关于BDC类中的属性temperature #15
Comments
|
您好！感谢关注。这个参数是论文中所指的温度参数，实际使用时，我们把它加到了BDC里计算特征的两两欧氏距离中。由于在求解BDC矩阵时算完特征之间的欧氏距离后，剩下的运算均是线性的，故实际上可认为加了一个具有一定初始值的温度在logits上。 |
|
感谢解惑！ |
请问一下，这里的温度系数是一个可学习的参数吗 |
您好!感谢关注。此时的温度是一个可学习的参数。在以一定初始值初始化后进行预训练，对于Meta DeepBDC，元训练时先继承预训练的所学得的温度值后，再进行训练。 |
|
好的，非常感谢，那是否可以通过bdc矩阵来判断究竟是哪个像素与support区别较大<br/><br/><br/><html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
</head>
<body>
<div id="imail_signature" class="ntes-signature"> <a href="https://dashi.163.com/projects/signature-manager/detail/index.html?ftlId=1&name=yhbai%40zju.edu.cn&uid=yhbai%40zju.edu.cn&iconUrl=https%3A%2F%2Fmail-online.nosdn.127.net%2Fqiyelogo%2FdefaultAvatar.png&email=%E9%82%AE%E7%AE%B1%EF%BC%9Ayhbai%40zju.edu.cn&items=%5B%22%E9%82%AE%E7%AE%B1%EF%BC%9Ayhbai%40zju.edu.cn%22%5D&logoUrl=https%3A%2F%2Fmail-online.nosdn.127.net%2Fqiyeicon%2Fdb5d090b4290fa1cc857f03f01662859.png" style="display:block;background:#fff; max-width: 400px; _width: 400px;padding:15px 0 10px 0;text-decoration: none; outline:none;-webkit-tap-highlight-color:transparent;-webkit-text-size-adjust:none !important;text-size-adjust:none !important;">
<table cellpadding="0" style='width: 100%; max-width: 100%; table-layout: fixed; border-collapse: collapse;color: #9b9ea1;font-size: 14px;line-height:1.3;-webkit-text-size-adjust:none !important;text-size-adjust:none !important;'>
<tbody style="font-family: 'PingFang SC', 'Hiragino Sans GB','WenQuanYi Micro Hei', 'Microsoft Yahei', '微软雅黑', verdana !important; word-wrap:break-word; word-break:break-all;-webkit-text-size-adjust:none !important;text-size-adjust:none !important;">
<tr>
<td width="38" style="padding:0; box-sizing: border-box; width: 38px;">
<img width="38" height="38" style="vertical-align:middle; width: 38px; height: 38px; border-radius:50%;" src="https://mail-online.nosdn.127.net/qiyelogo/defaultAvatar.png" />
</td>
<td style='padding: 0 30px 0 10px; color: #31353b;'>
<div style="font-size: 16px;font-weight:bold; width:100%; white-space: nowrap; overflow:hidden;text-overflow: ***@***.***</div>
</td>
<td width="72" style="text-align:right; width: 72px;">
<img width="72" height="20" style="width: 72px; height: 20px;" src="https://mail-online.nosdn.127.net/qiyelogo/db5d090b4290fa1cc857f03f01662859.png" />
</td>
</tr>
<tr width="100%" style="font-size: 14px !important; width: 100%;">
<td colspan='3' style="padding:10px 0 0 0; font-size:14px !important; width: 100%;">
<div style="width: 100%;font-size: 14px ***@***.***</div>
</td>
</tr>
</tbody>
</table>
</a></div><br/><br/><br/><div class="ntes-mailmaster-quote" style="padding-top: 1px; padding-bottom: 1px" >
<div style=" margin-top: 20px; margin-bottom: 12px; font-size: 14px; line-height: 1.25; color: #89898c; " >---- 回复的原邮件 ----</div>
<div style=" margin-bottom: 12px; font-size: 13px; line-height: 1.25; color: #2c2c2e; padding: 2px; border-radius: 8px; background-color: #f0f0f0; " >
<table width="100%" cellpadding="0" cellspacing="10" border="0">
<tr>
<td valign="top" style=" width: 4em; font-size: 13px; line-height: 1.25; color: #89898c; white-space: nowrap; " >发件人</td>
<td valign="top" style=" font-size: 13px; line-height: 1.25; color: #2c2c2e; word-break: break-all; " ><a class="mail-from" style="color: #1c83eb; text-decoration: none" ***@***.***" >***@***.***></a></td>
</tr>
<tr>
<td valign="top" style=" width: 4em; font-size: 13px; line-height: 1.25; color: #89898c; white-space: nowrap; " >日期</td>
<td class="mail-date" valign="top" style=" font-size: 13px; line-height: 1.25; color: #2c2c2e; word-break: break-all; " >2023年07月13日 11:03</td>
</tr>
<tr style="">
<td valign="top" style=" width: 4em; font-size: 13px; line-height: 1.25; color: #89898c; " >收件人</td>
<td valign="top" style=" font-size: 13px; line-height: 1.25; color: #2c2c2e; word-break: break-all; " ><a class="mail-to" style="color: #1c83eb; text-decoration: none" ***@***.***" >***@***.***></a></td>
</tr>
<tr style="">
<td valign="top" style=" width: 4em; font-size: 13px; line-height: 1.25; color: #89898c; " >抄送至</td>
<td valign="top" style=" font-size: 13px; line-height: 1.25; color: #2c2c2e; word-break: break-all; " ><a class="mail-cc" style="color: #1c83eb; text-decoration: none" ***@***.***" >***@***.***></a>、<a class="mail-cc" style="color: #1c83eb; text-decoration: none" ***@***.***" >***@***.***></a></td>
</tr>
<tr>
<td valign="top" style=" width: 4em; font-size: 13px; line-height: 1.25; color: #89898c; " >主题</td>
<td class="mail-subject" valign="top" style=" font-size: 13px; line-height: 1.25; color: #2c2c2e; word-break: break-all; " >Re: [Fei-Long121/DeepBDC] 关于BDC类中的属性temperature (Issue #15)</td>
</tr>
</table>
</div>
<div><p></p>
<blockquote>
<blockquote>
<p dir="auto">您好！感谢关注。这个参数是论文中所指的温度参数，实际使用时，我们把它加到了BDC里计算特征的两两欧氏距离中。由于在求解BDC矩阵时算完特征之间的欧氏距离后，剩下的运算均是线性的，故实际上可认为加了一个具有一定初始值的温度在logits上。</p>
</blockquote>
<p dir="auto">请问一下，这里的温度系数是一个可学习的参数吗</p>
</blockquote>
<p dir="auto">您好!感谢关注。此时的温度是一个可学习的参数。在以一定初始值初始化后进行预训练，对于Meta DeepBDC，元训练时先继承预训练的所学得的温度值后，再进行训练。</p>
<p style="font-size:small;-webkit-text-size-adjust:none;color:#666;">—<br />Reply to this email directly, <a href="#15 (comment)">view it on GitHub</a>, or <a href="https://github.com/notifications/unsubscribe-auth/A3YTUTS35XKPDDPC3DQ7V4LXP5QOXANCNFSM6AAAAAARMU2IHM">unsubscribe</a>.<br />You are receiving this because you commented.<img src="https://github.com/notifications/beacon/A3YTUTXCZN4ITUKROWQA2UTXP5QOXA5CNFSM6AAAAAARMU2IHOWGG33NNVSW45C7OR4XAZNMJFZXG5LFINXW23LFNZ2KUY3PNVWWK3TUL5UWJTTBLTSDU.gif" height="1" width="1" alt="" /><span style="color: transparent; font-size: 0; display: none; visibility: hidden; overflow: hidden; opacity: 0; width: 0; height: 0; max-width: 0; max-height: 0; mso-hide: all">Message ID: <span><Fei-Long121/DeepBDC/issues/15/1633477690</span><span>@</span><span>github</span><span>.</span><span>com></span></span></p>
<script type="application/ld+json">[
{
***@***.***": "http://schema.org",
***@***.***": "EmailMessage",
"potentialAction": {
***@***.***": "ViewAction",
"target": "#15 (comment)",
"url": "#15 (comment)",
"name": "View Issue"
},
"description": "View this Issue on GitHub",
"publisher": {
***@***.***": "Organization",
"name": "GitHub",
"url": "https://github.com"
}
}
]</script></div>
</div>
</body>
</html>
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
在BDC类的构造函数中，最后几行对属性temperature进行了赋值
self.temperature = nn.Parameter(torch.log((1. / (2 * input_dim[1]*input_dim[2])) * torch.ones(1,1)), requires_grad=True)请问该参数对应于论文中的哪个公式呢？
The text was updated successfully, but these errors were encountered: