SimplyUnknown

For a longer read, I can highly recommend the book "The Man from the Future" by Ananyo Bhattacharya. It's quite an easy read and extensively discusses the life and discoveries by John von Neumann in chronological order.

Indeed, EPEX spot prices are between €0.30 and €0.39 today including taxes where I live. I've seen them go up to €1.10 during peak demand.

Huh? That was University of Utah/Brigham Young University right. That is, if you're referring to Pons and Fleischman.

Obviously not, but there is larger potential than that. A microphone in a watch or phone can pick up the signal and relay the trigger to another server via for example MQTT. Then, you can have homeassistant act upon this signal.

The flow is then Open door --> Signal triggers --> Signal caught by microphone --> Microphone device relays to server --> Server turns on smart lighting in that room.

There is a lot more that you can do, but this is just an example. The microphone to pick it up is a limitations, but the gain is:

- reduced power usage as the signal generators are not power-operated

- less waste as no empty batteries need to be discarded (if non-rechargeable batteries are used)

- improved security because the signal generators are not connected to the internet, reducing the attack surface.

I'm still not sure I get it. I think it is

1. Put the BWT string in the right-most empty column

2. Sort the rows of the matrix such that the strings read along the columns of the matrix are in lexicographical order starting from the top-row????

3. Repeat step 1 and 2 until matrix is full

4. Extract the row of the matrix that has the end-delimiter in the final column

It's the "sort matrix" step that seems under-explained to me.

I have the feeling that B-splines would be a good solution for this problem. Given that they have a continuous zeroth (i.e., the function is continuous), first, and second derivative, the motion will always be smooth and there will be no kinks. However, maybe it's moving the problem because now you must tune the coefficients of the B-spline instead of damping parameters (even though a direct mapping between these must exist but this mapping may not be trivial).

Multiple reasons, while technically better and more benign compression artifacts, it is computationally more expensive, limited quality improvements, encumbered by patents, poor Metadata format, poor colorspace support... In the end, the benefits aren't great enough compared to jpeg to change the default format

I really like einops. This works for numpy, pytorch and keras/tensorflow and has easy named transpose, repeat, and eimsum operations.

Full paper link for the interested: https://ehdijrb3629whdb.tiiny.site

In medical imaging, data are often acquired using anisotropic resolution. So a pixel (or voxel in 3D) can be an averaged signal sample originating from 2mm of tissue in one direction and 0.9mm in another direction.

Conda indeed is slow. However, mamba is a drop in replacement for Conda and uses a way faster solver, which makes it a lot more palatable.

Not quite what you are looking but if you're interested in Operation Market Garden: for the Dutch maps there is https://www.topotijdreis.nl, which gives you historical maps with a year slider. This can at least help one visualize how cities, villages, and topography at through the years.

CGP Grey also made an excellent video about it, which he dubbed the NaPoVoInterCo: https://www.youtube.com/watch?v=tUX-frlNBJY

But Chinese (or mandarin) is not a context-free grammar whereas I believe that encoding a language on a turing machine implies a context-free grammar so this example doesn't hold.

I think maybe it's poorly phrased. As far as I can tell, their linear regression example for eq. 2 has an unique solution, but I think they state I that when optimizing for cosine similarity you can find non-unique solutions. But I haven't read in detail.

Then again, you could argue whether that is a problem when considering very high dimensional embeddings. Their conclusions seem to point in that direction but I would not agree on that.

The thing is you use synthetic data to when it is difficult to obtain real data. For example, in medical imaging, it is very expensive to collect MRI scans to build a large dataset. Not to mention the potential privacy issues and obtaining informed consent to publish the dataset. Synthetic datasets can help here to, for example, pretrain your model and fine tune on real data afterwards. I'm then assuming that collecting tabular data can face similar issues which prevent building large datasets.

Surely not, but if you practicing to get better at programming using AoC, LLMs are unlikely to help you.

Maybe better programmer not so much, but I'd say the spirit of AoC is to crack a puzzle and translate the solution into code. Solving this using LLM will not help you getting better at solving puzzles, solving puzzles, or grasp the main concepts of a programming language. In that sense, ChatGPT is not a pedagogical asset in this particular instance.

Obviously you can, but in the grand scheme of things people should share more details about their method so people can improve on it in the future, no?

I mean, you can't just share the weights of the model and call it a day, right? You have to share details on what and why you are doing. You must communicate this somehow. In theory, you might be able to do this in a github readme, but a paper-style document on arxiv is nicely suited for this.

I guessed as much, but I wasn't able to confirm. Thanks for the search.

So the solution would be to continue emitting greenhouse gases and ensure it will also get worse in 30 or 50 years?

> The best time to plant a tree was 20 years ago, the second best time is now.

Said to be a Chinese proverb, but I haven't extensively verified.

The only viable solution right now to combat global warming is to immediately stop emitting greenhouse gases, even if there is a lag effect.

First of all, great blog post with great examples. Reminds me of distill.pub used to be.

Second, the article correctly states that typically L2 weight decay is used, leading to a lot of weights with small magnitudes. For models that generalize better, would it then be better to always use L1 weight decay to promote sparsity in combination with longer training?

I wonder whether deep learning models that only use sparse fourier features rather than dense linear layers would work better...

Not all of these are mutually exclusive, but there is also a fourth option:

- ADHD was severely underdiagnosed in the past and got more accurately diagnosed in recent years

Most Europeans also have a lower cost of living. So maybe the correct metric is average purchasing power, which goes down if the wages go down (that said, maybe only when viewed for lower or middle class incomes. It could be that rich become as much richer as the poor are becoming poorer, yielding zero change in purchasing power on average)

> "Google is inserting AI into art history."

The phrasing of the subtitle implies that active interference or malicious intent is occurring while I don't believe the article shows this is the case.

The first edition of the C programming language was released in 1978, the second version in 1988. The first time something on Unicode was mentioned was also 1988, and the consortium was founded in 1991. UTF-8 was proposed in 1992.

Simply put, the book doesn't deal with alternatives to ASCII because there hardly were alternative text encodings at the time of writing in the western world, which is clearly the focus of an English book written by a Canadian and American.

Moreover, the point of the book was to propose the C programming language and showcase how the language works. It's not a book on best practices or how to use C in the real world today. There are other resources for that.

How so? Supply goes down while demand is the same --> Price goes up --> Profit margin for the farmer increases --> Profits of mustard > profits of rapeseed --> Farmer produces mustard instead of rapeseed. This is exactly the consequence of supply and demand.

The only noteworthy fact is that we got in this situation because it was not a completely free market but a regulated market due to subsidies for rapeseed, artificially increasing their profit margin.

If they died from a virus, why wouldn't they have fossilized like the earlier or later perished cats? It is then more likely that they haven't died there (e.g., migrated away) or died under circumstances that they died under conditions that they couldn't have fossilized (e.g., volcanic eruption).

It's definitely not perfect, but I'd argue that it is better than a "blurry front paste". For example, it doesn't place breasts on the back. Also, for the polo shirt example the model is doing significantly better than copying the v-line of the front of the shirt and shows a round collar. I think it is pretty impressive and one could argue the model has somewhat learned what a human looks like from the back given the clothing of the front, even though it is quite blurry.