This post might give some answers [1]. FL is a machine-learning framework where models can be trained while keeping users' data on their device rather than sending it to a server.
I knew about work in anomaly detection for state of health and state of charge in batteries, where you can somewhat model the physical effects (in a data driven manner). However, this description of the problem made me think that meta-learning might be suitable for the problem you’re describing. I’ve only seen it applied in computer vision though (and more recently in speech).
When I'm having a hard time writing a document I usually do the following:
- Go back to basics: Review what an introduction, body, and conclusion should have (and what they shouldn't). This of course will depend on the kind of document, since the structure and content of a research paper, a proposal, or an essay will be different.
- Make an outline: Lay out the different ideas in sentences. At some point, you will have all you need and you will just need to connect the sentences to create paragraphs and then sections. This specifically helps me to write concisely, as my document is a flowing outline.
- Revisit: Finally, if I have the time I like taking a break and revisit (even if it is the next morning). This gives me the time to come back with fresh eyes and spot the parts that do not flow, need more detail, or are redundant.
I've noticed that if I struggle creating the outline, it's because I don't understand well yet what I am trying to write.
I've been using tldr [1] instead of man pages lately to get started with a command (or to remind myself how to use one). I've learned a lot just by reading the examples shown, and then read the man pages if I am missing something.
"Machine Learning: a Probabilistic Perspective" is more an encyclopedia of algorithms I would say, and it has lots of typos. I personally would not recommend it (except for the amount of algorithms that it covers, many of which are usually not found in other books).
I agree that saying that he used calculus is a bit of a stretch. However, calculus is essential for all the gradient-based methods (or algorithms) in optimization. Moreover, if you want to analyze these algorithms, you have to know about convex analysis.
I think what the author _meant_ was that he used the intuition derived from calculus to tackle the problem he was framing as a mathematical optimization problem.
Also, you're right, he missed that the global maximum might not be unique in a non-concave function.
Some time ago I looked into this topic, and in my opinion the blog post misses two points: (1) Conversion to gray scale won't be good because the scale is not linear, which is important for printing, and (2) In Jet (and also, Turbo), high and low values of a measurement don't map well to colors. This is well described in the paper introducing cividis [1] (check out [2] for code).
The zipper of a down jacket I bought from them would get stuck. They repaired it twice for me, and the third time I took it for repair, they gave me a new one. This was 4 years after I bought it, no receipt needed. I would assume that the old jacket went into their recycling program.
The new jacket came with a sturdier zipper, so it also showed that they acknowledged the design flaw. I've been using this jacket for the last 5 years, and it is still holding well (and never again had a problem with the zipper).
There's certainly interest on this topic from the neuroscience community. At least for mice vocalizations, the first (unsolved) problem is to find a dictionary of possible vocalizations. This is a highly non-trivial problem, mainly because it depends on having a good similarity between (noisy) vocalizations, but also because we don't have a ground truth or gold standard.
The vocalizations vary between strains, and some of them are more less clean than others in their spectrogram representations. Different strains vocalize in different frequency ranges.
Assuming that you've found a dictionary, then you'd have to learn how to map dictionary elements to behavior. Behavior labeling is done by human annotators that spend hundreds (or more) hours looking at mice behavior and learning how to identify different these. This, by nature, is a noisy process as well (and possibly biased).
Given the difficulties, the problem itself is very useful for people in neuroscience because mice only vocalize in social situations, so they see it as a window into studying social behavior of, for example, mice with autistic behavior.