True. https://t.co/Rs5YFJrHYN
— Yann LeCun (@ylecun) January 26, 2019
True. https://t.co/Rs5YFJrHYN
— Yann LeCun (@ylecun) January 26, 2019
Many businesses choose to take a secret sauce approach to data.
— Neil Lawrence (@lawrennd) January 24, 2019
So secret, in fact, that no one even tastes the sauce.
When the bottle is eventually opened, the data turns out to be rancid. #DataReadinessLevels #RancidData
If you are learning the skill of posing interesting and well-structured analytics questions out of the mess of reality and w.r.t. the data that's available, one trick is to make yourself write down what you think the answer will be before you write code. https://t.co/Z9x1cyjoga
— Hilary Mason (@hmason) January 22, 2019
Data scientists can fail by:
— Emily Robinson (@robinson_es) January 18, 2019
❌not saying no enough
❌not providing anything more than a cursory analysis
❌assuming PM knows enough to ask question in the right way and not collaborating with them
❌caring more about using fancy method than solving business problems#rstudioconf
Lessons learned from growing a data science organization: #rstudioconf
— Julia Silge (@juliasilge) January 18, 2019
❌Don't hire data scientists before you have data
😳Getting it wrong allows you to do it right next time
🌱Adapt as a business grows
❤️Scale systems that allow your team to work together (your values)
If you’re a junior data scientist, you should be valued by your organization! You bring the ability to question dogma. And it’s okay to say “I don’t know but I can figure it out.” @AngeBassa #rstudioconf
— Emily Robinson (@robinson_es) January 18, 2019
#Prediction
— Andrew Trask (@iamtrask) January 15, 2019
The notion of a neural "architecture" is going to disappear thanks to meta learning
Instead, we will learn and relearn tasks over and over until we find the optimal connection of weights/layers to learn said tasks with the least number of data points
Brain-esque...
Nature stuff all around us (plants, animals, etc) are best thought of as basically super advanced alien technology. These are nanotechnology devices magically grown in ambient conditions with complex information processing. Synthetic bio is tinkering with / hijacking this tech.
— Andrej Karpathy (@karpathy) January 14, 2019
The generator matters. The perhaps most dangerous form of fake news are statistics/tables/charts used to make things look “scientific” - Statistics are meaningless without a description of how exactly they were derived. The generator.
— Denny Britz (@dennybritz) January 13, 2019
Reading lots of ML papers may make you an expert in writing ML papers, but it won’t make you an expert in implementing and applying ML to real world problems. The skills are related and somewhat complementary, but not at all equivalent.
— Denny Britz (@dennybritz) January 11, 2019
"If you’re building an NLP algorithm today, don’t do it on your own! Start from the best open-sourced algorithms [...] and then give back to the community by open-sourcing your improved algorithm and ideas."
— Sebastian Ruder (@seb_ruder) January 11, 2019
Good advice from @huggingface's @Thom_Wolfhttps://t.co/WMUEbJuaZN
The core weakness of reCAPTCHA is its near-universal use, which creates a considerable incentive to invest in a breaker. If instead everyone used a variety of different, reasonably effective spam prevention mechanisms, the lives of spammers would be much harder. https://t.co/jfMGbu7xtZ
— François Chollet (@fchollet) January 7, 2019