This catch and release pattern is quite common in many wildlife surveys.
For example, when estimating tiger populations, instead of painting a white dot, rangers set up camera traps to take photos. Then they can use stripe patterns as a signature for subsequent re-appearances. Quite an interesting intersection for image AI with statistical counting methods.
The current government came into power riding on a wave of popularity, largely driven by their "IT cells". Now comfortably in power, they are probably dismantling the path, to avoid others who might try to follow.
The government is not just in a stand-off with big-tech companies. It continues to view its own citizens as adversaries with frequent internet shutdowns [1].
Got it. I was looking for input on how generalizable (the ability of weights to change/adapt) when the training labeled data is 100x smaller than the initial pre-training dataset?
Also, I don't understand the need to be so defensive though and the relevance between my employer and my post?
I understand that you do mention the pre-training / transfer learning approach clearly, but isn't it disingenuous to claim that you provide better performance based on (only) 100 labeled examples, when the pre-training dataset (Wikitext-103) actually contains 103M words?
Until this interesting video [1] by Tom Scott, I had never considered sand as a non-renewable resource. But now, it is clear that we now have one more resource that we might run out of.
Proactively thinning selective regions of forests and controlled burns seem to be the route forward to prevent (rather than react to) large-scale fires.
This short video from MinuteEarth, while not going into much detail, does have a lot of good references