@ProofOfLogic … "that image is a letter 'A'", "that image is a letter 'B'", "that image is a letter 'C'", etc … cc: @AlleleOfGene
-
-
Replying to @reiver
@ProofOfLogic … I might use a type of Artificial Neural Network for this. For the sake of example, let's say I'm using … cc:@AlleleOfGene2 replies 0 retweets 2 likes -
Replying to @reiver
@reiver@ProofOfLogic@AlleleOfGene These can deliver effective classifiers, with low computational cost but they don't model reality well.2 replies 0 retweets 3 likes -
Replying to @MakerOfDecision
@MakerOfDecision By "modelling reality" are you referring to interpretability of the model? … cc:@ProofOfLogic@AlleleOfGene1 reply 0 retweets 2 likes -
Replying to @reiver
@MakerOfDecision …If "yes", then I'd agree; ANNs are infamous for not being interpretable :-) cc:@ProofOfLogic@AlleleOfGene1 reply 0 retweets 2 likes -
Replying to @reiver
@reiver@ProofOfLogic@AlleleOfGene True, but no; For a model to be used for decision making, it needs to be causal. Neural nets can predict1 reply 0 retweets 3 likes -
Replying to @MakerOfDecision
@reiver@ProofOfLogic@AlleleOfGene but they are not for representing causal effects, just accurate predictions conditional on training.1 reply 0 retweets 0 likes -
Replying to @MakerOfDecision
@reiver@ProofOfLogic@AlleleOfGene Unless the NN is being used to model a system that actually works that way (say, a brain,) if you use1 reply 0 retweets 1 like -
Replying to @MakerOfDecision
@reiver@ProofOfLogic@AlleleOfGene the model to decide what to change, it is unlikely to model the effects of that intervention correctly.1 reply 0 retweets 1 like -
Replying to @MakerOfDecision
@reiver@ProofOfLogic@AlleleOfGene In fact, the only ways I know to model a causal phenomenon is either to replicate the dynamics, or...1 reply 0 retweets 2 likes
@reiver @ProofOfLogic @AlleleOfGene to represent causal effects via statistical causal model. (regardless of if it is Rubin's, or Pearl's.)
-
-
Replying to @MakerOfDecision
@reiver@ProofOfLogic@AlleleOfGene In theory, you can train a NN to predict a causal phenomenon, with the right training data - but1 reply 0 retweets 1 like -
Replying to @MakerOfDecision
@reiver@ProofOfLogic@AlleleOfGene it's not generally the right tool for the job. (It still works because NNs are a universal approximator)0 replies 0 retweets 2 likes
End of conversation
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.