×

The Variational Bayes Method i

消耗积分:0 | 格式:rar | 大小:6656 | 2009-02-17

LOVEyou99

分享资料个

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 How to be a Bayesian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The Variational Bayes (VB) Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 A First Example of the VB Method: Scalar Additive Decomposition 3
1.3.1 A First Choice of Prior. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.2 The Prior Choice Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 The VB Method in its Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 VB as a Distributional Approximation . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6 Layout of the Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.7 Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Bayesian Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1 Bayesian Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1 Off-line vs. On-line Parametric Inference . . . . . . . . . . . . . . . . 14
2.2 Bayesian Parametric Inference: the Off-Line Case . . . . . . . . . . . . . . . 15
2.2.1 The Subjective Philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.2 Posterior Inferences and Decisions . . . . . . . . . . . . . . . . . . . . . . 16
2.2.3 Prior Elicitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.3.1 Conjugate priors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Bayesian Parametric Inference: the On-line Case . . . . . . . . . . . . . . . . 19
2.3.1 Time-invariant Parameterization . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2 Time-variant Parameterization . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.3 Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3 Off-line Distributional Approximations and the Variational Bayes
Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1 Distributional Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 How to Choose a Distributional Approximation . . . . . . . . . . . . . . . . . 26
3.2.1 Distributional Approximation as an Optimization Problem . . 26
3.2.2 The Bayesian Approach to Distributional Approximation . . . 27

3.3 The Variational Bayes (VB) Method of Distributional Approximation 28
3.3.1 The VB Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.2 The VB Method of Approximation as an Operator . . . . . . . . 32
3.3.3 The VBMethod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3.4 The VB Method for Scalar Additive Decomposition . . . . . . . 37
3.4 VB-related Distributional Approximations . . . . . . . . . . . . . . . . . . . . . . 39
3.4.1 Optimization with Minimum-Risk KL Divergence . . . . . . . . 39
3.4.2 Fixed-form (FF) Approximation . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4.3 Restricted VB (RVB) Approximation . . . . . . . . . . . . . . . . . . . 40
3.4.3.1 Adaptation of the VB method for the RVB
Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4.3.2 The Quasi-Bayes (QB) Approximation . . . . . . . . . 42
3.4.4 The Expectation-Maximization (EM) Algorithm . . . . . . . . . . 44
3.5 Other Deterministic Distributional Approximations . . . . . . . . . . . . . . 45
3.5.1 The Certainty Equivalence Approximation . . . . . . . . . . . . . . . 45
3.5.2 The Laplace Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.5.3 The Maximum Entropy (MaxEnt) Approximation . . . . . . . . . 45
3.6 Stochastic Distributional Approximations . . . . . . . . . . . . . . . . . . . . . . 46
3.6.1 Distributional Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.7 Example: Scalar Multiplicative Decomposition . . . . . . . . . . . . . . . . . . 48
3.7.1 Classical Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.7.2 The Bayesian Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.7.3 Full Bayesian Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.7.4 The Variational Bayes (VB) Approximation . . . . . . . . . . . . . . 51
3.7.5 Comparison with Other Techniques . . . . . . . . . . . . . . . . . . . . . 54
3.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4 Principal Component Analysis and Matrix Decompositions . . . . . . . . . 57
4.1 Probabilistic Principal Component Analysis (PPCA) . . . . . . . . . . . . . 58
4.1.1 Maximum Likelihood (ML) Estimation for the PPCA Model 59
4.1.2 Marginal Likelihood Inference of A. . . . . . . . . . . . . . . . . . . . . 61
4.1.3 Exact Bayesian Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.1.4 The Laplace Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 The Variational Bayes (VB) Method for the PPCA Model . . . . . . . . . 62
4.3 Orthogonal Variational PCA (OVPCA) . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3.1 The Orthogonal PPCA Model . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3.2 The VB Method for the Orthogonal PPCA Model . . . . . . . . . 70
4.3.3 Inference of Rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.3.4 Moments of the Model Parameters . . . . . . . . . . . . . . . . . . . . . . 78
4.4 Simulation Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4.1 Convergence to Orthogonal Solutions: VPCA vs. FVPCA . . 79
4.4.2 Local Minima in FVPCA and OVPCA . . . . . . . . . . . . . . . . . . 82
4.4.3 Comparison of Methods for Inference of Rank . . . . . . . . . . . . 83
4.5 Application: Inference of Rank in a Medical Image Sequence. . . . . . 85
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

5 Functional Analysis of Medical Image Sequences . . . . . . . . . . . . . . . . . 89
5.1 A Physical Model for Medical Image Sequences . . . . . . . . . . . . . . . . 90
5.1.1 Classical Inference of the Physiological Model . . . . . . . . . . . 92
5.2 The FAMIS Observation Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.2.1 Bayesian Inference of FAMIS and Related Models . . . . . . . . 94
5.3 The VB Method for the FAMIS Model . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.4 The VBMethod for FAMIS: Alternative Priors . . . . . . . . . . . . . . . . . . 99
5.5 Analysis of Clinical Data Using the FAMIS Model . . . . . . . . . . . . . . 102
5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6 On-line Inference of Time-Invariant Parameters . . . . . . . . . . . . . . . . . . 109
6.1 Recursive Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.2 Bayesian Recursive Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.2.1 The Dynamic Exponential Family (DEF) . . . . . . . . . . . . . . . . 112
6.2.2 Example: The AutoRegressive (AR) Model . . . . . . . . . . . . . . 114
6.2.3 Recursive Inference of non-DEF models . . . . . . . . . . . . . . . . . 117
6.3 The VB Approximation in On-Line Scenarios . . . . . . . . . . . . . . . . . . . 118
6.3.1 Scenario I: VB-Marginalization for Conjugate Updates . . . . 118
6.3.2 Scenario II: The VB Method in One-Step Approximation . . . 121
6.3.3 Scenario III: Achieving Conjugacy in non-DEF Models via
the VB Approximation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.3.4 The VB Method in the On-Line Scenarios . . . . . . . . . . . . . . . 126
6.4 Related Distributional Approximations . . . . . . . . . . . . . . . . . . . . . . . . 127
6.4.1 The Quasi-Bayes (QB) Approximation in On-Line Scenarios 128
6.4.2 Global Approximation via the Geometric Approach . . . . . . . 128
6.4.3 One-step Fixed-Form (FF) Approximation . . . . . . . . . . . . . . . 129
6.5 On-line Inference of a Mixture of AutoRegressive (AR) Models . . . 130
6.5.1 The VBMethod for ARMixtures. . . . . . . . . . . . . . . . . . . . . . . 130
6.5.2 Related Distributional Approximations for AR Mixtures . . . 133
6.5.2.1 The Quasi-Bayes (QB) Approximation . . . . . . . . . . 133
6.5.2.2 One-step Fixed-Form (FF) Approximation . . . . . . . 135
6.5.3 Simulation Study: On-line Inference of a Static Mixture . . . . 135
6.5.3.1 Inference of a Many-Component Mixture . . . . . . . . 136
6.5.3.2 Inference of a Two-Component Mixture . . . . . . . . . 136
6.5.4 Data-Intensive Applications of Dynamic Mixtures . . . . . . . . . 139
6.5.4.1 Urban Vehicular Traffic Prediction . . . . . . . . . . . . . . 141
6.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7 On-line Inference of Time-Variant Parameters . . . . . . . . . . . . . . . . . . . . 145
7.1 Exact Bayesian Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
7.2 The VB-Approximation in Bayesian Filtering . . . . . . . . . . . . . . . . . . . 147
7.2.1 The VB method for Bayesian Filtering . . . . . . . . . . . . . . . . . . 149
7.3 Other Approximation Techniques for Bayesian Filtering . . . . . . . . . . 150
7.3.1 Restricted VB (RVB) Approximation . . . . . . . . . . . . . . . . . . . 150
7.3.2 Particle Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

7.3.3 Stabilized Forgetting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.3.3.1 The Choice of the Forgetting Factor . . . . . . . . . . . . . 154
7.4 The VB-Approximation in Kalman Filtering . . . . . . . . . . . . . . . . . . . . 155
7.4.1 The VB method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
7.4.2 Loss of Moment Information in the VB Approximation . . . . 158
7.5 VB-Filtering for the Hidden Markov Model (HMM) . . . . . . . . . . . . . 158
7.5.1 Exact Bayesian filtering for known T . . . . . . . . . . . . . . . . . . . 159
7.5.2 The VB Method for the HMM Model with Known T . . . . . . 160
7.5.3 The VB Method for the HMM Model with Unknown T . . . . 162
7.5.4 Other Approximate Inference Techniques . . . . . . . . . . . . . . . 164
7.5.4.1 Particle Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
7.5.4.2 Certainty Equivalence Approach . . . . . . . . . . . . . . . 165
7.5.5 Simulation Study: Inference of Soft Bits . . . . . . . . . . . . . . . . . 166
7.6 The VB-Approximation for an Unknown Forgetting Factor . . . . . . . 168
7.6.1 Inference of a Univariate AR Model with Time-Variant
Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
7.6.2 Simulation Study: Non-stationary AR Model Inference via
Unknown Forgetting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
7.6.2.1 Inference of an AR Process with Switching
Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
7.6.2.2 Initialization of Inference for a Stationary AR
Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
8 The Mixture-based Extension of the AR Model (MEAR) . . . . . . . . . . . . 179
8.1 The Extended AR (EAR) Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
8.1.1 Bayesian Inference of the EAR Model. . . . . . . . . . . . . . . . . . . 181
8.1.2 Computational Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
8.2 The EAR Model with Unknown Transformation: the MEAR Model 182
8.3 The VB Method for the MEAR Model . . . . . . . . . . . . . . . . . . . . . . . . . 183
8.4 Related Distributional Approximations for MEAR . . . . . . . . . . . . . . . 186
8.4.1 The Quasi-Bayes (QB) Approximation . . . . . . . . . . . . . . . . . . 186
8.4.2 The Viterbi-Like (VL) Approximation . . . . . . . . . . . . . . . . . . . 187
8.5 Computational Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
8.6 The MEAR Model with Time-Variant Parameters . . . . . . . . . . . . . . . . 191
8.7 Application: Inference of an AR Model Robust to Outliers . . . . . . . . 192
8.7.1 Design of the Filter-bank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
8.7.2 Simulation Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
8.8 Application: Inference of an AR Model Robust to Burst Noise . . . . 196
8.8.1 Design of the Filter-Bank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
8.8.2 Simulation Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
8.8.3 Application in Speech Reconstruction . . . . . . . . . . . . . . . . . . . 201
8.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

声明:本文内容及配图由入驻作者撰写或者入驻合作网站授权转载。文章观点仅代表作者本人,不代表电子发烧友网立场。文章及其配图仅供工程师学习之用,如有内容侵权或者其他违规问题,请联系本站处理。 举报投诉

评论(0)
发评论

下载排行榜

全部0条评论

快来发表一下你的评论吧 !

'+ '

'+ '

'+ ''+ '
'+ ''+ ''+ '
'+ ''+ '' ); $.get('/article/vipdownload/aid/'+webid,function(data){ if(data.code ==5){ $(pop_this).attr('href',"/login/index.html"); return false } if(data.code == 2){ //跳转到VIP升级页面 window.location.href="//m.obk20.com/vip/index?aid=" + webid return false } //是会员 if (data.code > 0) { $('body').append(htmlSetNormalDownload); var getWidth=$("#poplayer").width(); $("#poplayer").css("margin-left","-"+getWidth/2+"px"); $('#tips').html(data.msg) $('.download_confirm').click(function(){ $('#dialog').remove(); }) } else { var down_url = $('#vipdownload').attr('data-url'); isBindAnalysisForm(pop_this, down_url, 1) } }); }); //是否开通VIP $.get('/article/vipdownload/aid/'+webid,function(data){ if(data.code == 2 || data.code ==5){ //跳转到VIP升级页面 $('#vipdownload>span').text("开通VIP 免费下载") return false }else{ // 待续费 if(data.code == 3) { vipExpiredInfo.ifVipExpired = true vipExpiredInfo.vipExpiredDate = data.data.endoftime } $('#vipdownload .icon-vip-tips').remove() $('#vipdownload>span').text("VIP免积分下载") } }); }).on("click",".download_cancel",function(){ $('#dialog').remove(); }) var setWeixinShare={};//定义默认的微信分享信息,页面如果要自定义分享,直接更改此变量即可 if(window.navigator.userAgent.toLowerCase().match(/MicroMessenger/i) == 'micromessenger'){ var d={ title:'The Variational Bayes Method i',//标题 desc:$('[name=description]').attr("content"), //描述 imgUrl:'https://'+location.host+'/static/images/ele-logo.png',// 分享图标,默认是logo link:'',//链接 type:'',// 分享类型,music、video或link,不填默认为link dataUrl:'',//如果type是music或video,则要提供数据链接,默认为空 success:'', // 用户确认分享后执行的回调函数 cancel:''// 用户取消分享后执行的回调函数 } setWeixinShare=$.extend(d,setWeixinShare); $.ajax({ url:"//www.obk20.com/app/wechat/index.php?s=Home/ShareConfig/index", data:"share_url="+encodeURIComponent(location.href)+"&format=jsonp&domain=m", type:'get', dataType:'jsonp', success:function(res){ if(res.status!="successed"){ return false; } $.getScript('https://res.wx.qq.com/open/js/jweixin-1.0.0.js',function(result,status){ if(status!="success"){ return false; } var getWxCfg=res.data; wx.config({ //debug: true, // 开启调试模式,调用的所有api的返回值会在客户端alert出来,若要查看传入的参数,可以在pc端打开,参数信息会通过log打出,仅在pc端时才会打印。 appId:getWxCfg.appId, // 必填,公众号的唯一标识 timestamp:getWxCfg.timestamp, // 必填,生成签名的时间戳 nonceStr:getWxCfg.nonceStr, // 必填,生成签名的随机串 signature:getWxCfg.signature,// 必填,签名,见附录1 jsApiList:['onMenuShareTimeline','onMenuShareAppMessage','onMenuShareQQ','onMenuShareWeibo','onMenuShareQZone'] // 必填,需要使用的JS接口列表,所有JS接口列表见附录2 }); wx.ready(function(){ //获取“分享到朋友圈”按钮点击状态及自定义分享内容接口 wx.onMenuShareTimeline({ title: setWeixinShare.title, // 分享标题 link: setWeixinShare.link, // 分享链接 imgUrl: setWeixinShare.imgUrl, // 分享图标 success: function () { setWeixinShare.success; // 用户确认分享后执行的回调函数 }, cancel: function () { setWeixinShare.cancel; // 用户取消分享后执行的回调函数 } }); //获取“分享给朋友”按钮点击状态及自定义分享内容接口 wx.onMenuShareAppMessage({ title: setWeixinShare.title, // 分享标题 desc: setWeixinShare.desc, // 分享描述 link: setWeixinShare.link, // 分享链接 imgUrl: setWeixinShare.imgUrl, // 分享图标 type: setWeixinShare.type, // 分享类型,music、video或link,不填默认为link dataUrl: setWeixinShare.dataUrl, // 如果type是music或video,则要提供数据链接,默认为空 success: function () { setWeixinShare.success; // 用户确认分享后执行的回调函数 }, cancel: function () { setWeixinShare.cancel; // 用户取消分享后执行的回调函数 } }); //获取“分享到QQ”按钮点击状态及自定义分享内容接口 wx.onMenuShareQQ({ title: setWeixinShare.title, // 分享标题 desc: setWeixinShare.desc, // 分享描述 link: setWeixinShare.link, // 分享链接 imgUrl: setWeixinShare.imgUrl, // 分享图标 success: function () { setWeixinShare.success; // 用户确认分享后执行的回调函数 }, cancel: function () { setWeixinShare.cancel; // 用户取消分享后执行的回调函数 } }); //获取“分享到腾讯微博”按钮点击状态及自定义分享内容接口 wx.onMenuShareWeibo({ title: setWeixinShare.title, // 分享标题 desc: setWeixinShare.desc, // 分享描述 link: setWeixinShare.link, // 分享链接 imgUrl: setWeixinShare.imgUrl, // 分享图标 success: function () { setWeixinShare.success; // 用户确认分享后执行的回调函数 }, cancel: function () { setWeixinShare.cancel; // 用户取消分享后执行的回调函数 } }); //获取“分享到QQ空间”按钮点击状态及自定义分享内容接口 wx.onMenuShareQZone({ title: setWeixinShare.title, // 分享标题 desc: setWeixinShare.desc, // 分享描述 link: setWeixinShare.link, // 分享链接 imgUrl: setWeixinShare.imgUrl, // 分享图标 success: function () { setWeixinShare.success; // 用户确认分享后执行的回调函数 }, cancel: function () { setWeixinShare.cancel; // 用户取消分享后执行的回调函数 } }); }); }); } }); } function openX_ad(posterid, htmlid, width, height) { if ($(htmlid).length > 0) { var randomnumber = Math.random(); var now_url = encodeURIComponent(window.location.href); var ga = document.createElement('iframe'); ga.src = 'https://www1.elecfans.com/www/delivery/myafr.php?target=_blank&cb=' + randomnumber + '&zoneid=' + posterid+'&prefer='+now_url; ga.width = width; ga.height = height; ga.frameBorder = 0; ga.scrolling = 'no'; var s = $(htmlid).append(ga); } } openX_ad(828, '#berry-300', 300, 250);