40 Matching Annotations
  1. Dec 2019
    1. empirical comparison


    2. utterance-level representations (Section5.2) and analysis


    3. comparison


    4. empirical evaluation


    5. estimation


    6. formulation



    1. Softmax


    2. During the training, we want the embedding of each utteranceto be similar to the centroid of all that speaker’s embeddings, whileat the same time, far from other speakers’ centroids.


      1. 一个说话人的聚类中心足够靠近他的话音向量
      2. 不同说话人的聚类中心相互足够远离.
    3. cj


    4. xji


    1. torch.optim provides standard ways of updating the model so that the output starts tomore closely resemble the answers specified in the training data.

      与...相似 closely resemble

    2. adapt the model to


    3. The core PyTorch modules for building neural networks are located in torch.nn, which provides common neural network layers and other architectural components.


    4. Different nodes canbe invoked according to conditions on the outputs of the preceding nodes, for exam-ple, without a need for such conditions to be represented in the graph itself


  2. Nov 2019
  3. Jun 2019
    1. 要被印出的结构,在哪里被印出,第三个参数通常可以忽略


    1. 如何理解JavaScript语言是基于对象的,而不是面向对象的?
      1. 可以根据需要创建自己的对象
      2. 以原型对象为基础的继承机制
      3. 没有正式的类的概念
      4. 没有提供以类为基础的抽象,继承,重载等典型功能
      5. 弱类型的
    2. JavaScript语言有何特点
      1. 脚本语言
      2. 基于对象
      3. 简单性
      4. 安全性
      5. 动态性
      6. 跨平台性



    1. Pintos is a simple operating system framework for the 80x86 archi-tecture. It supports kernel threads, loading and running user programs, and a file system,but it implements all of these in a very simple way.

      简介: pintos是一个简单的80x86架构的操作系统框架.它支持内核线程,载入和运行用户程序,以及一个文件系统,但是所有的实现都十分简单.


  4. May 2019
    1. > (union '(a b c) '(c b s)) (A C B S) > (intersection '(a b c) '(b b c)) (B C) > (set-difference '(a b c d e) '(b e)) (A C D)

      union 并集 intersection 交集 set-difference 补集

  5. Apr 2019
    1. Computes sparse softmax cross entropy between logits and labels.


    2. labels: Tensor of shape [d_0, d_1, ..., d_{r-1}] (where r is rank of labels and result) and dtype int32 or int64. Each entry in labels must be an index in [0, num_classes). Other values will raise an exception when this op is run on CPU, and return NaN for corresponding loss and gradient rows on GPU.


    3. logits: Unscaled log probabilities of shape [d_0, d_1, ..., d_{r-1}, num_classes] and dtype float16, float32, or float64.


    4. trainable: If True, the default, also adds the variable to the graph collection


  6. Mar 2019
    1. ip命令可以说是无比强大了,它完全可以替换ifconfig、netstat、route、arp等命令,比如查看网卡eth1 IP地址:
    1. An X driver is responsible for translating kernel events into XInput events that can be used by applications such as GIMP. We provide an X driver that virtualizes the tablet into logical devices (e.g. stylus, eraser, touch) which report extended data like pressure and high-resolution coordinates.

      X driver 用来将内核事件转换为能被像GIMP这样的应用使用的Xinput事件.我们提供一个X driver 将数位板虚拟为支持扩展数据(像压感,高分辨率的定位)的数字设备. 如果应用程序没有响应数位板或者提供不了压感试试安装或者更新xf86-input-wacom.

    1. Tablet recognized but xsetwacom and similar tools do not display it


      问题可能来自xorg ,因为xf86-input-wacom配置xorg的配置文件在/usr/share/X11/xorg.conf.d/文件夹中,而xorg先执行此文件夹内容的东西,随后执行/etc/X11/xorg.conf.d文件夹中的配置内容.后面的配置会覆盖前面的内容,也就是说/etc/X11/xorg.conf.d中的某个内容会覆盖正确的配置内容.

    1. tf.reducemean


    2. 带指数衰减的学习率设置、使用正则化来避免过度拟合,以及使用滑动平均模型来使得最终模型更加健壮。

      指数衰减的学习率可以有效提高学习的速率同时减少震荡. 正则化将描述模型复杂度的函数添加到损失函数中.使用滑动平均模型....


  7. Feb 2019
    1. 可以便模型在测试数据上更健壮(robust)的方法一一滑动平均模型。

      tf.train.ExponentialMovingAverage 滑动平均模型的应用?和原理?

    2. 在预测商品销量时,如果预测多了(预测值比真实销量大),商家损失的是生产商品的成本;而如果预测少了(预测值比真实销量小〉,损失的则是商品的利润


    3. So位max回归

      softmax回归:  一般作为额外的处理层,将神经网络前向传播的结果转化为概率分布: 在[0,1]内相加为1的数.

    4. 深层神经网络
      1. 介绍深度学习与神经网络的概念
      2. 损失函数
      3. 反向传播算法


    1. Meanwhile, the reason for the problem is known: the newest wacom driver for linux does also support bluetooth for this device. However, on my laptop it is not possible to use the latest driver on a regular installation. I compliled and installed the driver, but the new custom kernel is blocked by secure boot. And without secure boot, I cannot boot my system. I can only boot live systems on pendrives without secure boot. So far I have tried ubuntu and mint. Not tried manjaro yet. In theory it is possible to sign custom kernel modules so they are not blocked by secure boot, but that task is way above my linux level.

      最新的Linux驱动支持蓝牙,但是在我的笔记本狄安娜不能够通过正常安装使用最新驱动.我编译安装了驱动但是新的custom kernel 被secure boot屏蔽了,但是没有secure boot,我又不能启动我的系统.

      https://github.com/linuxwacom/xf86-input-wacom/issues/32 网站中有网友表示:Although the USB portion of the tablet will work with as old as a 4.10 kernel, the Bluetooth functionality I believe specifically requires Linux 4.17 or later. If you install our "input-wacom" driver, the device should be detected and work correctly. 更新内核后,发现问题解决.

    1. Just experimented and discovered it's the kcm-wacomtablet that's the culprit. Removed that and wacomtablet-kderc and bluetooth works as expected. I


    1. A kernel driver is necessary to initalize the tablet and translate hardware-specific protocols into standard input events. We contribute patches directly to the Linux kernel, as well as maintain an out-of-tree kernel module that enables old kernels to work with many newer tablets.