<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[JasonSparrow_1的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/JasonSparrow_1</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; JasonSparrow_1]]></copyright><item><title><![CDATA[Unity Shader入门精要(第三，四章)]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/102552390</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/102552390</guid><author>JasonSparrow_1</author><pubDate>Mon, 21 Oct 2019 21:00:11 +0800</pubDate><description><![CDATA[Unity Shader 概述

Shader类型：
Standard Surface Shader：标准光照Shader
Unlit Shader：不包含光照
Image Effect Shader：实现屏幕后处理基本模板
Compute Shader：利用并行性进行一些物馆计算
ShaderLab：


...]]></description><category></category></item><item><title><![CDATA[Construction and Evaluation of Chinese Emotion Classification Model]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/102625853</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/102625853</guid><author>JasonSparrow_1</author><pubDate>Sat, 19 Oct 2019 21:14:15 +0800</pubDate><description><![CDATA[目录AbstractIntroductionEmotion Word ClassifyingDatabase ConstructionDatabase StructureEmotion Classification ModelLexical Analysis(词法分析)Syntax Analysis(语法分析)
Abstract

利用我们所构建的情感同义词典来获取情感信息的半自动获取技术
情绪同...]]></description><category></category></item><item><title><![CDATA[汉语普通话情感言语的声学分析]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/102620421</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/102620421</guid><author>JasonSparrow_1</author><pubDate>Fri, 18 Oct 2019 11:35:07 +0800</pubDate><description><![CDATA[Abstract

愤怒、恐惧、喜悦和悲伤情绪的声音表达与中性言语的关系
针对汉语语音信号中带有情绪唤起的词重音进行研究

Introduction

汉语愤怒、恐惧、喜悦和悲伤的声学实现有两个维度，即语音和韵律
首先分析了声学特征，包括持续时间、短时振幅和音高，不仅在句子层面，而且在单词层面。第二，将检验两个水平之间的比较结果。第三，我们分析了普通话情感话语中的重音。

Emotional sp...]]></description><category></category></item><item><title><![CDATA[Unity Shader入门精要(第二章)]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/102454323</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/102454323</guid><author>JasonSparrow_1</author><pubDate>Mon, 14 Oct 2019 17:08:49 +0800</pubDate><description><![CDATA[渲染流程：

应用阶段(Application Stage)

准备好场景数据，摄像机位置等
粗粒度剔除，把不可见的物体剔除出去
设置渲染状态，材质，纹理，shader等
最重要的是输出渲染图元(点，线，三角面等)，给几何阶段


几何阶段(Geometry Stage)
把顶点坐标变换到屏幕空间中，输出屏幕空间的二维顶点坐标、每个顶点对应的深度值、着色等
光栅化阶段(Rasterizer Sta...]]></description><category></category></item><item><title><![CDATA[脑机交互促进学习有效发生的路径及实验研究]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/100859472</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/100859472</guid><author>JasonSparrow_1</author><pubDate>Mon, 16 Sep 2019 10:57:54 +0800</pubDate><description><![CDATA[研究问题
开发了基于脑机交互技术的注意力监控与促进系统
“整体型”：结构化信息
“序列型”：偏向遵循一个精心设计的序列
系统工作原理
如果学生的注意力下降到某一个设定的阈值(40)，学习系统将自动暂停学习视频的播放， 同时启动注意力提升系统并向学生提问， 这些问题都是已经播放的视频中的相关内容。如果学生回答正确，视频则能继续播放，进入下一阶段的学习；如果学生未能正确回答问题，视频将会自动倒退5秒（...]]></description><category></category></item><item><title><![CDATA[虚拟现实和脑电波反馈系统对于创造力表现影响的实证研究]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/100835156</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/100835156</guid><author>JasonSparrow_1</author><pubDate>Sun, 15 Sep 2019 16:29:10 +0800</pubDate><description><![CDATA[研究的问题：

虚拟现实是否有助于个体创造力表现？主要体现在哪些方面？
什么样的脑波序列与个人创造力表现紧密相关？
在虚拟现实环境中，基于脑电波的声音反馈是否有助于个人创造力表现？


...]]></description><category></category></item><item><title><![CDATA[Personalized Estimation of Engagement from Videos Using Deep Reinforcement Learning]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/95474578</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/95474578</guid><author>JasonSparrow_1</author><pubDate>Fri, 19 Jul 2019 14:57:39 +0800</pubDate><description><![CDATA[Introduction

能够准确的识别用户的参与状态能很好的使一些系统能够在恰当的时间提供必要的交互来完成交互的任务。
参与评估的一个基本挑战是人们如何激发参与以及如何在计算参与模型中表示参与。
传统的方法使用非语言的参与线索，如注视模式、身体姿势、韵律、面部表情、空间关系和任务上下文行为。
因为个体与个体之间的差异性较大，所以造成训练集训练出来的模型利用测试集进行测试的结果效果很差。
然而，...]]></description><category></category></item><item><title><![CDATA[A fuzzy physiological approach for continuously modeling emotion during interaction with play game]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/94020307</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/94020307</guid><author>JasonSparrow_1</author><pubDate>Tue, 02 Jul 2019 17:02:01 +0800</pubDate><description><![CDATA[Introduction
传统的针对productivity environments(task表现)的度量手段不能很好的适应目前的affective environments，因为我们现在考虑的不是performance，而是通过现在的游戏环境，能够带给我们什么样的情绪。
Affective evaluation of entertainment technologies

传统的评估情绪的方法...]]></description><category></category></item><item><title><![CDATA[Learning Latent Dynamics for Planning from Pixels]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/92811546</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/92811546</guid><author>JasonSparrow_1</author><pubDate>Tue, 25 Jun 2019 11:32:03 +0800</pubDate><description><![CDATA[Abstract

The Deep Planning Network (PlaNet), a purely model-based agent that learns the environment dynamics from images and chooses actions through fast online planning in latent space.
PlaNet一个纯的mo...]]></description><category></category></item><item><title><![CDATA[Human-level performance in 3D multiplayer games with population based reinforcement learning]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/90759067</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/90759067</guid><author>JasonSparrow_1</author><pubDate>Tue, 18 Jun 2019 17:11:42 +0800</pubDate><description><![CDATA[Background

Played on both indoor- and outdoor-themed maps that are randomly generated for each game.
The team with the greatest number of flag captures after five minutes wins.
Agents can tag opponen...]]></description><category></category></item><item><title><![CDATA[Learning From Video DRL &&OpenAI Five]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/89504602</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/89504602</guid><author>JasonSparrow_1</author><pubDate>Thu, 02 May 2019 20:49:01 +0800</pubDate><description><![CDATA[Overcoming sparse rewards in Deep RL: Curiosity, hindsight &amp; auxiliary tasks
Sparse rewards
比较离散的reward signal，也就是说，在玩儿一些特定游戏的时候无法确定是否通过一个reward就能明确判断结果的输赢。也就是说需要连续的决策动作序列才能到达成功的情况。如下图所示：

Reinfor...]]></description><category></category></item><item><title><![CDATA[CS294-第二讲 监督学习和模仿学习]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/88936618</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/88936618</guid><author>JasonSparrow_1</author><pubDate>Sat, 06 Apr 2019 15:20:20 +0800</pubDate><description><![CDATA[监督学习和模仿学习

]]></description><category></category></item><item><title><![CDATA[Reinforcement Learning易忘知识点总结]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/88849117</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/88849117</guid><author>JasonSparrow_1</author><pubDate>Fri, 29 Mar 2019 15:29:12 +0800</pubDate><description><![CDATA[value function


常用的公式：

表示可以对未来的期望打出一定的折扣，着重考虑一些固定的reward。
当γ\gammaγ的值为0时为只考虑立即不考虑长期；当值为1时立即和长期考虑同等重要。

状态值函数(state value function)和动作值函数(action value function Q函数)是不同的


...]]></description><category></category></item><item><title><![CDATA[Planning by Dynamic Programming]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/85721783</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/85721783</guid><author>JasonSparrow_1</author><pubDate>Thu, 21 Mar 2019 11:13:28 +0800</pubDate><description><![CDATA[…

]]></description><category></category></item><item><title><![CDATA[强化学习ActorCritic]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/88232753</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/88232753</guid><author>JasonSparrow_1</author><pubDate>Fri, 08 Mar 2019 15:10:02 +0800</pubDate><description><![CDATA[ActorCritic
Actor负责进行动作的奖惩，而Critic将对奖惩进行评估，从而对下一步的奖惩做出影响
Actor的算法基础是PolicyGradients，Critic的算法基础是Q-learning
Actor只能回合更新，而Critic部分可以单步更新
缺点就是空间的连续性，从而导致神经网络学不到东西，相关性比较强
为了解决这种问题，DeepMind将ActorCritic和DQN...]]></description><category></category></item><item><title><![CDATA[强化学习PolicyGradients]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/88029002</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/88029002</guid><author>JasonSparrow_1</author><pubDate>Tue, 05 Mar 2019 17:51:12 +0800</pubDate><description><![CDATA[Policy Gradients
通过奖惩机制reward，来对选择进行反向传递，从而确定是否需要进行加大，或者减小选择的力度

]]></description><category></category></item><item><title><![CDATA[强化学习DQN]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/87781373</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/87781373</guid><author>JasonSparrow_1</author><pubDate>Thu, 28 Feb 2019 14:47:25 +0800</pubDate><description><![CDATA[DQN
针对Q-learning做出改进。原Q-learning的算法不能支持较大的Q表的查询和遍历，而DQN则直接将状态和动作当作神经网络的输入值，在分析后得到Q值；或者只输入状态值，输出动作值，从而根据Q-learning的原则进行动作的选择。

针对第二种进行分析，如何更新NN中的参数


两个让DQN能够更好的学习的方法

Experience replay
随机抽取经历进行学习，从而打乱...]]></description><category></category></item><item><title><![CDATA[Tensorflow(二)]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/87858706</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/87858706</guid><author>JasonSparrow_1</author><pubDate>Sun, 24 Feb 2019 17:34:28 +0800</pubDate><description><![CDATA[Optimizer
Stochastic Gradient Descent (SGD)
基础方法，是mini batch中的一种，加了随机
Momentum
W+=−LearningRate∗dxW+=-LearningRate *dxW+=−LearningRate∗dx
即原始的WWW加上一个负的学习效率乘上一个矫正值dxdxdx
数学公式如下：

一定程度上保留之前更新的方向(像一个醉汉，随...]]></description><category></category></item><item><title><![CDATA[Tensorflow(一)]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/87785407</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/87785407</guid><author>JasonSparrow_1</author><pubDate>Wed, 20 Feb 2019 21:05:09 +0800</pubDate><description><![CDATA[…

]]></description><category></category></item><item><title><![CDATA[强化学习Sarsa]]></title><link>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/87776937</link><guid>https://blog.youkuaiyun.com/JasonSparrow_1/article/details/87776937</guid><author>JasonSparrow_1</author><pubDate>Wed, 20 Feb 2019 10:03:51 +0800</pubDate><description><![CDATA[Sarsa
算法如下：

相比Q-learning而言，Sarsa会比较胆小，不希望自己掉进陷阱里，从而做出的决策都比较谨慎；而Q-learning会选择一条虽然比较危险但是可以成功的道路
# off-policy
class QLearningTable(RL):
    def __init__(self, actions, learning_rate=0.01, reward_decay=0...]]></description><category></category></item></channel></rss>