One way to think of it is that each point in your data follows your model but with gaussian iid noise shifting them away. The likelihood is then product of gaussians mean shifted and rescaled by variance. Minimize the log-likelihood then becomes reducing the sum of (x-mu)^2 for each point, which is essentially least squares.
My question is how much of the operations in JAX here can be done with reduced precision and can utilize training accelerators i.e. TPUs. I've noticed a lot of research coming out in physics, where everything is simulated in at least double float, being augmented with ML approaches where precision is traded for dynamic range.
A bit off-topic, but are there any Tailscale users here who consistently use it on mobile? How was the speed and battery life? I see that Tailscale offers a lot of nifty features like this that go beyond setting up a VPN, but if the overhead is too big it might not be worth running at all.
Even more than that, the structure of these networks imply that targeting influencers may be ineffective. There's an interview [0] with Duncan Watts discussing this.
The article is a bit vague on these so I'll take a stab:
Random network are generated randomly, i.e. drawn from some distribution. An example would be the Erdos-Renyi (ER) network, where edges are drawn from a Bernoulli distribution between each node in the network. The simplest way to measure random networks is to determine the degree distribution. ER networks hence result to degree distributions that follow the Binomial distribution (Poisson as N->infty, Np->const). Since the edge probabilities are rather simple and drawn independently, one can take the ER network as sort of a "null model" of random networks.
Scale-free networks, on the other hand, do not follow this degree distribution. Rather, they follow a power-law distribution p(k) ~ x^-a. A process to explain this is that the network is generated by preferential attachment: nodes are more likely to be connected to nodes which are highly connected in the first place. Most of the discussion and the following arguments on this are in the article.
If I remember correctly, Mathematica started the notion of having a notebook interface when working with code, Theodore Gray even has the patent for this [0]. Even though the post reads like a longform ad for Mathematica and the Wolfram language, I do agree the notion of code written to present an idea/solution as a story. Sadly, with Jupyter notebooks being free/open source and compatible with other language kernels, there seems to be a decline in the mindshare for Mathematica and its notebook interface among researchers.
[0] Patent US 8407580 B2 - Method and system for presenting input expressions and evaluations of the input expressions on a workspace of a computational system
(https://www.google.com/patents/US8407580)
I'm not sure how different this is with Cython since you can also write C/C++ code with it, and hence use any C/C++ libraries. I say that the benefit offered by the example is more of bringing the Rust ecosystem to Python than solving performance issues.
For the new AMD releases there's none yet, but for Ryzen 7 Phoronix did one comparison for code compiled with -mavx2 and Ryzen did really poorly. The other benchmarks which use more integer math or relies heavily in multithreading puts AMD in an advantage. The post was in 18 May 2017 and AMD may have released some microcode updates that nullifies the results.
People are currently discussing about the legal implications of storing images coming from federated Mastodon instances. A lot of this stems from an influx of users at pawoo.net, a Mastodon instance set up by Pixiv, posting CP and CP-like content. While Pixiv explains that these images are legal in Japan where the servers are hosted, a lot of the instances federated with pawoo have jurisdictions setting such images as illegal, or at least immoral. As response, a lot of instances have defederated from pawoo.
To comment on Apple's "cloud first" strategy, while they do provide local file storage in their APIs, their guideline suggests avoiding creating an option for storing local files since users "expect all of their files to be available on all of their devices." (https://developer.apple.com/ios/human-interface-guidelines/i...)
My rule of thumb is to always have a healthy dose of skepticism regardless of how a study was published, whether in a peer-reviewed journal, preprints, or even a blog post. About bleeding edge research, publishing machine learning methods immediately on arXiv seem to be hot among academics nowadays.
I'm not the author, but I can happily provide some practical applications to his discussion of dynamical systems and chaos.
One of the nice application of the logistic equation is its use as a simple model for population dynamics. In fact, it is in this context that Robert May started the study of the logistic maps [1]. Looking at the logistic map equation, the next value (population) is based on the previous value at the right-hand side. We can provide interpretation to this: with rx(1-x), the r*x-factor tells us that the population should change proportional to the current population as they reproduce. However, the environment may have limited resources and overpopulation may hinder its growth, thus the (1-x)-factor. The bifurcation diagram which the author showed should tell us some things about the population dynamics. At particular growth rates (0<r<1), the population will simply collapse to zero, which makes sense as the growth factor r rather tells that the population should decrease being a fraction of the current population. The arching region between 1 and 3 tells us that there should be a stable population (a fixed-point value) where, from the initial value, the population would always go to. The weird structure we see beyond r>3 shows how the population would oscillate from different values, which we call strange attractor: the population, after a long time, neither goes to a specific value nor increases uncontrollably: the population just goes around at different values.
Another application would be pseudorandom number generators. For example, the most common implementation for quickly creating pseudorandom numbers is the Mersenne twister [2]. In its simplest explanation, the twister looks like a very complicated feedback system, which generates the next value using the previous values. Unlike the logistic map, the twister has a very long period. Plotting it on a bifurcation diagram would cover the entire range of values, which would be ideal as you would want to cover your range of numbers uniformly. You wouldn't want to get only two values as random numbers, unlike at around r=3.3 of the logistic map.
Although you may say that you have weak background in mathematics, it is not impossible to appreciate and use these concepts. You can look for any video of Steven Strogatz on YouTube explaining these topics. His discussions are really accessible and easy to understand for any person.
[1] May, R. M. (1976). Simple mathematical models with very complicated dynamics. Nature, 261(5560), 459-467.
[2] Matsumoto, M., & Nishimura, T. (1998). Mersenne twister: a 623-dimensionally equidistributed uniform pseudo-random number generator. ACM Transactions on Modeling and Computer Simulation (TOMACS), 8(1), 3-30. (But I suggest reading the wiki page for the twister)