Beginning Anomaly Detection Using

 What Is a Temporal Convolutional Network?

Temporal convolutional networks refer to a family of architectures that incorporate 

one-dimensional convolutional layers. More specifically, these convolutions are 

causal, 

meaning no information from the future is leaked into the past. In other words, the 

model only processes information going forward in time. One of the problems with 

recurrent neural networks in the context of language translation is that it reads sentences 

from left to right in time, leading it to mistranslate in some cases where the order of the 

sentence is switched around to create emphasis. To solve this, bi-directional encoders 

were used, but this meant future information would be considered in the present. 

Temporal convolutional networks don’t have this problem because they don’t rely on 

information from previous time steps, unlike recurrent neural networks, thanks to their 

casuality. Additionally, TCNs can map an input sequence of any length to an output 

sequence with the same length, just as a recurrent neural network (RNN) can do.

---

258

Basically, temporal convolutional networks seem to be a great alternative to RNNs. 

These are the advantages of TCNs, specifically considering RNNs in general:

•