加载中…
博文
标签:

杂谈

天眼大项目、登月等太空项目、霍金摄星计划:  科技是基础; 大项目本身是很必须的; 大项目有很大的溢出效益-反哺民生的左右 ; 复杂系统大项目有练兵检验自身组织和协同等软实力的能力; 质疑代表监督的力量也很重要。 显微镜和疾病:兴趣带来的解决问题的方法思路工具; 帮助的直接和间接,授人以鱼不如授人以渔; 突破的来源:大挑战激发创新,燃起想象力行动力。 大项目相当于战争对科技的促进。 锻炼个人及民族国家人类的信心。


原文:http://tech.sina.com.cn/d/s/2016-07-04/doc-ifxtsatn8036983.shtml

阅读  ┆ 评论  ┆ 转载 ┆ 收藏 
标签:

杂谈

http://www.jianshu.com/p/db87c51de510


OpenAI 的核心志向是开发出算法和技术可以赋予计算机理解世界的能力.

 Richard Feynman 名言的含义:``What I cannot create, I do not understand.''

模型能够发现和有效地内化(internalize)数据的本质,从而才可以生成这些数据.

他们有着自动学习数据集的自然特征的潜力,包含类型、维度或者其他特征.


InfoGAN 对该空间引入了额外的结构通过增加一个新的包含最大化表示向量和观测值的小的子集的互信息目标函数. 这个方法给出相当出色的结果. 例如,在 3D 人脸图像中,我们变动代码的一个连续的维度,保持其他维度不变. 很明显从 5 个提供的例子中(每一行),代码的结果维度刻画了可解释的维度,并且模型在没有告诉这些重要特征存在的情形下可能已经理解到是存在摄像头角度、面部变化等等:



这项工作更深的启示是,在训练生成式模型的过程中,我们最终会赋予计算机对

阅读  ┆ 评论  ┆ 转载 ┆ 收藏 
标签:

杂谈

http://www.jianshu.com/p/db87c51de510

OpenAI 的核心志向是开发出算法和技术可以赋予计算机理解世界的能力.

 Richard Feynman 名言的含义:``What I cannot create, I do not understand.''

模型能够发现和有效地内化(internalize)数据的本质,从而才可以生成这些数据.

他们有着自动学习数据集的自然特征的潜力,包含类型、维度或者其他特征.


InfoGAN 对该空间引入了额外的结构通过增加一个新的包含最大化表示向量和观测值的小的子集的互信息目标函数. 这个方法给出相当出色的结果. 例如,在 3D 人脸图像中,我们变动代码的一个连续的维度,保持其他维度不变. 很明显从 5 个提供的例子中(每一行),代码的结果维度刻画了可解释的维度,并且模型在没有告诉这些重要特征存在的情形下可能已经理解到是存在摄像头角度、面部变化等等:



 

阅读  ┆ 评论  ┆ 转载 ┆ 收藏 
标签:

杂谈

For example, given a dataset of indoor videos, a representation that explicitly rep-resents whether or not the lights are on is more disentangled than a representationcomposed of raw pixels. This is because for the common transformation of flippingthe light switch, the first representation will only change in only that single dimension(light on or off), whereas the second will change in every single dimension (pixel).

 

In a certain light, all of science is one big unsupervised learning problem in whichwe search for the most disentangled representation of the world around us


A human should

阅读  ┆ 评论  ┆ 转载 ┆ 收藏 
标签:

杂谈

字典是大脑功能的全息映射,每个字词代表的概念对应到大脑的物理结构:口语对应了人的某种认识、某个习惯、某个动作、对外界的某些认识、定义、识别;书面语言让人对口语的概念可以反思;反思即引发更高级抽象的思考;字典即人类对世界认识的一种全息抽象,人类思考世界的概念集合,字典包含从具体(树)到抽象(思考)的人类相关认知各种层面的概念;深度网络可以识别从二维图片内容、不同角度、3d3维旋转、空间物体识别、运动特征识别、3维空间运动旋转不同角度物体的识别到为物体命名;这些是具体物体的识别

同一场景多个物体的识别:如何对多个物体的关系进行识别-建立在前面物体识别基础上的神经网络,比如大小比较的识别、上下概念的学习;从基本3d物体的识别,到基本的抽象概念的识别或分类;

从视觉从最基础的光线开始构建机器对世界的认识,从平面图像、双目摄像头、3d物体、3d场景;从具体物体到抽象关系(大小上下);
        视觉抽象层次:光线的点、线、边、角到物体的部分轮廓-到物体的大致轮廓-到物体的整体识别(深度学习已经实现)-到运动物体识别
阅读  ┆ 评论  ┆ 转载 ┆ 收藏 
标签:

杂谈

Extreme learning machine (ELM) has gained increasing interest from various research fields recently. Inthis review, we aim to report the current state of the theoretical research and practical advances on thissubject. We first give an overview of ELM from the theoretical perspective, including the interpolationtheory, universal approximation capability, and generalization ability. Then we focus on the variousimprovements made to ELM which further improve its stability, sparsity and accuracy under generalor specific conditions. Apart from classification and regression, ELM has recently been extended forclustering, feature selection, representational learning and many other learning tasks. These newlyemerging algorithms greatly expand the applications of ELM. From implementation aspect, hardwareimplementation and parallel computation techniques have substantially sped up the training of ELM,making it feasible for big data pr
阅读  ┆ 评论  ┆ 转载 ┆ 收藏 
标签:

杂谈

Deep learning is a broad set of techniques that uses multiple layers of representation to automatically learn relevant features directly from structured data. Recently, such techniques have yielded record-breaking results on a diverse set of difficult machine learning tasks in computer vision, speech recognition, and natural language processing. Despite the enormous success of deep learning, relatively little is understood theoretically about why these techniques are so successful at feature learning and compression. Here, we show that deep learning is intimately related to one of the most important and successful techniques in theoretical physics, the renormalization group (RG). RG is an iterative coarse-graining scheme that allows for the extraction of relevant features (i.e. operators) as a physical system is examined at different length scales. We construct an exact mapping
阅读  ┆ 评论  ┆ 转载 ┆ 收藏 
标签:

杂谈

Deep learning is a broad set of techniques that uses multiple layers of representation to automatically learn relevant features directly from structured data. Recently, such techniques have yielded record-breaking results on a diverse set of difficult machine learning tasks in computer vision, speech recognition, and natural language processing. Despite the enormous success of deep learning, relatively little is understood theoretically about why these techniques are so successful at feature learning and compression. Here, we show that deep learning is intimately related to one of the most important and successful techniques in theoretical physics, the renormalization
阅读  ┆ 评论  ┆ 转载 ┆ 收藏 
标签:

奇点大学

百度百家

大脑

神经

教育

 1124,百度百家举办第六次线下活动,邀请了奇点大学的三位讲师在太庙开公开课。


左:

阅读  ┆ 评论  ┆ 转载 ┆ 收藏 

Frontiers is Open Science

Frontiers is an online platform for the scientific community to publish open-access articles and network with colleagues.

    阅读  ┆ 评论  ┆ 转载 ┆ 收藏 
    个人资料
    CreateAMind
    CreateAMind
    • 博客等级:
    • 博客积分:0
    • 博客访问:9,605
    • 关注人气:13
    • 获赠金笔:0支
    • 赠出金笔:0支
    • 荣誉徽章:
    访客
    加载中…
    好友
    加载中…
    评论
    加载中…
    分类
      

    新浪BLOG意见反馈留言板 欢迎批评指正

    新浪简介 | About Sina | 广告服务 | 联系我们 | 招聘信息 | 网站律师 | SINA English | 会员注册 | 产品答疑

    新浪公司 版权所有