highd

... so how do API users enable 1hr caching? I haven't found a setting anywhere.

I'm fairly confident that the new reddit React implementation can be improved in performance by a factor of 3x to 10x. I would be interested to hear others who have good reason to explain why not. I can certainly imagine React-like systems that are capable of statically determining DOM influence sufficient to make comment-collapsing negligible.

How are you doing your train/test split?

If you think identifying a hypothetical back door in something as complex as a router is not a difficult process you are simply uninformed.

More like classic Bayes. Taleb's pop-sci came a few decades later.

Do you have any details on the experiment procedures? E.g. hardware, training time, loss curves? It is difficult to confidently reproduce research without at least some of these details.

If you released any of the software you are using to do this I would find it extremely interesting!

You can do Q-Learning with a transformer. You simply define the state space as the observation sequence. This is in fact natural to do in partially observed settings. So your distinction does not make sense.

TFP cites decision transformers. Just using a transformer does not bypass the credit assignment problem. Transformers are an architecture for solving sequence modeling problems, e.g. the credit assignment problem as arises in RL. There have been many other such architectures.

The hardness of the credit assignment problem is a statement about data sparsity. Architecture choices do not "bypass" it.

Can you provide some technical details on what you do? Do you divide the number of channels on each real layered network by 4? I don't see anything describing this in the paper.

Golem is solving exactly this problem.

It's actually using just one day's data (one point) to predict the next day's data, let alone predicting 5 days.

Most of the research in compressed sensing type applications now is focused on more sophisticated prior distributions than sparse/Laplacian. These more general bayesian approaches provide significant performance improvements over LASSO etc.

Not to mention that it appears they're comparing against networks of the same architecture. If you build your quaternion components with with same number types as your reals you effectively have 4 times the number of parameters, which could be most of the benefit. They should also benchmark against similar architectures with equivalent parameter counts.