St Louis, Aspirin, AI discussed on This Week in Machine Learning & AI

Automatic TRANSCRIPT

Ones we measuring the brain train. In some ways you can feel like neural network you. You have some laws on the top layer. Which is what the class by this stubble. Latent variables are lead free to do whatever they want right so we try to put some constraints in this middle layers in St Louis. Trying one of these layers to make it look a little bit more like the brain of a mouse and what we found was that that modell became more robust two participations. It was interesting because it became more robust to high frequency noise perturbations which are now. Starting we can expand the simpler way that it may be like. You know the more sensitive to low spatial frequencies but this is at least a proof of concept that you know even if these cases may be a simpler scenario that you should if you know that you could learn something of their presentation a level in a system and then translate Dover we have the tools to sort of movie daughter because we have a model over the brain deep neural network. And he's trained on data and then we have a neural network model that tries to Show that both models. They're both build of the same ingredient which is neurons. We sign up says ceiling can try now transfer information from one to the now but still not anywhere in my personal taste even if we were to do these in something. Incredible it's his Attack it was still not very satisfying. We're GONNA White Sobe. You know the mayor copying operation itself. You know you may word but for me personally to beside his fight I would like to understand why working and end up getting goes back to the some aspirin to predictability inside wage year. It almost kind of points to at least you're you're you're limited experiment. You know kind of calls to mind this future where instead of pulling a You know building a network up from conversational layers and pooling layers. Where you know. I'll use a mouse layer. And a flat worm layer players controlling. Yeah Yeah and I think the other thing is to show the ut lead you of these systems we need to come up with the right machine learning benchmark tests and right now the the focusing machine learning in the last ten years has been gained to create benchmark. Were very difficult in the nineties in the eighties. For computers. Like ultra Right Bud there hasn't been as much or more now but where you train. You want to build a visual of system as it possible visual tasks riding sigma More tracking action recognition can be doing where we started giving like the way. Biology system works right and fro main. Any male she'll have to who is working on the problem like if you're having genie are especially working in a company. They want to build on was driving. They don't care that sees damn the study group General because they'll never be tested on Sunday. You're is right. The or that same system is not the one that's going to talk to you to the driver right may be so but from if you're interested from the AI..

Coming up next