Google and OpenAI publish new research on how neural networks build and arrange visual concepts using an “activation atlas”
Neural networks can learn to classify images … Christine Fisher / Engadget : Google is sharing a tool to keep your data anonymous from AI Alasdair Allan / Hackster Blog : Say “Hello” to Google Coral Tweets: Sundar Pichai / @sundarpichai : TensorFlow Privacy, using techniques based on differential privacy, will make it easier for developers to train ML models with privacy and better protect users' data in their AI development. #GoogleAI http://www.theverge.com/... Reza Zadeh / @reza_zadeh : Feature visualizations, letting us see through the eyes of the neural network. The hidden layers of neural networks are quite fun to inspect. Great interactive article by Distill. http://distill.pub/... http://twitter.com/... Zeynep Tufekci / @zeynep : Because I keep having a typo in the hashtag, reposting. Here's a comparison between regular TensorFlow code and tf-encrypted—with privacy-preserving python lib addition. I'm constantly surprised how relatively small these research programs are given their importance. #AMLD2019 pic.twitter.com/BTFkaC4Uff @wired : Neural networks are tremendously successful at tasks like identifying objects in images, but how they do so has remained largely a mystery—until today, when a paper was released that offers some insight into how neural networks “learn” visual concepts http://www.wired.com/... Zeynep Tufekci / @zeynep : First pic, regular TensorFlow. Second pic, tf-encrypted: a python lib for privacy-preserving ML. This is the track that needs *all* the attention. ML community can't control that its output will be used for good (it often won't be) but it can develop tools that do good. #AMDL2019 pic.twitter.com/SKu0G54pdq Ada Rose Cannon / @lady_ada_king : This makes me very excited, one of my biggest concerns with AI (besides how it acts as a magnifier for our own prejudices) is that it is often run on a machine owned by a 3rd party who can mishandle the data. http://twitter.com/... @openai : In collaboration with Google, we're releasing Activation Atlases: a new technique for visualizing what interactions between neurons can represent. Blog: https://blog.openai.com/... Paper: https://distill.pub/... Code: https://github.com/... Demo: https://distill.pub/... pic.twitter.com/7c6Q6GvNLq