efavdb

10 ms·

151.
Three lessons from safety stock theory(varietyiq.com)
1 points by efavdb 1y ago|0 comments
152.
by efavdb 1y ago
If this attitude prevails I would think the value of degrees will quickly diminish.
153.
by efavdb 1y ago
Suggests an opportunity for hybrids, where the diffusion model might be responsible for large scale structure of response and the next token model for filling in details. Sort of like a multi scale model in dynamics simulations.
154.
A look at some of the creative ways companies try to dodge high tariffs(npr.org)
2 points by efavdb 1y ago|0 comments
155.
by efavdb 2y ago
Pretty interesting. I’m surprised that the throughput drops with traffic, especially all the way to zero in the first plot. Definitely frustrating to drive through these and great point that it’s bad for efficiency.
156.
by efavdb 2y ago
Example?
157.
by efavdb 2y ago
Thanks watched one and it was neat! Q: can anyone tell me what these solo people do when they need to sleep and it’s too deep to anchor?
158.
by efavdb 2y ago
awesome insight. i've heard this referred to as the "site merchandising" problem. There are some products that are there to attract people / give them a choice, but they don't sell themselves.
159.
by efavdb 2y ago
In fact, there is demand for colorful products. However, the way businesses measure demand today is through the aggregate unit demand. In effect, you get the lowest common denominator products rising to the top, and people with specific p
160.
Garment Color and Lighting(varietyiq.com)
2 points by efavdb 2y ago|0 comments
161.
by efavdb 2y ago
I mostly had to teach throughout my PhD. Curious if funding of that sort is also at risk or if it comes out of tuition from undergrads.
162.
by efavdb 2y ago
I personally don’t think much of the maximum entropy principle. If you look at the axioms that inform it, they don’t really seem obviously correct. Further, the usual qualitative argument is only right in a certain lens: namely they say
163.
by efavdb 2y ago
Supporting anecdata: I was interested to see that chatgpt doesn’t know how to use one of my (small, not too popular) open source python packages — despite having blog posts and documentation on them, all from more than five years back.
164.
by efavdb 2y ago
Noob question: It sounds like the statement is “we learned it’s cheap to copy an already built model” from its outputs, but is it still expensive to train a new (better) base model? If so, is this mostly a concern because there’s little mo
165.
by efavdb 2y ago
Curious if this “just enough to remind the body you still need something” relates to the spaced repetition idea for learning / memorizing facts.
166.
Where to park your car [pdf](sites.santafe.edu)
1 points by efavdb 2y ago|0 comments
167.
Mathematics of Parking Tickets(efavdb.com)
4 points by efavdb 3y ago|0 comments
168.
by efavdb 3y ago
Thank you! I’ll definitely check this out.
169.
by efavdb 3y ago
Yeah, backprop gives efficient gradient calcs to train at a given net architecture. To find the best architecture though people try various other methods, eg random search or Gaussian processes, which don’t evaluate derivatives wrt archite
170.
by efavdb 3y ago
Yes sometimes it’s hard to measure a derivative. Eg when doing hyperparameter tuning in ML, you can read out a metric at a given choice of parameters, but it’s generally not easy to get a gradient. Shameless plug: I happen to have recentl
171.
Inchwormrf – a Python package for finding roots of 1-d functions(github.com)
3 points by efavdb 4y ago|0 comments
172.
Spaced repetition can allow for infinite recall(efavdb.com)
281 points by efavdb 4y ago|169 comments
173.
by efavdb 4y ago
I happened to write a paper on just this issue (shameless plug)! You're right: With a sphere, the metabolic consumption grows like R^3 but nutrient capture goes like the surface area, which scales like R^2. The result is a max radius
174.
Counting the number of ways to make change for a trillion dollars(efavdb.com)
9 points by efavdb 5y ago|0 comments
175.
by efavdb 5y ago
Shameless plug: I previously wrote up a blog post on how to use an unsupervised feature selection analog of PCA to avoid many of the issues you point out here, and an associated python package to carry it out ("linselect", which
176.
Generalized Dollar Cost Averaging(efavdb.com)
4 points by efavdb 6y ago|1 comments
177.
by efavdb 6y ago
I’d love to see a historical trace of some reasonable political bias score quantified via NLP, eg of the nytimes over the last few decades.
178.
Physics-based proof of the duality theorem for linear programs(efavdb.com)
1 points by efavdb 6y ago|0 comments
179.
by efavdb 6y ago
Agree. Not always easy to be rewarded within a company. However, others will notice you do good work (assuming you can make it visible) and they’ll be a great resource when seeking next positions.
180.
by efavdb 6y ago
A related point is that we could improve how good work is bubbled up. Another poster pointed out Allison’s work that led to the Nobel prize, and he claimed it was appreciated by just one group and that helped it get broader notice. How ma

More ›