pcovington

6 ms·

1.
by pcovington 10y ago
Your intuition is correct - there are other ways to capture the non-stationary nature of this particular problem. We thought that the example age approach is neat because it is a general technique for removing bias inherent to any machine l
2.
by pcovington 10y ago
Thanks! This model handles new users gracefully because it can fallback to demographic/geographic priors and gradually specialize as the user watches videos. New items are difficult because of the fixed output vocabulary and batch trai
3.
by pcovington 10y ago
Figure 3 illustrates that the variable sized watch history is combined with an average operation. This is partially why the embeddings need to be so large - in order to retain information after averaging, you need lots of dimensions to spre
4.
by pcovington 10y ago
This is a very natural avenue and an active area of research at Google/Deep Mind. Stay tuned...
5.
by pcovington 10y ago
word2vec did inspire earlier iterations of the model, but the key insight is that embeddings are learned jointly with all other model parameters. There is no separate source of embeddings. This way, embeddings are specialized for the the sp
6.
by pcovington 10y ago
There are many close collaborations between product and research, as well as direct exchanges between different product areas. Close collaboration is key because those working directly on the product understand best the data, serving system
7.
by pcovington 10y ago
YouTube has used machine learning in recommendations for many years. We have struggled with interpretability, both while debugging mistakes made by the system and exposing plausible "reasons" to users. There was a fascinating disc
8.
by pcovington 10y ago
The video embeddings in the paper are learned purely based on observing what users co-watch in sessions. In this sense, they can be thought of as latent factors in more traditional collaborative filtering approaches. When we inspect them, n
9.
by pcovington 10y ago
Yes, it's a reasonable proxy. It was challenging to set up similar experiments with the old system because it was trained to approximate a different "surrogate" problem. We've also found that recommendation systems are v
10.
by pcovington 10y ago
Author here - happy to answer questions about the techniques in the paper. We're super excited to finally share this work externally. Feedback about YouTube recommendations in general also welcome.