RobustML

Let's make #MachineLearning Robust

[RobustML @robustML]

@twitter @jack you asked what we think of new twitter... it sucks and you know it.

[RobustML @robustML]

between #supervised and #unsupervised learning approaches there is a whole gray matter where the #robustml resides

[RobustML @robustML]

Some say - #DeepLearning
We say - not-so-robust-learning :)
More evidence is piling.
https://t.co/78C4sDHkWd
#robustML

[Dan Hendrycks @hendrycks]

Natural Adversarial Examples are real-world and unmodified examples which cause classifiers to be consistently confused. The new dataset has 7,500 images, which we personally labeled over several months.
Paper: https://t.co/L5YMIHZGbz
Dataset and code: https://t.co/4UzQWC4TjG

[RobustML @robustML]

"Machine learning will always produce an answer" is an uneducated statement by academics eager to one up on others. It is true in absence of systems around ML. #robustML takes care of all this

[RobustML @robustML]

Reliability is one of the metric of #RobustML
MachieLearning is a two step process - Training step, Inferencing (production) step. Reliability metrics are required in both. A modern compute/software infrastructure will play a very important role in addition to data/methods.

[RobustML @robustML]

Blockchain technology can add to the robustness of ML technology #robustML - #traceability

[RobustML @robustML]

CNNs are designed to process and classify images for computer vision and many other tasks. But slight modifications — say, a few darker pixels within an image — may cause a CNN to produce a drastically different classification. NOT a #robustML

[RobustML @robustML]

Enabling machines to process data from various sources to identify fraud or detect defects ("Anomaly Detection Automation") was the most appealing idea to lenders, followed by "Borrower Default Risk Assessment."