这是一篇来自Twitter的研究员给ICLR 2017投递的论文。目前论文已经被录用而且还获得了ICLR委员会的高度评价。想知道研究员在论文中又提出了什么新的研究思路？或者ICLR委员会对这项工作的评价如何？那就赶紧跟随小编来看看吧。以下内容是AI科技评论根据论文内容进行的部分编译。论文摘要虽然当下深度神经网络算法在大规模数据集上取得了巨大的成功，但是这类算法在少量学习（Few-shot Learning）任务中的表现却往往不如人意，这是因为在少量学习任务中要求分类器能够在经过每个类别的少量样本训练之后快速具备泛化能力。现在大家通常都认为，要使得一个基于梯度下降优化算法的分类器能具备较好的表现就需要在大量训练样本上进行多次训练迭代。而在本次的论文工作中，来自Twitter的研究员推出了基于LSTM的元学习模型（LSTM-based meta-learner model），该模型可以学习到精确的优化算法来训练另外一个用于少量学习任务的神经网络分类器。图1：该图展示了元学习（Meta-learning）过程的一个示例。图片分为两大部分，上半部分的表示元训练 ...继续阅读 (19)

为期3天的ICLR 2017深度学习顶会正在法国土伦召开，6场Invited Talk，15场Contributed Talk（oral paper）, 310个论文海报展示让任何人置身其中的时候，都不免产生一种信息量巨大，怎么看也看不完的错觉，雷锋网［AI科技评论］团队实地考察后，发现我们可以从现场的亮点中来理出一些对大会深入学习的蛛丝马迹。大牛“满天飞”ICLR 2017正式召开前，大会官网议程表里显示的只是几个简单的Invited talk(主题演讲)和Contributed Talk（口头论文演讲）的演讲者信息，丝毫不见其它大牛讲者的踪影。不过这个平凡感没维持太久，第一天（4月24号）上午的第2个Contributed Talk结束后，3楼主会议室的700多名听众不约而同地涌出会议室进入到外面的自由交流空间进入Poster Session（论文自由交流）环节，正好让刚刚“湮没”在主会议室的Yann LeCun, Yoshua Bengio, Ian Goodfellow等大牛都跻身在3楼自由讨论空间和大家来个“偶然邂逅”。譬如，雷锋网［AI科技评论］在3楼某个不起眼的角落就看见了Yann LeCun正在和一个纽约大学的教授讨论刚刚结束的演讲主题，因为隔得比较近，我们顺势请教了LeCun教授关于ICLR的几个问题并要求了一个Yann LeCun教授的签名。论文自由交流时间占比 ...继续阅读 (21)

Neural and Evolutionary ComputingTapping the sensorimotor trajectoryOswald Berthold,Verena HafnerComments: Paper submitted to the ICDL-Epirob 2017 conference. Currently under reviewSubjects:Neural and Evolutionary Computing (cs.NE)We propose sensorimotor tappings, a new graphical technique that explicitlyrepresents relations between the time steps of an agent’s sensorimotor loop anda single training step of an adaptive model that the agent is using internally.In the simplest case this is a relation linking two time steps. In realisticcases these relations can extend over several time ste ...继续阅读 (31)

1. 我爱机器学习(1)Machine Learning Daily: 2017-04-262. 人工智能学家(4)腾讯区块链方案白皮书：底层技术平台及五大场景解决方案10位科技互联网领袖的趋势预测”独手人”大脑揭示神经学新组织理论浅说深度学习：核心概念 ...继续阅读 (29)

Neural and Evolutionary ComputingReinforcement Learning Based Dynamic Selection of Auxiliary Objectives with Preserving of the Best Found SolutionIrina Petrova,Arina BuzdalovaComments: this is a full version of a paper which has been accepted as a student workshop paper to GECCO conference 2017Subjects:Neural and Evolutionary Computing (cs.NE); Learning (cs.LG)Efficiency of single-objective optimization can be improved by introducingsome auxiliary objectives. Ideally, auxiliary objectives should be helpful.However, in practice, objectives may be efficient on some optimization stagesbut obstru ...继续阅读 (42)

packtpub.com 是一家电子书网站，目前在上面它们每天都会赠送一本电子书，今天送的是这本Learning PHP 7。注册账户后就可以免费获取了，网站提供了 ePub、Mobi 和 PDF 三种格式，同时也提供了书中的源码。获取地址：https://www.packtpub.com/packt/offers/free-learning截止时间：2017 年 4 月 25 日 8 时 来自反斗限免您可能也喜欢：免费获取电子书 R Deep Learning Essentials[$39.99→0]免费获取电子书 Learning Reactive Programming with Java 8[$35.99→0]免费获取电子书 Flask By Example免费获取电子书 Learning C++ by Creating Games with UE4[$26.99→0]免费获取电子书 Python Machine Learning[$35.99→0]无觅 ...继续阅读 (11)

一、机器学习概述信息图二、机器学习演变信息图三、机器学习方法信息图原文链接：1http://usblogs.pwc.com/emerging-technology/a-look-at-machine-learning-infographic/2http://usblogs.pwc.com/emerging-technology/machine-learning-evolution-infographic/3http://usblogs.pwc.com/emerging-technology/machine-learning-methods-infographic/您可能也喜欢的文章：游戏机手柄演变史--信息图一张图了解深度学习的前世今生——信息图Email演变史--信息图公关的演变与进化--信息图哪一种学习类型适合你——信息图无觅 ...继续阅读 (15)

Neural and Evolutionary ComputingFeed-forward approximations to dynamic recurrent network architecturesDylan Richard MuirComments: 8 pages, 10 figuresSubjects:Neural and Evolutionary Computing (cs.NE); Neurons and Cognition (q-bio.NC)Recurrent neural network architectures can have useful computationalproperties, with complex temporal dynamics. However, evaluation of recurrentdynamic architectures requires solution of systems of differential equations,and the number of evaluations required to determine their response to a giveninput can vary with the input, or can be indeterminate altogether in ...继续阅读 (27)

Neural and Evolutionary ComputingGenetic Algorithm Based Floor Planning SystemHamide Ozlem Dalgic,Erkan Bostanci,Mehmet Serdar GuzelSubjects:Neural and Evolutionary Computing (cs.NE)Genetic Algorithms are widely used in many different optimization problemsincluding layout design. The layout of the shelves play an important role inthe total sales metrics for superstores since this affects the customers’shopping behaviour. This paper employed a genetic algorithm based approach todesign shelf layout of superstores. The layout design problem was tackled byusing a novel chromosome representat ...继续阅读 (30)

本文是系列文章中的第一篇，是对深度增强学习/深度强化学习的基本介绍以及对实现通用人工智能的探讨。现在但凡写人工智能的文章，必提 Alpha Go。也正是因为Alpha Go在围棋人机大战中里程碑式的胜利，人工智能迎来了新的春天。 本文也不免俗套，从Alpha Go说起，但希望能指明一些被忽视的但对Alpha Go棋力有深远影响的技术。 围棋人工智能大致可以分为三个阶段 [1]：第一阶段以启发式算法为主，水平低于业余初段，代表软件即以静态势力函数为强项的 手谈； 第二阶段以 蒙特卡洛树搜索算法为代表，水平最高达到业余5段，比如说 Zen， Crazy Stone；第三阶段以深度学习（ Deep Learning）以及增强学习（ Reinforcement Learning，也称强化学习）算法为突破，并战胜了人类职业九段棋手李世乭，这也就是Alpha Go的故事了。每每提到Alpha Go卓越的能力，往往归咎于深度学习的强大，但实际上增强学习算法也功不可没。这二者的结合被称之为深度增强学习（ Deep Reinforcement Learning，DRL），最初见于DeepMind在Nature上发表的 Human-level control through deep reinforcement learning。 本文试图从深度增强学习的角度来探讨通用人工智能的实现，并简要介绍了深度 ...继续阅读 (10)

Neural and Evolutionary ComputingDiscovering Evolutionary Stepping Stones through Behavior DominationElliot Meyerson,Risto MiikkulainenComments: To Appear in Proceedings of the Genetic and Evolutionary Computation Conference (GECCO 2017)Subjects:Neural and Evolutionary Computing (cs.NE)Behavior domination is proposed as a tool for understanding and harnessingthe power of evolutionary systems to discover and exploit useful steppingstones. Novelty search has shown promise in overcoming deception by collectingdiverse stepping stones, and several algorithms have been proposed that combinenovelty w ...继续阅读 (31)