eutectic

16 ms·

151.
by eutectic 5y ago
I meant more the caramelization - if you're going to nuke all the aromatics, why not just use sugar?
152.
by eutectic 5y ago
Sounds like a waste of good honey.
153.
by eutectic 5y ago
I guess this wouldn't work as a cure for HIV if it only kills actively replicating virus. Still amazing if it works in practice.
154.
by eutectic 5y ago
Not sure if this was intended as a compliment or an insult :)
155.
by eutectic 5y ago
For that drastic a difference you need to play about 30 games on average to be 99% sure that p_win>0.02 in the Bayesian model. Agreed that win rates are not the only data available, and that players probably should have noticed the diffe
156.
by eutectic 5y ago
Assuming your win rate goes from 2% -> 4%, you need around 200 samples to be 90% confident that something is up, 270 to be 95% confident, and 450 to be 99% confident, if you start with a Beta(2, 98) prior. Obviously results will change s
157.
by eutectic 5y ago
Maybe someone put it back up inbetween?
158.
by eutectic 5y ago
No, chaining is presented as an alternative to Cuckoo hashing.
159.
by eutectic 5y ago
Neat!
160.
by eutectic 5y ago
But does that really require an absolute time reference?
161.
by eutectic 5y ago
An example of a bad case for random exploration would be a narrow ridge where you die if you fall off and you only receive reward if you get to the end. So it's a worst case result with respect to the MDP, but expected time/high p
162.
by eutectic 5y ago
It's proven that if you want polynomial sample complexity in the size of the state space you need directed exploration. The algorithms 'flail' because they are initialized with random policies.
163.
by eutectic 5y ago
I wonder if a transformer with 1 recurrent layer (or a transformer Deep Equilibrium Model) could work well. Transformers are almost like a physics simulation in that they sum vector-valued interactions which depend on distance in some spac
164.
by eutectic 5y ago
What was it?
165.
by eutectic 5y ago
I don't really see the dichotomy. Surely the features you need to learn depend on the task? Or do you mean running linear / shallow rl on top of unsupervised learning?
166.
by eutectic 5y ago
Seems like it might be worth taking some lessons from SAT solvers. e.g. backjumping and clause learning.
167.
by eutectic 5y ago
You have to integrate which means discretizing time. And you have to deal with collision and solve constraints. There are no exact solutions in physics simulation.
168.
by eutectic 5y ago
They are a way of training generative models similar to GANs or autoencoders. My understanding is that if you train an autoencoder with a gaussian likelihood then you will tend to get fuzzy samples, but using an iterative process where each
169.
by eutectic 5y ago
Also a good episode of the podcast '99% Invisible' on the topic (although not specifically in the context of COVID-19).
170.
by eutectic 5y ago
Physics might be simple, but it's not easy. Especially if you want to make predictions for complex systems.
171.
by eutectic 5y ago
What's the mechanism? Sounds unlikely to do much.
172.
by eutectic 5y ago
This is why it's a good idea to have a real integer type.
173.
by eutectic 5y ago
I assumed it was an intentional pun.
174.
by eutectic 5y ago
I would say that the space shuttle is the perfect example of the risks of over-engineering.
175.
by eutectic 5y ago
OK, agreed.
176.
by eutectic 5y ago
I wouldn't say power series are the 'reason' they are related. The more intuitive reason for Euler's identity is the combination of e^x being its own derivative, and the relationship between complex multiplication and ro
177.
by eutectic 5y ago
The normal distribution is special in being closed under linear transformation.
178.
by eutectic 5y ago
I would say it's more a commentary on the fact the the gradient is effectively based on L2 distance in parameter space, and so can be a bad/inefficent direction to move in even if you have access to the full gradient. Hence the mo
179.
by eutectic 5y ago
I wonder if its feasible to identify the optimal (or at least a near-optimal) strategy if you treat it as a game of pure logic, where the players agree beforehand which attributes from a limited vocabulary apply to which words.
180.
by eutectic 5y ago
The model trained on TSIT gives nicer results.

More ›