The AI Index 2019 Report

https://hai.stanford.edu/news/introducing-ai-index-2019-report

by smt1 • 7 years ago

103 23 7 years ago

hai.stanford.edu

> 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.

grammarxcore• 7 years ago

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.

account734668• 7 years ago

> 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.

scottlocklin2• 7 years ago

I'm pretty sure there is a ~2002 "Nanotech index report" out there which looks a lot like this. Oh looksie; I am right:

http://www.nanotech-now.com/CMP-reports/NOR_White_Paper-July...

account734668• 7 years ago

True but I am not sure "nanotech" was used by top S&P 500 companies and their engineers on a daily basis.

scottlocklin2• 7 years ago

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.

nostromo• 7 years ago

Actually it’s quite analogous.

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.

keithyjohnson• 7 years ago

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.

ummonk• 7 years ago

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).

antpls• 7 years ago

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)

cyorir• 7 years ago

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.

hooande• 7 years ago

read the actual report instead of just the bullet points. the speed improvement is a function of cost on cloud hardware

utopian3• 7 years ago

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)

The AI Index 2019 Report

Related Stories