>> Diversifying AI faculty along gender lines has not shown great progress, with women comprising less than 20% of the new faculty hires in 2018.
I am wondering what is % of women doing PhD in the same field and whether it is growing. Without the latter number growing it would be hard to have a greater % of women among new faculty hires.
We'd also need to see numbers on new positions and freshly available positions. Even if the number of candidates is relatively stagnant, they can't get jobs if their number is larger than the number of positions available. Many of the doctoral candidates I knew in school couldn't find positions in academia because of market saturation.
>> Many of the doctoral candidates I knew in school couldn't find positions in academia because of market saturation.
The situation is now drastically different compared to say 40 years ago. That is why so many PhDs leave for industry to be better paid. The latter is possible for CS guys but in other domains the situation is pretty bad since they don't have easily accessible highly paid industry jobs.
Newsflash: neither is "AI." Unless you expand the word "nanotech" to include things like toothpaste and paint, or the word "AI" to mean things like logistic regression and random forests.
I'm curious - how do you know it's used for most industries? I'm sure there are lots of industries that use ML, and lots that don't. Is there a report on this somewhere? I just get a lot of worthless hype articles when I google it.
FAANG isn't "most industries" by any stretch. This ignores healthcare, automotive, pharmaceutical, agricultural, and so many more industries.
All you've done is point out that there are a lot of ML papers coming from tech companies. Nobody is disputing that. Why do you assume this generalizes to most industries?
Go to Google Scholar, type your favorite company and "deep learning". If you find no paper from that company, chances are high their stock price is not beating S&P 500 over the last 5 years.
Also, top S&P 500 companies by market cap are "AAPL","MSFT","GOOG","AMZN","FB","BABA".
This is a No True Scotsman argument. Who said anything about beating the S&P 500? I said most industries, and every comment you are redefining what that means.
"1) I ask why this applies to most industries
2) You point out it applies to FAANG"
That is a lie because I said "often" in "You see that from papers published at top ML conferences. Top contributors are often from FAANG. Some papers are clearly applications." without saying that FAANG = most industries. Instead, I said you can see that from these conferences. Thus, you made it up.
"4) You say most industries must only include those better than the S&P 500"
That is a lie. I didn't say that. I said "If you find no paper from that company, chances are high their stock price is not beating S&P 500 over the last 5 years." I only mentioned a correlation and said nothing about your "most industries".
You misinterpret the text. Fortunately, the text is written so it was easy to prove you are liar, boy.
When nanotech became a hot concept, the goal posts were moved so that companies could claim they were nanotech companies even though they were using pretty basic chemistry. No nanobots required.
And now we’re seeing a lot of statistics being rebranded as AI because the latter attracts investors and journalists.
AI is the term used by marketing division and clueless public. Researchers don't use that term without a special need. FAANG engineers actually use ML approaches.
I get that these bullets points are answering What instead of Why but for those that are more readily discernible, like "In a year-and-a-half, the time required to train a large image classification system on cloud infrastructure has fallen from about three hours in October 2017 to about 88 seconds", what's causing this? Are models getting smaller without a loss in accuracy? Is training distributed over a greater amount of cheaper machines? Personally, I'd be more excited about the former rather than the latter. We can't all afford MegatronLM-type experiments - https://nv-adlr.github.io/MegatronLM.
Both. Companies are certainly building bigger and bigger clusters for training.
At the same time though, consumer GPUs have gotten significantly faster (compare e.g. an Nvidia 2080TI to a 980TI), and learning algorithms keep improving / better learning algorithms become more widely used (e.g. Adam instead of stochastic gradient descent).
And also, architectural search allowed for neural networks to use more efficient builtin blocks, using many less parameters, and achieving the same accuracy with smaller models (and lowering training cost)
The improvements in the report are mainly from improvements in cloud infrastructure, but that's not to say there haven't been improvements in developing small, efficient models as well. One notable model that was introduced in 2017 was MobileNet, which aimed to create a model that could function on a mobile device without much loss in accuracy. There have been many more attempts to shrink models for use on devices with limited resources since 2017. These smaller models tend to have lower training times as well.
> In a year-and-a-half, the time required to train a large image classification system on cloud infrastructure has fallen from about three hours in October 2017 to about 88 seconds in July 2019. During the same period, the cost to train such a system has fallen similarly.
3 hours to 88 seconds? Wow... I wonder if that's been further reduced today (December 2019)
25 comments
[ 3.3 ms ] story [ 73.2 ms ] threadI am wondering what is % of women doing PhD in the same field and whether it is growing. Without the latter number growing it would be hard to have a greater % of women among new faculty hires.
The situation is now drastically different compared to say 40 years ago. That is why so many PhDs leave for industry to be better paid. The latter is possible for CS guys but in other domains the situation is pretty bad since they don't have easily accessible highly paid industry jobs.
http://www.nanotech-now.com/CMP-reports/NOR_White_Paper-July...
All you've done is point out that there are a lot of ML papers coming from tech companies. Nobody is disputing that. Why do you assume this generalizes to most industries?
Also, top S&P 500 companies by market cap are "AAPL","MSFT","GOOG","AMZN","FB","BABA".
I said.
>> and every comment you are redefining what that means
That is false.
You could learn something. Instead, you wasted my time.
2) You point out it applies to FAANG
3) I say FAANG isn't most industries
4) You say most industries must only include those better than the S&P 500
5) I point out a flaw in your reasoning
6) You insist you are right, upset that I tried to ask a question
7) I summarize what happened and exit this thread
That is a lie because I said "often" in "You see that from papers published at top ML conferences. Top contributors are often from FAANG. Some papers are clearly applications." without saying that FAANG = most industries. Instead, I said you can see that from these conferences. Thus, you made it up.
"4) You say most industries must only include those better than the S&P 500"
That is a lie. I didn't say that. I said "If you find no paper from that company, chances are high their stock price is not beating S&P 500 over the last 5 years." I only mentioned a correlation and said nothing about your "most industries".
You misinterpret the text. Fortunately, the text is written so it was easy to prove you are liar, boy.
When nanotech became a hot concept, the goal posts were moved so that companies could claim they were nanotech companies even though they were using pretty basic chemistry. No nanobots required.
And now we’re seeing a lot of statistics being rebranded as AI because the latter attracts investors and journalists.
At the same time though, consumer GPUs have gotten significantly faster (compare e.g. an Nvidia 2080TI to a 980TI), and learning algorithms keep improving / better learning algorithms become more widely used (e.g. Adam instead of stochastic gradient descent).
3 hours to 88 seconds? Wow... I wonder if that's been further reduced today (December 2019)