Hardware is the least of our worries. Having high quality open datasets in areas that matter is vastly more important issue.
To be able to do research, you have to spend ~$1500 amortized over 1-2 years. And AI hardware is improving much faster than Moore's law.https://twitter.com/NandoDF/status/782276800038727680 …
-
-
-
remind me where you work?
End of conversation
New conversation -
-
-
I agree. Also, there's a lot more to ML/DL research than running ImageNet-scale experiments.
-
ImageNet is done. Many important research problems involve reasoning and memory, not perception, and do not require big datasets
- Show replies
New conversation -
-
-
@NandoDF What would you consider to be a reasonable research platform hardware wise?Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
my challenge is access to data and interpretations that makes sense for my domain since gathering this information is costly.
-
This Tweet is unavailable.
- Show replies
New conversation -
-
-
@NandoDF $1500 seems too few (Titan X around 1000£, less on eBay), 20k£ seems too much, more than Nvidia Devbox. My price:5k€Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I regularly use a 3 year old $2k laptop with a nice GPU for research projects, good enough to try ideas on (esp during talks!)
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@NandoDF It depends on what kind models. E.g. good luck doing NMT research with $1.5k gpu. Hardware is an issue in some areas. -
Getting data is almost never an issue except for regulated areas like healthcare, or if you want to productize.
- Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.