vladf

14 ms·

91.
by vladf 6y ago
Alternatively, one could get rid of the memory used by optimizers entirely by switching to vanilla SGD. I haven’t tried this on transformers and maybe that’s what breaks down here but in “classic” supervised settings I’ve found SGD with sch
92.
by vladf 6y ago
I can't wait to see the vector DBs we're going to have 5 years from now when everyone needs to serve their embeddings. It's clearly early but frothy right now. Also check out this similar co I ran into: https://www
93.
by vladf 6y ago
I agree, I was really only proposing PRESS for the "parameter efficiency" part of your comment. It'd be interesting to see some modern takes on IRLS. I think generally this goes against the grain of the Cheap Gradient Princip
94.
by vladf 6y ago
> they do not quite care about parameter efficiency. Google Research is pretty big, I used to think like you did but I think it's mostly b/c DeepMind just hogs all the spotlight. Check out PRESS [0] for example. [0]: https:&#x
95.
Fast sparse matrix loading in Python with Rust(github.com)
6 points by vladf 6y ago|0 comments
96.
by vladf 6y ago
See also my note on reddit refering to the generalization: https://www.reddit.com/r/MachineLearning/comments/hm25mv/com...
97.
Complex Hash Collisions(vladfeinberg.com)
10 points by vladf 6y ago|2 comments
98.
by vladf 6y ago
What about drone-based delivery instead?
99.
by vladf 6y ago
Great comment. I can see how the Bayesian setting uniquely equips you for dealing with non-point testing settings in which reasoning about alternatives to do power, type m, and type s design would be hard. I'd be very curious about two
100.
by vladf 6y ago
No, it's not auto-completing function blocks, just simple expressions that are easy to validate. E.g., let lo = 0; let hi = vec.len(); let mid = lo + (hi will autocomplete to `(hi - lo) / 2` as the second autocomp
101.
by vladf 6y ago
I happen to use slightly less fancy and expensive GPT-2 based autocomplete, and it's amazing. https://tabnine.com
102.
by vladf 6y ago
Horner’s method for polynomial evaluation is used because of numerical stability, not speed.
103.
Solving the General RCT2 Maze Problem with Sympy(vladfeinberg.com)
6 points by vladf 6y ago|0 comments
104.
by vladf 6y ago
This is an interesting idea, with the main non-trivial win really being vectorized GBDT inference. Instead of serially going down a DT, you can convert it to vectorizable GEMM code. By a cursory look at hummingbird/ml/operator_con
105.
by vladf 6y ago
I think "vast majority of sales" is too cynical, though there's undoubtedly some merit to your sentiment in certain cases. I manage a high-output team, and we _love_ Mode. Could we write our own connectors to our BI warehouse
106.
by vladf 7y ago
https://vladfeinberg.com/ I mostly post about stats or programming topics. I only really try to put something up if it's a particularly hot take on a useful topic, like * a reduction from causal to statistical inferenc
107.
by vladf 7y ago
That’s not a really fair way of putting it. Unless you’re suggesting that there’s something deeply magical about our human “aesthetic” intuition why can’t that itself be coded into a set of syntactic indicators (e.g., favor short equations,
108.
by vladf 7y ago
While generally I agree with your conclusion (synthetic data doesn’t have better privacy guarantees, probably will hurt training if you use it naively), I wouldn’t be so pessimistic. At the risk of digressing from TFA, Candes’ knockoffs, fo
109.
Stop Anytime Multiplicative Weights(vladfeinberg.com)
1 points by vladf 7y ago|0 comments
110.
by vladf 7y ago
Re independence: identifying whether or not these physical assumptions covary is not that easy. That's my point: assumptions a, b, c, d could easily have some mutual incompatibility that makes them non-independent. It's an active
111.
by vladf 7y ago
I think that you may have missed my point. First off, why are assumptions independent? Why are you allowed to factor p(a|e) * p(b|e) = p(a,b|e)? Assumptions aren't just independent binary variables -- it's not necessarily true tha
112.
by vladf 7y ago
I don't understand how we can reason with "probabilities of failure" about such fundamental things such as laws of physics. Under the frequentist interpretation of probability, 95% of having a "correct assumption" m
113.
by vladf 7y ago
You use Snowflake for OLTP? Can you comment more on how/why?
114.
by vladf 7y ago
I think there's a really important caveat to OFU here. Namely, we have your higher-level interpretation in the GGP ("great-grandparent post"): > The intuition here is that if your optimism turns out to be correct, you can
115.
The Semaphore Barrier(vladfeinberg.com)
1 points by vladf 7y ago|0 comments
116.
The Sempahore Barrier
1 points by vladf 7y ago|1 comments
117.
Weekly Financial Summaries with Postgres(vladfeinberg.com)
2 points by vladf 7y ago|0 comments
118.
by vladf 7y ago
Interesting -- could you elaborate on your use case here?
119.
by vladf 8y ago
Can you elaborate on why other things are "better" than crossfit? I've found it to be very time-efficient and well-rounded.
120.
by vladf 8y ago
Which websites have you requested deletion from?

More ›