Neural and Evolutionary ComputingFast Genetic AlgorithmsBenjamin Doerr,Huu Phuoc Le,Régis Makhmara,Ta Duy NguyenSubjects:Neural and Evolutionary Computing (cs.NE)For genetic algorithms using a bit-string representation of length~(n), thegeneral recommendation is to take (1/n) as mutation rate. In this work, wediscuss whether this is really justified for multimodal functions. Taking jumpfunctions and the ((1+1)) evolutionary algorithm as the simplest example, weobserve that larger mutation rates give significantly better runtimes. For the(jump_{m,n}) function, any mutation rate between (2/ ...继续阅读 (30)

1. 我爱机器学习(1)arXiv Paper Daily: Fri, 10 Mar 20172. 新智元(5)WSDM 2017精选论文解读GAN 作者 Goodfellow：我为什么离开 OpenAI 重回谷歌大脑CMU：深度学习自然语言处理，神经机器翻译与 seq2seq 模型汇总，6 大类型附部署技巧【Science 封面】世界首个人工再造真核生命体问世，三分之二中国造2017 中国人工智能产业图谱：盘点最受 VC 青睐的 AI 公司3. 机器之心(5)四大深度学习框架+四类GPU+七种神经网络：交叉性能评测学界 | 5.5%语音识别词错率究竟如何炼成？IBM发布相关研究论文教程 | 谷歌官博详解XLA：可在保留TensorFlow灵活性的同时提升效率盘点 | 机器学习入门算法：从线性模型到神经网络学界 | CMU论文：神经机器翻译和Seq2seq模型导论4. 大数据文摘(3)还在纠结该从哪门编程语言学起？先来聊聊JavaScript3个月手把手云环境实训 系统学习数据科学 | 早鸟优惠最后1天维基解密再爆猛料：CIA利用漏洞入侵全球数十亿个人电子设备5. 深度学习世界(2)四大深度学习框架+四类GPU+七种神经网络：交叉性能评测世界首个人工再造真核生命体问世，三分之二中国造 ...继续阅读 (20)

Neural and Evolutionary ComputingDeep Reservoir Computing Using Cellular AutomataStefano Nichele,Andreas MolundSubjects:Neural and Evolutionary Computing (cs.NE); Emerging Technologies (cs.ET)Recurrent Neural Networks (RNNs) have been a prominent concept withinartificial intelligence. They are inspired by Biological Neural Networks (BNNs)and provide an intuitive and abstract representation of how BNNs work. Derivedfrom the more generic Artificial Neural Networks (ANNs), the recurrent ones aremeant to be used for temporal tasks, such as speech recognition, because theyare capable of memorizing ...继续阅读 (21)

1. 我爱机器学习(1)arXiv Paper Daily: Thu, 9 Mar 20172. 新智元(5)薛定谔的滚与深度学习中的物理【李飞飞演讲全文】谷歌的开源与生态：谷歌云正将AI民主化史上最全：TensorFlow 好玩的技术、应用和你不知道的黑科技起底中国科大 AI 与大数据兵团，57 员大将，不断更新滴滴成立美国研究院，烧钱补贴大战之后力图争夺自动驾驶人才3. 机器之心(4)业界 | 滴滴美国研究院落户硅谷，揽入全球高端人才现场直击 | 李飞飞首度亮相谷歌云大会：发布全新API，解读AI民主化业界 | 谷歌云官方正式宣布收购数据科学社区Kaggle机器之心专访杜克大学Lawrence Carin教授：十年后，你的工作会被人工智能取代吗？4. 大数据文摘(3)MIT的新成果震惊了科学界，我们真的可以用意念控制机器人了智能哲学：“第三问题”与图灵的“模仿游戏”谷歌宣布收购全球最大数据科学社区Kaggle5. CSDN大数据(3)WSDM 2017精选论文解读Visual Studio 2017全面上线！CSDN开启Visual Studio移动跨平台开发转型大作战Shanghai Apache Spark Meetup第十二次聚会6. 深度学习世界(2)IBM宣称人类语音识别词错率实际应为5.1%，自家系统已突破至5.5%业界 | 谷歌使用深度学习帮助病理学家检测癌症，算 ...继续阅读 (24)

Neural and Evolutionary ComputingEnd-to-End Prediction of Buffer Overruns from Raw Source Code via Neural Memory NetworksMin-je Choi,Sehun Jeong,Hakjoo Oh,Jaegul ChooComments: 6 pages + 1 appendix, 5 figuresSubjects:Software Engineering (cs.SE); Neural and Evolutionary Computing (cs.NE)Detecting buffer overruns from a source code is one of the most common andyet challenging tasks in program analysis. Current approaches have mainlyrelied on rigid rules and handcrafted features devised by a few experts,limiting themselves in terms of flexible applicability and robustness due todiverse bug patte ...继续阅读 (27)

Neural and Evolutionary ComputingLearning across scales – A multiscale method for Convolution Neural NetworksEldad Haber,Lars Ruthotto,Elliot HolthamSubjects:Neural and Evolutionary Computing (cs.NE); Computer Vision and Pattern Recognition (cs.CV)In this work we explore the connection between Convolution Neural Networks,partial differential equations, multigrid/multiscale methods and and optimalcontrol. We show that convolution neural networks can be represented as adiscretization of nonlinear partial differential equations, and that thelearning process can be interpreted as a control ...继续阅读 (20)

Neural and Evolutionary ComputingOn the Behavior of Convolutional Nets for Feature ExtractionDario Garcia-Gasulla,Ferran Parés,Armand Vilalta,Jonatan Moreno,Eduard Ayguadé,Jesús Labarta,Ulises Cortés,Toyotaro SuzumuraComments: Submitted to Journal of Artificial Intelligence Research Special Track on Deep Learning, Knowledge Representation, and ReasoningSubjects:Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Learning (cs.LG); Machine Learning (stat.ML)Convolutional neural networks (CNN) are representation learning techniquesthat achieve state-of- ...继续阅读 (25)

Neural and Evolutionary ComputingEvolving Deep Neural NetworksRisto Miikkulainen,Jason Liang,Elliot Meyerson,Aditya Rawal,Dan Fink,Olivier Francon,Bala Raju,Arshak Navruzyan,Nigel Duffy,Babak HodjatSubjects:Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI)The success of deep learning depends on finding an architecture to fit thetask. As deep learning has scaled up to more challenging tasks, thearchitectures have become difficult to design by hand. This paper proposes anautomated method, CoDeepNEAT, for optimizing deep learning architecturesthrough evolution. By extendi ...继续阅读 (33)

Neural and Evolutionary ComputingLearning A Physical Long-term PredictorSebastien Ehrhardt,Aron Monszpart,Niloy J. Mitra,Andrea VedaldiSubjects:Artificial Intelligence (cs.AI); Neural and Evolutionary Computing (cs.NE)Evolution has resulted in highly developed abilities in many naturalintelligences to quickly and accurately predict mechanical phenomena. Humanshave successfully developed laws of physics to abstract and model suchmechanical phenomena. In the context of artificial intelligence, a recent lineof work has focused on estimating physical parameters based on sensory data anduse them in ...继续阅读 (26)

Computer Vision and Pattern RecognitionPredicting Slice-to-Volume Transformation in Presence of Arbitrary Subject MotionBenjamin Hou,Amir Alansary,Steven McDonagh,Daniel Rueckert,Ben Glocker,Bernhard KainzComments: 8 pages, 4 figures, 6 pages supplemental material, currently under review for MICCAI 2017Subjects:Computer Vision and Pattern Recognition (cs.CV)This paper aims to solve a fundamental problem in intensity-based 2D/3Dregistration, which concerns the limited capture range and need for very goodinitialization of state-of-the-art image registration methods. We propose aregression appro ...继续阅读 (31)

Neural and Evolutionary ComputingLow-Precision Batch-Normalized ActivationsBenjamin GrahamComments: 16 pagesSubjects:Neural and Evolutionary Computing (cs.NE); Computer Vision and Pattern Recognition (cs.CV)Artificial neural networks can be trained with relatively low-precisionfloating-point and fixed-point arithmetic, using between one and 16 bits.Previous works have focused on relatively wide-but-shallow, feed-forwardnetworks. We introduce a quantization scheme that is compatible with trainingvery deep neural networks. Quantizing the network activations in the middle ofeach batch-normalizati ...继续阅读 (29)

Neural and Evolutionary ComputingControl of the Correlation of Spontaneous Neuron Activity in Biological and Noise-activated CMOS Artificial Neural MicrocircuitsRamin M. Hasani,Giorgio Ferrari,Hideaki Yamamoto,Sho Kono,Koji Ishihara,Soya Fujimori,Takashi Tanii,Enrico PratiSubjects:Neural and Evolutionary Computing (cs.NE); Neurons and Cognition (q-bio.NC); Quantitative Methods (q-bio.QM)There are several indications that brain is organized not on a basis ofindividual unreliable neurons, but on a micro-circuital scale providing Legoblocks employed to create complex architectures. At such an in ...继续阅读 (24)

Neural and Evolutionary ComputingBidirectional Backpropagation: Towards Biologically Plausible Error Signal Transmission in Neural NetworksHongyin Luo,Jie Fu,James GlassSubjects:Neural and Evolutionary Computing (cs.NE); Learning (cs.LG)The back-propagation (BP) algorithm has been considered the de facto methodfor training deep neural networks. It back-propagates errors from the outputlayer to the hidden layers in an exact manner using feedforward weights. Inthis work, we propose a more biologically plausible paradigm of neuralarchitecture according to biological findings. Specifically, we pr ...继续阅读 (25)

Neural and Evolutionary ComputingRobustness to Adversarial Examples through an Ensemble of SpecialistsMahdieh Abbasi,Christian GagnéComments: Submitted to ICLR 2017 Workshop TrackSubjects:Neural and Evolutionary Computing (cs.NE); Learning (cs.LG)We are proposing to use an ensemble of diverse specialists, where specialityis defined according to the confusion matrix. Indeed, we observed that foradversarial instances originating from a given class, labeling tend to be doneinto a small subset of (incorrect) classes. Therefore, we argue that anensemble of specialists should be better able to ...继续阅读 (31)

Neural and Evolutionary ComputingOnline Representation Learning with Multi-layer Hebbian Networks for Image Classification TasksYanis Bahroun,Andrea SoltoggioComments: 8 pages, 6 figuresSubjects:Neural and Evolutionary Computing (cs.NE); Computer Vision and Pattern Recognition (cs.CV)Unsupervised learning allows algorithms to adapt to different data thanks tothe autonomous discovery of discriminating features during the training. Whenthese algorithms are reducible to cost-function minimisation, betterinterpretations of their learning dynamics are possible. Recently, newHebbian-like plasticity ...继续阅读 (22)

Neural and Evolutionary ComputingOnline Multi-Task Learning Using Biased SamplingSahil Sharma,Balaraman RavindranComments: 9 pagesSubjects:Neural and Evolutionary Computing (cs.NE); Learning (cs.LG)One of the long-standing challenges in Artificial Intelligence is to build asingle agent which can solve multiple tasks. Recent progress in multi-tasklearning for learning behavior in many goal-directed sequential tasks has beenin the form of distillation based learning wherein a single student networklearns from multiple task-specific teacher networks by mimicking thetask-specific policies of the t ...继续阅读 (30)

Neural and Evolutionary ComputingToward Abstraction from Multi-modal Data: Empirical Studies on Multiple Time-scale Recurrent ModelsJunpei Zhong,Angelo Cangelosi,Tetsuya OgataComments: Accepted by IJCNN 2017Subjects:Neural and Evolutionary Computing (cs.NE)The abstraction tasks are challenging for multi- modal sequences as theyrequire a deeper semantic understanding and a novel text generation for thedata. Although the recurrent neural networks (RNN) can be used to model thecontext of the time-sequences, in most cases the long-term dependencies ofmulti-modal data make the back-propagation thr ...继续阅读 (27)

1. 新智元(5)深度学习下一个大突破：机器阅读北大 AI 公开课：徐小平、雷鸣——人工智能的发展、挑战与机遇【Nature】世界最薄芯片拯救摩尔定律：仅一层原子厚的集成电路，液态金属纳米打印【Bengio vs 谷歌】深度学习兄弟对决，神经网络泛化本质之争挑战这 25 道面试题，成为硅谷机器学习工程师2. 机器之心(5)学界 | iPOP：首个基于个性化大数据的个性化医学研究模型学习全面概述：利用机器学习查找软件漏洞资源 | 5 个Python 库，照亮你的机器学习之路一周论文 | 增强学习在Image Caption任务上的应用深度 | 机器学习很有趣Part6：怎样使用深度学习进行语音识别 ...继续阅读 (29)

Neural and Evolutionary ComputingUnbiased Online Recurrent OptimizationCorentin Tallec,Yann OllivierComments: 11 pages, 5 figuresSubjects:Neural and Evolutionary Computing (cs.NE); Learning (cs.LG)The novel Unbiased Online Recurrent Optimization (UORO) algorithm allows foronline learning of general recurrent computational graphs such as recurrentnetwork models. It works in a streaming fashion and avoids backtracking throughpast activations and inputs. UORO is a modification of NoBackTrack thatbypasses the need for model sparsity and makes implementation easy in currentdeep learning frameworks, ...继续阅读 (26)

Neural and Evolutionary ComputingBuilding Robust Stochastic Configuration Networks with Kernel Density EstimationDianhui Wang,Ming LiComments: 16 pagesSubjects:Neural and Evolutionary Computing (cs.NE); Learning (cs.LG); Machine Learning (stat.ML)This paper aims at developing robust data modelling techniques usingstochastic configuration networks (SCNs), where a weighted least squares methodwith the well-known kernel density estimation (KDE) is used in the design ofSCNs. The alternating optimization (AO) technique is applied for iterativelybuilding a robust SCN model that can reduce some nega ...继续阅读 (28)

Neural and Evolutionary ComputingExploring loss function topology with cyclical learning ratesLeslie N. Smith,Nicholay TopinComments: Submitted as an ICLR 2017 Workshop paperSubjects:Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)We present observations and discussion of previously unreported phenomenadiscovered while training residual networks. The goal of this work is to betterunderstand the nature of neural networks through the examination of these newempirical results. These behaviors were identified through the application ofCyclical Learning Rates (CLR) and linear network int ...继续阅读 (28)

Neural and Evolutionary ComputingFeature Space Modeling Through Surrogate IlluminationAdam Gaier,Alexander Asteroth,Jean-Baptiste MouretSubjects:Neural and Evolutionary Computing (cs.NE); Computational Engineering, Finance, and Science (cs.CE); Machine Learning (stat.ML)The MAP-Elites algorithm produces a set of high-performing solutions thatvary according to features defined by the user. This technique has thepotential to be a powerful tool for design space exploration, but is limited bythe need for numerous evaluations. The Surrogate-Assisted Illuminationalgorithm (SAIL), introduced here, in ...继续阅读 (22)

Neural and Evolutionary ComputingStochastic Configuration Networks: Fundamentals and AlgorithmsDianhui Wang,Ming LiComments: 19 pages, 4 figures and 28 referencesSubjects:Neural and Evolutionary Computing (cs.NE)This paper contributes to a development of randomized methods for neuralnetworks. The proposed learner model is generated incrementally by stochasticconfiguration (SC) algorithms, termed as Stochastic Configuration Networks(SCNs). In contrast to the existing randomised learning algorithms for singlelayer feed-forward neural networks (SLFNNs), we randomly assign the inputweights and bia ...继续阅读 (27)

Neural and Evolutionary ComputingEnergy Saving Additive Neural NetworkArman Afrasiyabi,Ozan Yildiz,Baris Nasir,Fatos T. Yarman Vural,A. Enis CetinComments: 8 pages (double column), 2 figures, 1 tableSubjects:Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Learning (cs.LG)In recent years, machine learning techniques based on neural networks formobile computing become increasingly popular. Classical multi-layer neuralnetworks require matrix multiplications at each stage. Multiplication operationis not an energy efficient operation and consequently it drains the batter ...继续阅读 (26)

Neural and Evolutionary ComputingA Historical Review of Forty Years of Research on CMACFrank Z. XingSubjects:Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI)The Cerebellar Model Articulation Controller (CMAC) is an influentialbrain-inspired computing model in many relevant fields. Since its inception inthe 1970s, the model has been intensively studied and many variants of theprototype, such as Kernel-CMAC, Self-Organizing Map CMAC, and Linguistic CMAC,have been proposed. This review article focus on how the CMAC model isgradually developed and refined to meet the dema ...继续阅读 (25)

Neural and Evolutionary ComputingEstimation of classrooms occupancy using a multi-layer perceptronEugénio Rodrigues,Luísa Dias Pereira,Adélio Rodrigues Gaspar,Álvaro Gomes,Manuel Carlos Gameiro da SilvaComments: 6 pages, 2 figures, conference articleSubjects:Neural and Evolutionary Computing (cs.NE); Learning (cs.LG)This paper presents a multi-layer perceptron model for the estimation ofclassrooms number of occupants from sensed indoor environmental data-relativehumidity, air temperature, and carbon dioxide concentration. The modellingdatasets were collected from two classr ...继续阅读 (32)

1. 我爱机器学习(1)arXiv Paper Daily: Tue, 7 Feb 20172. 新智元(4)谷歌开源最大手动注释视频数据集和 TensorFlow 模型性能调优工具AAAI 2017讲座：8大主流深度学习框架超详细对比（90PPT）微软开源生物模型分析器（BMA）：基于云的生物研究工具Swarm AI 精确预测超级碗最终比分，大逆转剧情难逃群集智能预测3. 机器之心(5)开源 | 微软开源生物模型分析器：一款基于云的生物研究工具资源 | 谷歌开源BoundingBoxes，迄今为止最大的带注释图像数据集现场报道 | AAAI-17 Workshop：自动驾驶与人工智能深度 | 谷歌Lukasz Kaiser为你解读深度学习如何变革自然语言处理Face++旷视科技首席科学家孙剑首次接受专访：计算机视觉亟待解决哪些问题？4. 大数据文摘(3)小白学数据之常用Python库“小抄表”（附小抄表PDF下载）视频更新｜斯坦福CS231n深度学习与计算机视觉课时23-卷积神经网络工程实践技巧与注意点（下）超级碗另一面：大逆转背后，你没看到的人工智能大PK5. CSDN大数据(3)【Cloudaily】Oracle进击云计算，“超级碗”交付一场高科技盛宴回顾与展望：从软硬角度看 VR 开发技术难点基于Apple IDFA的开源项目——SimulateIDFA6. 36大数据(3 ...继续阅读 (27)

Neural and Evolutionary ComputingDistributed Evolutionary k-way Node SeparatorsPeter Sanders,Christian Schulz,Darren Strash,Robert WilligerComments: arXiv admin note: text overlap witharXiv:1509.01190,arXiv:1110.0477Subjects:Neural and Evolutionary Computing (cs.NE); Distributed, Parallel, and Cluster Computing (cs.DC); Data Structures and Algorithms (cs.DS)Computing high quality node separators in large graphs is necessary for avariety of applications, ranging from divide-and-conquer algorithms to VLSIdesign. In this work, we present a novel distributed evolutionary algorithmtackling the k-wa ...继续阅读 (22)

Neural and Evolutionary ComputingDistributed Evolutionary k-way Node SeparatorsPeter Sanders,Christian Schulz,Darren Strash,Robert WilligerComments: arXiv admin note: text overlap witharXiv:1509.01190,arXiv:1110.0477Subjects:Neural and Evolutionary Computing (cs.NE); Distributed, Parallel, and Cluster Computing (cs.DC); Data Structures and Algorithms (cs.DS)Computing high quality node separators in large graphs is necessary for avariety of applications, ranging from divide-and-conquer algorithms to VLSIdesign. In this work, we present a novel distributed evolutionary algorithmtackling the k-wa ...继续阅读 (33)

Neural and Evolutionary ComputingRobust Particle Swarm Optimizer based on ChemomimicryCasey Kneale,Karl S. BookshComments: To be revised for formatting and submitted as a Letters style paperSubjects:Neural and Evolutionary Computing (cs.NE)A particle swarm optimizer (PSO) loosely based on the phenomena ofcrystallization and a chaos factor which follows the complimentary errorfunction is described. The method features three phases: diffusion, directedmotion, and nucleation. During the diffusion phase random walk is the onlycontributor to particle motion. As the algorithm progresses the contribu ...继续阅读 (30)