Love to see noise cameras all over the world.
Even more, similar to air pollution maps, it would be great to have noise pollution map, at least over cities.
444 karma · joined August 14, 2019
https://github.com/albumentations-team/AlbumentationsX
Stars help demonstrate real-world adoption when we apply for grants that fund open-source maintenance, testing, documentation, security improvements, and new features.
Even more, similar to air pollution maps, it would be great to have noise pollution map, at least over cities.
There is a good fiction book in Russian on the topic: https://ru.wikipedia.org/wiki/%D0%A1%D0%B2%D0%B5%D1%82_%D0%B...
Plot: After you die, you get to another place, call it heaven.
Every time someone who is alive remembers about you, you get a coin.
Coins are used in this afterlife community for food, entertainment, labor, etc
You cannot steal them, although can give as gift to another afterlife person.
Of course, in the beginning you are rich, people miss you. Then income drops...
When you are out of coins - you die second time, most likely forever.
Well written, good plot, etc.
But! In this after world Hitler and cavemen whose body is part of the exhibition in natural sciences museum live better and longer than any "normal" and "ordinary" person, especially in the long time perspective.
P.S. You still can do many things in this world to make people remember you for a long time after you are gone.
With respect to more mature research fields the entry point to ML is much lower.
Hence I always recommended people to have major in Physics, Chemistry, Biology etc but look for projects in these fields that could benefit in ML (I have a number of them about Physics).
So that argument was not novel.
But the fact that pure ML PhDs will have significantly lower multidisciplinary knowledge is a good one. It could be compensated by the fact that ML is growing fast and all kinds of people join the ride, but still.
Right now Google, YouTube, StackOverflow, PornHub, Netflix perform search in the database.
But imagine, that you ask for something on Netflix and a whole movie will be generated for you.
We can see glimpses of this in ChatGPT vs Google, and I hope that this direction will grow to the mature phase.
On the other I use free services like Google, YouTube, Facebook all the time.
They make money selling our data, on the other, the fact that it is free for everyone is the democratization of the access to information. I suspect that even a subscription of $1 / month will push away more than half of the users for these services.
I prefer to live in the world where people have free access to search engines and large social networks.
Allows searching for generated imagery at Ternaus.com using text or image queries.
Python code for the bot is included in the text.
One can use an image and modify color, style, or content
URLs to 10M images.
I need to create a thumbnail for each of them.
I.e. I need to download each image, resize and save in the webp format to the S3.
Right now I am doing this locally and it takes forever, as my images are in high resolution their total size is a few Tb.
Is there a tutorial that will describe how to do it? I was thinking about using AWS Lambda + SQS, but if DStack is an easier solution, I would be happy to check it out.
Computer games is a dopamine that you get, but did not earn in the real life.
I played computer games since 1992, if I spent all this time on writing code, it would be better.
But when you are stressed out or in pain, it is better to play computer games, rather look at alcohol.
P.S. When Witcher 4, Stalker 2, or Fallout 5 will be released I will play again :) P.P.S. This story is a great example of why NFTs in the game industry has some value.
Everything stopped working. His changes were reverted back.
But within a month all code base was covered with tests.
I am core developer of https://albumentations.ai/
We have web site, documentation, tutorials, blog posts.
It is not perfect, and could be improved but people use the library.
Metrics: 10k stars at GitHub, 11k downloads per day https://pypistats.org/packages/albumentations
We have 100s of bug reports / feature requests and someone needs to go through this. Even when someone picks a feature, writes code and creates pull request, core team member needs go through it, and typically there are a few iterations before the PR is merged.
In our case we would like to hire full time engineer to do all this stuff. But it is money.
=> donations may help more than advice from the blog post, although I agree with every word in it.
Five new transforms:
[1] RandomCropFromBorders - Crops image based on indents from image borders. [2] BBoxSafeRandomCrop - Crops image without loss of boxes. [3] Spatter - Simulates corruption which can occlude a lens in the form of rain or mud. [4] Defocus - Imitates lens defocusing. [5] ZoomBlur - Imitates lens blur on zooming.
Numerous improvements and bugfixes including added support for bounding boxes for GridDistortion, ElasticTransform, and OpticalDistortion.
Here is the summary.
Good books on the topic:
- Blue Ocean strategy: https://www.goodreads.com/book/show/23627395-blue-ocean-stra...
- Good Strategy Bad Strategy: The Difference and Why It Matters: https://www.goodreads.com/book/show/11721966-good-strategy-b...
I would say, that for good strategy you need:
[1] Understand what is happening in the market, why competitors are doing well / not doing well.
[2] Guiding policy. How to address company's issues from 1. or how to differentiate from the competition.
[3] A set of coherent actions to implement [2].
Hypothesis about future are important, but it is only a small part of the strategic thinking.
Let me quote myself from https://www.toolbox.com/tech/big-data/articles/lyft-on-why-r...
> I believe Physics and Machine Learning are really well aligned.
> First, the level of Math that you learn in Physics is a few decades away from what you use in modern Machine Learning. This makes the theory behind machine learning easy to understand.
> Second, and the most important is the mindset. In Physics, people are maneuvering between rigorous theory and actual data, trying to connect them sensibly. Machine Learning is very similar.
> I may be biased about the alignment of Physics and machine learning. Still, I know at least 20 Kaggle Grandmasters that have Physics majors.
P.S. I got PhD in theoretical Physics at UC Davis, but moved to Bay Area for a Machine Learning job.
I love Physics, and can get back to it after retirement, but right now now I am paid well, and more flexible. I like that I can choose company and geography. Postdocs and faculty members do not have this luxury.
But yes, it is a material sum of money that I prefer to spend on other activities.
1. Doctors estimate how good the insurance is. 2. If it is good they create a huge bill. 3. Health insurance negotiates the discount and it can be big. 4. Regardless of your insurance you pay plus/minus the same.
Why? Unclear.
About being in US for a long time and not having the insurance.
I like rock climbing, skiing, snowboarding. The chance of an accident in these sports is rather high. I would feel uncomfortable doing them without insurance.
But I am very curious, what part of the scans can cost $30k.
Requested itemized statement, but did not get it yet.
I do not think that HN is not the place to discuss the limitations and advantages of the US healthcare system.
It just reminded me that there is still a big market opportunity and ways to reduce costs for machine learning in the field of medical imaging.
What games do not work on 2080Ti or even 1080Ti?
And from a Deep Learning perspective 3080 does not look that great. 10Gb of GPU ram is a big problem.
Let's take infinity, subtract other infinitites add a bit of alchemy and here is our prediction. <- from mathematical perspective this is a very sketchy business.
But it works. And it works in other fields. For example in the theory of phase transitions (I have PhD on the topic :))
And it is pretty common in Physics to work with sketchy constructions for decades before mathematicians catch up.
I love the example of "Generalized Functions" (examples are Delta function and Step Function) that were widely in Physics decades before a consistent mathematical theory was developed.
I hope, at some point we will have a similar story with Renormalization theory.
From time to time, I review scientific papers for Machine Learning journals.
In the modern machine learning theoretical papers of the type: We looked in problem A, here is a pattern B and it is the case because <math.>
Nearly all papers that I get for a review are applied papers: We took dataset A, model B, tweaked knob C and our result is better than paper D, but we do not know why. I am fine with this approach. Many good ML ideas were introduced in this way.
But I give reject if authors did not provide the dataset and code that reproduces results.
It is not an issue for other reviewers and my reject is often overridden by editors and papers get accepted, even if there is not way that anyone ever will be able to reproduce it.