通常,卷积层之后是池化层,池化层基本上总结了由池化的接收场确定的邻域中的输出特征图激活。 例如,一个 2 x 2 的接收场将收集四个相邻的输出特征图激活的本地信息。 对于最大池操作,将选择四个激活的最大值作为输出,而对于平均池化,将选择四个激活的平均值。 合并降低了特征图的空间分辨率。 例如,对于具有 2 x 2 接收场的 224 x 224 尺寸的特征图池化操作,特征图的空间尺寸将减小为 112 x 112。
通常,卷积层之后是池化层,池化层基本上总结了由池化的接收场确定的邻域中的输出特征映射激活。 例如,一个 2 x 2 的接收场将收集四个相邻的输出特征映射激活的本地信息。 对于最大池操作,将选择四个激活的最大值作为输出,而对于平均池化,将选择四个激活的平均值。 合并降低了特征映射的空间分辨率。 例如,对于具有 2 x 2 接收场的 224 x 224 尺寸的特征映射池化操作,特征映射的空间尺寸将减小为 112 x 112。
要注意的一件事是,卷积运算减少了每层要学习的权重数。 例如,如果我们有一个空间尺寸为 224 x 224 的输入图像,而下一层的期望输出为尺寸为 224 x 224 的尺寸,那么对于具有完整连接的传统神经网络来说,要学习的权重数 是 224 x 224 x 224 x 224.对于具有相同输入和输出尺寸的卷积层,我们需要学习的只是滤波器内核的权重。 因此,如果我们使用 3 x 3 过滤器内核,我们只需要学习 9 个权重即可,而不是 224 x 224 x 224 x 224 权重。 这种简化是有效的,因为局部空间邻域中的图像和音频之类的结构之间具有高度相关性。
`user_ID`和`movie_ID`从其相应的嵌入矩阵中提取用户和电影嵌入向量。 在该图中,**embedding_1** 代表用户 ID 的嵌入层,而 **embedding_2** 代表电影 ID 的嵌入层。 在 **dot_1**层中执行用户嵌入矢量和电影嵌入矢量的点积,以输出评分(一到五个)。 定义模型的代码如下所示:
`user_ID`和`movie_ID`从其相应的嵌入矩阵中提取用户和电影嵌入向量。 在该图中,`embedding_1`代表用户 ID 的嵌入层,而`embedding_2`代表电影 ID 的嵌入层。 在`dot_1`层中执行用户嵌入矢量和电影嵌入矢量的点积,以输出评分(一到五个)。 定义模型的代码如下所示:
| `@sprintcare` is the worst customer service. | `@__cname__` Hey there! Can you dm us your account's email address? We'll take a look backstage / dn https://t.co/ldfdzrinat. |
| `@__cname__` Y'all lie about your great connection. 5 bars LTE, still won't load something. Smh. | `@__cname__` We apologize for the trouble. Please send us a dm with your email address so we can connect. |
| `@__cname__` Whenever I contact customer support, they tell me I have shortcode enabled on my account, but I have never in the 4 years I've tried https://t.co/0G98RtNxPK. | `@__cname__` Hi there, we are sorry to hear this. Please send us a dm with your email address and phone number so we can look into this for. |
| Actually that's a broken link you sent me and incorrect information https://t.co/V4yfrHR8VI. | `@__cname__` Hi, we are sorry to hear this. Please dm us your contact info and we'll be happy to help. ^ kk |
| Yo `@Ask_Spectrum`, your customer service reps are super nice— but imma start trippin if y'all don't get my service going! | `@__cname__` We apologize for the trouble. Please send us a dm with your email address so we can connect. |
| My picture on `@Ask_Spectrum` pretty much every day. Why should I pay $171 per month? https://t.co/U6ptkQa5Ik. | `@__cname__` Hi, we are sorry to hear this. Please dm us your contact info and we'll be happy to help. ^ kk |
| Somebody from `@VerizonSupport` please help me �������� I'm having the worst luck with your customer service. | `@__cname__` We apologize for the trouble. Please dm us your contact info and phone number so we can assist. |
| `@VerizonSupport` My friend is without internet we need to play video games together please our skills diminish every moment without internetz. | `@__cname__` Hey there! Can you dm us your account's email address? we'll take a look backstage / dn https://t.co/ldfdzrinat |
| `@__cname__` tried to pay a bill for 60 days. No service, rude CS, and several transfers. Look up my equipment # and give me an acct #! | `@__cname__` Hi, we are sorry to hear this. Please dm us your contact info and we'll be happy to help. ^ kk |
| `@__cname__` is the worst ISP I've ever had. | `@__cname__` Hey there! Can you dm us your account's email address? We'll take a look backstage / dn https://t.co/ldfdzrinat |
| `@__cname__` fix your app it won't even open. | `@__cname__` Hi there! We don't have any info on this issue. Please send us a dm with your name and phone number and the full service address. |