A huge fan of putting this in papers (from the BigGAN paper https://t.co/HCcLIPMSZ4) pic.twitter.com/1IxLJZML4Q
— James Bradbury (@jekbradbury) September 29, 2018
A huge fan of putting this in papers (from the BigGAN paper https://t.co/HCcLIPMSZ4) pic.twitter.com/1IxLJZML4Q
— James Bradbury (@jekbradbury) September 29, 2018
"If someone shows you simulations that only show the superiority of their method, you should be suspicious. A good set of simulations will show where the method shines but also where it breaks." @BryanSmucker
— Edward Tufte (@EdwardTufte) September 28, 2018
More generally, THE "MY DOG IS THE BEST DOG IN THE WORLD" FALLACY
Today's life tip: Don't be afraid to give yourself permission to reach for your dreams. If a single humble linear regression can become "AI", what's stopping you? Ask for that promotion. Put that tech you've been working with for three weeks on your resume. Shoot for the stars.
— Vicki Boykis (@vboykis) September 27, 2018
MISTAKE #8
— 👩💻 @DynamicWebPaige 🔜 #APICityConf 🌇 (@DynamicWebPaige) September 27, 2018
Use known best practices when designing and implemting your experiments.
🎲Randomization = assign by chance, not by choice
⚖Blinding = mask information from participants+tester
👩🔬Controlling = minimize the effects of variables other than the independent variable
Assumptions of independence are rarely testable and have huge effects on statistical inferences: https://t.co/7B16CpUdiL
— John Myles White (@johnmyleswhite) September 25, 2018
Advice for aspiring data scientists: learn SQL, communication is a technical skill, start a blog, teach others, don't worry about learning everything, find your community, stay curious, have fun, don't be afraid to use Google, and remember that everyone is winging it.
— Caitlin Hudon👩🏼💻 (@beeonaposy) September 25, 2018
Most paradoxical neural net phenomena are unexplained only in the same sense that UFO sightings are unexplained. If you dig deeper, there's almost always a mundane explanation in terms of principles we already understand.
— Roger Grosse (@RogerGrosse) September 23, 2018
We have yet to discover the screwdriver. https://t.co/62FEd4oUQh
— hardmaru (@hardmaru) September 20, 2018
I‘m in the minority but I take the complete opposite view from this Kai Fu Lee dude. The greatest discoveries are yet to be made. The current algorithms are not really “AI”, but rather just machine learning. They are not well known and people are still figuring out how they work.
— hardmaru (@hardmaru) September 20, 2018
I have a different controversial opinion: ML development is a different kind of software development and has a different set of best practices.
— Ian Goodfellow (@goodfellow_ian) September 19, 2018
Overfitting is always a good first step. Overfitting on training data and making sure the results look reasonable and that there are no obvious bugs is what everyone should do first for a new task/model :)
— Denny Britz (@dennybritz) September 17, 2018
For anyone debating the value of using physical simulations to improve forecasts, here's Hurricane Florence's actual track compared to all other tropical systems in the area in historic Septembers
— Ben Hamner (@benhamner) September 17, 2018
A simulation-based approach nailed the forecast track. Pure ML would fare terribly pic.twitter.com/QSaNzaLGMc