Відео

I would never explain Arithmetic Coding, with floats.

I remember you way back on the days of KA. Awesome to see what you're making now. Keep it up!!

bro, you really dropped 1000 search terms in the first 5 minutes and said "it's not important"

THIS VIDEO IS SOOO GOOOOD!

But jpegs do lose information

JPGs and MP3s use lossy encoding.

love this video

Very well done

I know I do too much tech stuff when I saw images and thought about ISO files.

I think it would be nice if in future you put what people need to know beforehand. This was monumentally complex with how much terminology was thrown at the viewer as well as incredibly fast equation transformations. There was pretty much zero edge case examples (the ones you use to establish how it works and build on top of). I find this video pedagogically horrendous for an average viewer and there was no attempt whatsoever to dissuade anyone and inform them that it was not for them. Many other comments in the comment section here reflect that - the people who choose to watch this are likely a little familiar with computer science and math to what I'd argue high school degree and maybe a little more. I would expect the level of explanation to be as such. Huffman encoding is fairly straight forward to explain. This felt like you'd needs an entire degree just to grasp it when in actuality it's not that complex when you go and experiment to try and understand it on your own. The entire section in the middle that explains information density was, for me, impossible to understand. Part of it is the excessive use of formulas and their transformations rather than raw data examples. This felt like a thesis paper and not an educational video. Pedagogy is a skill. And the best way to improve at it is to practice and test. Did you even show this to anyone who isn't as knowledgeable on the subject matter? They would point out dozens of points where they got completely lost. That is where you ask questions and figure out a way to explain the concept you are talking about in more clear terms. Maybe it's he words being used. Maybe it's the pacing. Maybe it's the visuals. Maybe it's all of the above. Or maybe it's something else entirely. But you will never have the intuition until you develop it. And to me, this video shows a very deep lack of said intuition. I love teaching. I love learning. I think you have the skills and knowledge to remake this into something amazing. But as it is now, it's extremely lackluster. This almost makes me want to learn these algorithms deeper just so I could put out a better video on the subject since this, to my instinct, turns people away from using it rather the encouraging. I'll personally take more time to learn this since I find lossless data compression potentially very valuable in what I do for a living. I'm happy you showcased these even exist. And I hope this comment encourages you to make more stuff in the future rather than stop completely. This is made as constructive criticism. Take your time and show someone your work before you publish it. Preferably a few people with different levels of understanding so you know how approachable it is and which group to cater to. The production was very good. So good luck and I hope you inspired more people to pursue these types of algorithms.

Huff man + +Latent Encoding then all you need is a magic box to unfuck the tree on the decoding side

Patents are fundamentally terrible

DjVu employs arithmetic coding, and DjVuLibre includes the lossless compression utility bzz. I don't think it's received much attention to make it fast, though; zstd is likely a more practical choice today.

🥲 ahora entendi por que nadie los usa

2:40 edit : nvm he adress it later lol i think huffmann lost because it also store number with 0 in front as posible number, where as the conversion method store 00123 the same as 123

never underestimate the human ability to claim ownership over anything they possibly can

Nice video! Just one suggestion: don't let the video black-out for so long while you're talking. I was confused thinking something was wrong with the image, because you were still talking regardless hahaha

Huffman also accepts that "01" can be multiple characters like "you"? Also H(X) is sometimes called Shannon's measurement of Information (SMI)

I don't think anything can beat middle out compression....

I recommend anybody using this method to also implement try to implement more layers, compressing with algorithms and compressing neighbour pattern delimiters with some form of character. This will greatly improve the compression and complexity of your data! <Bull:<Bull_Chunk:cGQ=SBbScxHn(D)Ib=GdgsvlsC(I)JZVxJMW|15:>:>

I got lost in the first 10 minutes haha, all the math killed me

is there any legal risk to bake the ideas pointed at in this video into a commercial product ?

Someone should patent patent-trolling. "Method of using fraudulent patents as money extortion and corporate sabotage"

it looks sooo similar to binary search, interesting

How 8:45 do you store all messages and allow easy lookup??? There are a few REALLY one excellent, and many adaptive ones, approach here 8:45

Patents stifle innovation

3 minutes into the video and i knew.....my brain is smooth, very smooth

Another proof that patents should be abolished

“i reject your rejection”

zstd uses tANS

This video is a bit too complicated for me😭

Interesting topic, good explanation, nice animations, and a special thanks to Frédéric Chopin. UA-cam, is as bad as the US patent office, and doesn't recognise all authors of the music. YT didn't recognise Satie either.

Es gibt nur eine Person die ins solch ein Straflager lebenslänglich reingesteckt gehört …

To be honest this should've been an hour long video, it really went over my head

It is an interesting video, but you go too fast for this to be understand es without pausing the video. I recommend going a little bit slower

Interesting topic but terribly explained, I've written a toy ANS implementation and I couldn't follow what this guy was trying to say half the time, good visuals though.

Intellectual property is the attempt to own somebody else's capacity to reproduce work. Ownership over brains or body of somebody else is abolished and so should patents and the whole concept of intellectual property as well be.

Imagine a world with no patent laywers

Terrible video.

i wish i could intergrate this into my audio format but i cannot due to its way of storing data is by string and cannot be binary in any way

this video was very difficult to digest, I don't have a math background, but I also think that the video should focus more on what something means rather then what it is

I first heared of this about 30 years ago and i was so excited that i wrote it in C. I can acknowledge, it's compressing better than huffmann, but twice? Maybe with very specific data. But it was quite slow. Most probably because i wrote the adaptive version. It should be written in assembly, because it has some very special needs that no higher programming language provides. And writing a workaround to get that missing feature in C is a pain in the afternoon and also slow. Maybe using libgnump will be helpfull, if you don't want to use assembly - libgnump is written in assembly, as far as i know.

I stumbled upon ANS for the first time in this video, so thanks, but I didn't understanding it at all. I had to watch Stanford's EE 274 on the subject, where it seemed ... approachable, at least. Not that a video of this length was going to fully explain such fairly advanced topics, the pace is quite fast, but it was an excellent introduction to the subject, and it motivated me for further research, so goal achieved I guess. Also the effort in making it really shows.

I almost understood this but there was a damn plane kept flying over my head

if the akinator was a compression algorithm lmao

When auto-tune was invented all the talentless acts became instant celebrities.

What about markov compression? 😮

This was a heroic effort at an explanation video, but it should really have been a 10 video series at least.

Erm, pretty sure I’m watching the video encoded with arithmetic coding. It’s the basis of h.265 and h.264 video codecs, s as long with their still image profiles. H.264 baseline profile is closer to Huffman, but the main profiles expect to use context adaptive binary arithmetic coding.