BC INTERNET
Saved by @CodyyyGardner

Here's an overview of key adoption metrics for deep learning frameworks over 2020: downloads, developer surveys, job posts, scientific publications, Colab usage, Kaggle notebooks usage, GitHub data.

TensorFlow/Keras = #1 deep learning solution.

 Note that we benchmark adoption vs Facebook's PyTorch because it is the only TF alternative that registers on the scale. Another option would have been sklearn, which has massive adoption, but it isn't really a TF alternative. In the future, I hope we can add JAX.
TensorFlow has seen 115M downloads in 2020, which nearly doubles its lifetime downloads. Note that this does *not* include downloads for all TF-adjacent packages, like tf-nightly, the old tensorflow-gpu, etc.
Also note that most of these downloads aren't from humans, but are automated downloads from CI systems (but none are from Google's systems, as Google doesn't use PyPI).

In a way, this metric reflects usage in production.
There were two worldwide developer surveys in 2020 that measured adoption of various frameworks: the one from StackOverflow, targeting all developers, and the one from Kaggle.
Note that the StackOverflow survey listed both TF and Keras; Keras had very strong metrics, and I suspect many people checked Keras without checking TF. So if "TF/Keras" was a choice, it would have significantly higher numbers here (probably around 15% overall usage).
Mentions in LinkedIn job posts is a metric that I'm not quite sure is meaningful, unfortunately. It doesn't reflect the stack of companies that hire, only the keywords tracked by recruiters.
We can track usage in the research community in two categories: ArXiv, which represents "pure deep learning" research, and Google Scholar, which represents all publications, including applications of deep learning to biology, medicine, etc.
Deep learning research is an important but small niche (~20k users of deep learning out of several millions in total) and it is the only niche where PyTorch is neck-to-neck with TensorFlow.
Finally, GitHub metrics. GitHub makes it possible to track new commits over the last year, but doesn't make it possible to track new stars/forks/watchers, hence why I'm displaying total numbers for these rather than 2020 increases.
Note that the GitHub metrics are only for the TensorFlow repo, not the dozens of large TensorFlow adjacent repos (like the Keras repo, etc).
Overall: 2020 has been a difficult year, in particular one during which many businesses have cut their exploratory investments in deep learning because of Covid, causing a slump from March to November. However, on balance, TF/Keras has still seen modest growth over the year.
Our current growth rate is solid, and our prospects for 2021 are looking bright! I'll post an update to these metrics in 2021. Here's to another year full of improvement, growth, and focusing on delighting our users :)

More from Internet

Marko \u26a1 Denic
Marko ⚡ Denic
@denicmarko
50 free tools and resources you're gonna love!

🧵

1. UI Garage
Daily UI inspiration & patterns for designers, developers to find inspiration, tools and the best resources for your project.

Link:

2. Remove bg
Remove Image Background: 100% automatically – in 5 seconds – without a single click – for free.

Link:

3. uiGradients
A handpicked collection of beautiful color gradients for designers and developers.

Link:

4. Blender
Free and Open 3D Creation Software.

Link:
INTERNET
Andy Pavlo
Andy Pavlo
@andy_pavlo
If you're interested in DB internals, stop what you're doing and watch @CMUDB Quarantine Talk from Nico+Cesar about @SQLServer's Cascades query optimizer: https://t.co/FCdsbHHEaD

Many talks this semester were good. This one is the best. My thread provides key takeaways

[6:50] Microsoft hired Goetz Graffe in the 1990s to help them rewrite original Sybase optimizer into Cascades. This framework is now used across all MSFT DB products (@SQLServer, @cosmosdb, @Azure_Synapse).

[14:43] The optimizer checks whether it has stats it will need before cost-based search. If not, it blocks planning until the DBMS generates them. This is different than other approaches we saw this semester where DBMS says "we'll do it live!" with whatever stats are available.

[21:05] Their Cascades' search starts small/simple and then they make decision on the fly whether to expand search based on the expected query runtime and performance benefit from more search.

[26:33] They explicitly have a property for Halloween Problem. Operators specify whether they protect from it and then optimizer ensures property is satisfied. This is mindblowing. I have never thought about using the optimizer for this but it makes sense.
INTERNET
Advertisement
Kristine Schachinger
Kristine Schachinger
@schachin
@curaffairs The author doesn't know enough about search to have written that article and the reason he's having those problems is AI algos like Rank Brain and Google's attempts to use natural language processing nothing to do with SEOs.

1/x

YMYL is not evaluated at the site level, but the query level.

Again this is on Google, not SEOs because they want to surface better sites.

And those devaluations are part of Core Updates which SEOs like me recover sites from.

2/

Featured Snippets are not optimized for - you can use certain HTML to be pulled in one but there is no definite use case where Featured Snippets are shown or that your site's will show for it. Also doesn't always come from the #1 result.

& that's Google not the SEO.

3/

Also there are many times when an SEO doesn't want their site in the Featured Snippet cuz it doesn't result in a click to the website. But we have no control over what Google pulls as Feature Snippets except to put in a meta tag to exclude us from them

4/

And Neil Patel is not an SEO. Most senior level SEOs see him as nothing more than a scammer because reportedly he's ripped them off by stealing their content and articles and passing them off as his.

5/
CULTURE , INTERNET , SEO
Amarnath Amarasingam
Amarnath Amarasingam
@AmarAmarasingam
1. (THREAD) So, it seems like the deplatforming debate is once again kicking off, so I thought I would introduce some of the earlier work that was done in this area back when ISIS was buck wild on social media. What have we learned over the last six years might be useful today:

2. One of the earliest studies that discussed the impact of suspensions of ISIS accounts was @intelwire and Morgan's piece: The ISIS Twitter Consensus.

They found that suspensions did have an impact on replies and retweets and overall dissemination.

3. After suspensions, the die-hard supporters dedicated themselves to creating new accounts, but others whittled away: “it appears the pace of account creation has lagged behind the pace of suspensions”

4. On the specific question of how suspensions impact the Twitter network, see this piece by @intelwire and @IntelGirl111, which explores how suspensions impact these groups, including major disruptions to dissemination and decline in follower count:

5. Another study by @Aud_Alexander similarly found that ISIS supporters were finding it hard to “gain traction” after Twitter took a harder stance on the group.
INTERNET , ALL
BlackRoomSec
BlackRoomSec
@blackroomsec
Or, you could let us know when you figure out why it was trending yesterday and the users are complaining which is why Vice wrote about it. Why I'm saying what I am.

There's an assumption here that this problem is fixed bc it was already hacked.

It's not.


We have ppl freezing and dying in TX right now because some ppl who thought they were really smart never spoke to anyone with actual experience with energy systems in extreme cold climates.

Texans are waiting for a solution to a preventable problem.

Farmers are saying that now, not during a crisis, they have to wait for a JD tech to arrive to help them.

The assumption that bc SOME farmers said screw this and used hacked firmware to get around that obstacle doesn't mean that all farmers are doing that.

If all farmers were using hacked firmware we wouldn't be discussing this right now would we?

Of course no one has pointed out that another issue here is that no one at John Deere has figured out they don't have enough staff to quickly and reliably SOLVE problems.

The locked firmware is just one of many issues here.

By not giving people a fast solution they're causing this and other issues.

So what I'm saying is somebody at John Deere needs to examine all of this.
INTERNET

You May Also Like

Cookie Cyboid
Cookie Cyboid
@CookieCyboid
“Why are Tumblr and Facebook cracking down on sex so much? And why is Facebook in so over zealous about it? It’s all Apples fault!” Etc etc

I can explain all. It is largely the fault of a set of US laws called SESTA and FOSTA.
TECH
Jaya_Upadhyaya
Jaya_Upadhyaya
@Jayalko1
SRI BANASHANKARI AMMAN, BANGALORE
#bharatmandir

Located in Southern Bangalore, this is one of the most popular places of worship in Bangalore City.
Devoted to Devi Banashankari, this shrine is visited specially during Rahukala which is believed to be an unfavorable time.


Somanna Shetty, a great devotee of Devi, brought this murti from Badami’s Banshankari Amma Kshetram.
It is said that many years ago, three ladies visited this shrine and requested the pujari to conduct puja for them. Since it was the inauspicious rahu kaal, the pujari refused.


However, upon their insistence, he conducted the puja. When he returned with prasadam, the ladies vanished. This made the people believe that they were none other than the Devi herself. Since then, the people perform poojas here during the Rahukaal.
@GunduHuDuGa


People use lemons as a diya. Lemons are cut from the centre and their pulps are removed, for pouring oil in them.
It is strongly believed that Devi removes all the troubles from the life of her devotees during the Rahukala although it is considered inauspicious for any pooja.
ALL
Advertisement
Dr Fish Philosopher Todd (Dr FPT)\U0001f41f
Dr Fish Philosopher To...
@ZoeSTodd
1. I was walking in the woods yesterday and I had an epiphany. It wasn’t one I wanted to have. I was thinking about today’s class, where we are reading Kimmerer’s ‘Gathering Moss’. I’ve taught Braiding Sweetgrass in multiple classes every year since 2016. It’s eminently teachable

2. Both books are beautifully written. Students uniformly enjoy them as assigned reading. They do a great job communicating to non-Indigenous folks the contours of basic Indigenous relationships to more-than-human beings. They’re wildly popular with white readers.

3. I’ve found Braiding Sweetgrass to be a great entry point for non-Indigenous folks to Indigenous scholarship. But this is where yesterday’s epiphany comes in. Neither Braiding Sweetgrass or Gathering Moss tends to the deep deep lineages of Indigenous scholarship in Canada or US

4. As the popularity of ‘Braiding Sweetgrass’ has soared, I’ve seen more and more white folks wax on about ‘braiding knowledge’. But they often don’t know any of the lineages of these concepts beyond what is presented in this one book. They confidently run with this.

5. Kimmerer does a commendable job of weaving in her own experiences, her family stories+nods to some very important knowledge keepers who deserve to be celebrated for their work. But Braiding Sweetgrass does not tend to the citational politics Sara Ahmed & others teach us about
SOCIETY
Roger Grosse
Roger Grosse
@RogerGrosse
Important paper from Google on large batch optimization. They do impressively careful experiments measuring # iterations needed to achieve target validation error at various batch sizes. The main "surprise" is the lack of surprises. [thread]

https://t.co/7QIx5CFdfJ


The paper is a good example of lots of elements of good experimental design. They validate their metric by showing lots of variants give consistent results. They tune hyperparamters separately for each condition, check that optimum isn't at the endpoints, and measure sensitivity.

They have separate experiments where the hold fixed # iterations and # epochs, which (as they explain) measure very different things. They avoid confounds, such as batch norm's artificial dependence between batch size and regularization strength.

When the experiments are done carefully enough, the results are remarkably consistent between different datasets and architectures. Qualitatively, MNIST behaves just like ImageNet.

Importantly, they don't find any evidence for a "sharp/flat optima" effect whereby better optimization leads to worse final results. They have a good discussion of experimental artifacts/confounds in past papers where such effects were reported.
MACHINE LEARNING
Aditya Todmal
Aditya Todmal
@AdityaTodmal
A THREAD ON @SarangSood

Decoded his way of analysis/logics for everyone to easily understand.

Have covered:
1. Analysis of volatility, how to foresee/signs.
2. Workbook
3. When to sell options
4. Diff category of days
5. How movement of option prices tell us what will happen

1. Keeps following volatility super closely.

Makes 7-8 different strategies to give him a sense of what's going on.

Whichever gives highest profit he trades in.


2. Theta falls when market moves.
Falls where market is headed towards not on our original position.


3. If you're an options seller then sell only when volatility is dropping, there is a high probability of you making the right trade and getting profit as a result

He believes in a market operator, if market mover sells volatility Sarang Sir joins him.


4. Theta decay vs Fall in vega

Sell when Vega is falling rather than for theta decay. You won't be trapped and higher probability of making profit.
OPTIONSLEARNINGS , OPTIONSBASICS , OPTIONSLEARNINGSSARANGSOOD