🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

What a fantastic opportunity for #myfirstTweet -- a group shot at the biennial Massachusetts Institute of Technology (MIT) brain & cognitive sciences community retreat!

Congrats #MIT2018! I know you’ll make us proud! x.com/mitbrainandcog…

I’m thrilled for Rishi Rajalingham, who successfully defended his PhD thesis on how the primate visual ventral stream supports core object recognition yesterday! Congrats Dr. Rajalingham!

One of my dreams is to help accelerate progress in the brain sciences and AI by defining shared goals and facilitating ways to work toward them as a larger community of data collectors and model builders. We share a concrete step here called Brain-Score: biorxiv.org/cgi/content/sh…

Five years ago, deep convolutional NNs inspired by brain science advanced to become the best models of the brain's visual processing. Since then, CV/ML has created more complex models, without regard to the brain. Here we aim to follow the brain science: doi.org/10.1101/408385

Trying to make good on today’s CCN discussion of how to head towards understanding.

Do deep ANN models of the brain's ventral visual stream constitute an "understanding" of that biological system? One test of "understanding" is the new power it gives us humans. Here is one example of such power -- non-invasive neural control: biorxiv.org/content/early/…

Great piece by Melanie Mitchell in the The New York Times. We share the belief that the intersection of the science of natural intelligence and the engineering of artificial intelligence is the way forward. So let’s do this! MIT Quest for Intelligence. nyti.ms/2D4wtAg

Do artificial deep network models constitute an "understanding" of visual processing? As a practical test of understanding, Bashivan and Kar show that the knowledge already embedding in these networks gives a new power of non-invasive neural control: doi.org/10.1126/scienc…

One week left to apply to our faculty search in computation... We seek candidates trying to understand natural intelligence by building artificially intelligent systems, constrained by neuroscience and cognitive science. Apply here: academicjobsonline.org/ajo/jobs/14277

In making precise comparisons of leading ANN vision models with biological V1 neurons, we noticed that models that were more brain-like were also more adversarially robust. That led to this very exciting new work by some amazing students and postdocs! biorxiv.org/content/10.110…

Some ANNs are descent models of the mechanisms of the adult ventral visual stream, but are not viable scientific hypotheses of its post-natal learning. This work led by Zhuang and Yamins is their latest exciting advance toward achieving that goal: biorxiv.org/content/10.110…

Incredibly proud of this new topographic deep ANN work by Hyo Lee! She and her collaborators show that the IT face processing network could be a simple consequence of general object recognition plus neural wiring costs. Congrats Hyo! biorxiv.org/content/10.110…

Announcing the ThreeDWorld: A Platform for Interactive Multi-Modal Physical Simulation. Project page, with code release: threedworld.org Paper link: arxiv.org/pdf/2007.04954…

If you love the nuts and bolts of computing and current machine intelligence, and your dream is to discover how biological intelligence actually works, please consider joining us as faculty on this quest at MIT.

Cool paper from colleagues Ratan Murty and Nancy Kanwisher just out now (rdcu.be/cx5qI): Using state-of-the-art models of the ventral stream to ask what really drives "face","place" and "body" fMRI patches in humans. Their models vs. experts comparison is super fun!

Pleased to share recent results from Guy Gaziv and Michael Lee showing that models of ventral visual processing can guide the design of photon patterns to predictably modulate human category perception. preprint: arxiv.org/abs/2308.06887 project webpage: himjl.github.io/pwormholes