François Chollet on Twitter: "Successful transfer learning from simulated environment to reality: e-sports pro beats F1 pro on the track Too bad it doesn't work like this for deep learning models… https://t.co/d8pNnW1dB0"

Country	Code	For customers of
United States	40404	(any)
Canada	21212	(any)
United Kingdom	86444	Vodafone, Orange, 3, O2
Brazil	40404	Nextel, TIM
Haiti	40404	Digicel, Voila
Ireland	51210	Vodafone, O2
India	53000	Bharti Airtel, Videocon, Reliance
Indonesia	89887	AXIS, 3, Telkomsel, Indosat, XL Axiata
Italy	4880804	Wind
3424486444	Vodafone
» See SMS short codes for other countries

1. mok33‏ @mokhtarmoussa 20 Jan 2019
  Replying to @fchollet
  
  I didn't get your point on deep learning, what does transfer learning to reality refers to with DL ?
  
  1 reply
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Andrew Stellman ‏ @AndrewStellman 20 Jan 2019
  Replying to @fchollet
  
  I wouldn't be surprised if @EnzoBonito17 has significantly more simulated track time under his belt than many F1 winners have in cars. And I bet he's seen conditions that rarely occur on a real-world track. It's like how a fighter pilot uses a flight simulator.
  
  1 retweet 6 likes
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. WeShouldAllWearMasks‏ @BradCanKook 20 Jan 2019
  Replying to @fchollet
  
  I'm now a believer in transferrable skills.
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Christopher Peters ‏ @statwonk 20 Jan 2019
  Replying to @fchollet
  
  Yep, the game's physics make it a much easier task for the e-sports pro.
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Chris‏ @ndx97 20 Jan 2019
  Replying to @fchollet
  
  That's the goal isn't it!
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Ketil‏ @ktlmld 20 Jan 2019
  Replying to @fchollet
  
  It does too work like that. Simulated data ftw!. Admittedly in a more constrained setting... :-) (Allken et al, ICES JMS, 2018)
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Ray Cao‏ @RayCao10 21 Jan 2019
  Replying to @fchollet
  
  Hi! There seem to be a bug in the keras.preprocessing.image, the flow_from_directory. when training using keras, the validation loss is still, while validation loss is decreasing using Tensorflow.
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo
1. Malik H. Altakrori‏ @AltakroriM 21 Jan 2019
  Replying to @fchollet
  
  I think training to drive "while in the car" prevents you from pushing the car to its limits due to the fear of crashing. On the other hand, you can try different things in a simulation, maybe crash a few times, then restart. Also, the point made above about the training time.
  Thanks. Twitter will use this to make your timeline better. Undo
  
  Undo