摘要:
Our results show that DLinear outperforms existing complex Transformer-based models in most cases by a large margin. In particular, for the Exchange-R 阅读全文
posted @ 2023-10-08 08:03
emanlee
阅读(77)
评论(0)
推荐(0)
摘要:
While some deep learning models discover dependencies in decomposed time series, they are not good at capturing local dynamics and long-term dependenc 阅读全文
posted @ 2023-10-08 08:03
emanlee
阅读(24)
评论(0)
推荐(0)
摘要:
https://blog.csdn.net/weixin_44790306/article/details/124064177 摘要本周一是对Informer论文的阅读,其关注的问题依然是长时间序列预测问题。也是从self-attention 机制的缺陷出发,做了一些优化于改进工作,像ProbSpa 阅读全文
posted @ 2023-10-08 08:02
emanlee
阅读(1657)
评论(0)
推荐(0)
摘要:
Informer: 一个基于Transformer的效率优化的长时间序列预测模型 Informer创新点介绍 ProbSparse self-attention self-attention蒸馏机制 一步Decoder 实验结果 总结 Informer: Beyond Efficient Trans 阅读全文
posted @ 2023-10-08 08:02
emanlee
阅读(900)
评论(0)
推荐(0)
摘要:
参数量方法一:pytorch自带方法,计算模型参数总量 参数量方法二: summary的使用:来自于torchinfo第三方库 参数量方法三: summary的使用:来自于torchsummary第三方库 计算量方法一:thop的使用,输出计算量FLOPs和参数量parameter我们通常要通过计算 阅读全文
posted @ 2023-10-08 08:01
emanlee
阅读(8446)
评论(0)
推荐(0)
摘要:
https://www.zhihu.com/question/547837668 https://www.zhihu.com/question/24021704 https://www.zhihu.com/question/24021704/answer/2245867156 傅立叶原理表明:任何连 阅读全文
posted @ 2023-10-08 08:01
emanlee
阅读(74)
评论(0)
推荐(0)
摘要:
https://blog.csdn.net/weixin_49967436/article/details/121736079 3.1.Multi-Head Attention(图2-红色圆圈部分,图3-红色长方体) 3.2.Self-sttention Distilling(图2-蓝色圆圈部分,图 阅读全文
posted @ 2023-10-08 07:59
emanlee
阅读(85)
评论(0)
推荐(0)
该文被密码保护。 阅读全文
posted @ 2023-10-08 07:53
emanlee
阅读(0)
评论(0)
推荐(0)
摘要:
pytorch permute permute(dims)将tensor的维度换位。参数:参数是一系列的整数,代表原来张量的维度。比如三维就有0,1,2这些dimension。例: import torch import numpy as np a=np.array([[[1,2,3],[4,5,6 阅读全文
posted @ 2023-10-08 07:53
emanlee
阅读(35)
评论(0)
推荐(0)
摘要:
https://www.bilibili.com/video/BV1m14y1a74s/?spm_id_from=333.337.search-card.all.click&vd_source=6292df769fba3b00eb2ff1859b99d79e 阅读全文
posted @ 2023-10-08 07:53
emanlee
阅读(117)
评论(0)
推荐(0)

浙公网安备 33010602011771号