search

zhuofupan / Tensorflow-Deep-Neural-Networks

用Tensorflow实现的深度神经网络。
136 stars 55 forks source link

高斯-伯努利RBM #11

Open cjlhxls opened 5 years ago

cjlhxls commented 5 years ago

您好，请问一下在guass-bin的重建过程中，act_function为什么用affine而不是guass? 即 def reconstruction(self,h): logits = tf.matmul(h, tf.transpose(self.W)) + self.bv prob_v=self.v_func(logits) return logits,prob_v

zhuofupan commented 5 years ago

因为，公式推导出来就是这样的，p(v|h)服从高斯分布，取这个分布的均值，也就是 linear(h)

发送自 Windows 10 版邮件https://go.microsoft.com/fwlink/?LinkId=550986应用

发件人: layout_toxin notifications@github.com 发送时间: Sunday, April 7, 2019 3:48:16 PM 收件人: fuzimaoxinan/Tensorflow-Deep-Neural-Networks 抄送: Subscribed 主题: [fuzimaoxinan/Tensorflow-Deep-Neural-Networks] 高斯-伯努利RBM (#11)

您好，请问一下在guass-bin的重建过程中，act_function为什么用affine而不是guass? 即 def reconstruction(self,h): logits = tf.matmul(h, tf.transpose(self.W)) + self.bv prob_v=self.v_func(logits) return logits,prob_v

― You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/fuzimaoxinan/Tensorflow-Deep-Neural-Networks/issues/11, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ANKxT0YqEoI4hB-Np3DJJSEzz7cZBqFtks5veaLAgaJpZM4cgnaB.

cjlhxls commented 5 years ago

不好意思还有个问题想要问您一下，DBN第一层为高斯-伯努利RBM的话，第二层还是高斯-伯努利RBM吗？或者是第二层应该是伯努利-伯努利型呢？ 谢谢！

------------------ 原始邮件 ------------------ 发件人: "Fuzz Pan"notifications@github.com; 发送时间: 2019年4月7日(星期天) 下午4:12 收件人: "fuzimaoxinan/Tensorflow-Deep-Neural-Networks"Tensorflow-Deep-Neural-Networks@noreply.github.com; 抄送: "470564568"470564568@qq.com; "Author"author@noreply.github.com; 主题: Re: [fuzimaoxinan/Tensorflow-Deep-Neural-Networks] 高斯-伯努利RBM (#11)

因为，公式推导出来就是这样的，p(v|h)服从高斯分布，取这个分布的均值，也就是 linear(h)

发送自 Windows 10 版邮件https://go.microsoft.com/fwlink/?LinkId=550986应用

发件人: layout_toxin notifications@github.com 发送时间: Sunday, April 7, 2019 3:48:16 PM 收件人: fuzimaoxinan/Tensorflow-Deep-Neural-Networks 抄送: Subscribed 主题: [fuzimaoxinan/Tensorflow-Deep-Neural-Networks] 高斯-伯努利RBM (#11)

您好，请问一下在guass-bin的重建过程中，act_function为什么用affine而不是guass? 即 def reconstruction(self,h): logits = tf.matmul(h, tf.transpose(self.W)) + self.bv prob_v=self.v_func(logits) return logits,prob_v

― You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/fuzimaoxinan/Tensorflow-Deep-Neural-Networks/issues/11, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ANKxT0YqEoI4hB-Np3DJJSEzz7cZBqFtks5veaLAgaJpZM4cgnaB.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

zhuofupan commented 5 years ago

你想要他是什么那就是什么