BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Jenelle Feather (MIT) DTSTART:20201002T160000Z DTEND:20201002T050000Z DTSTAMP:20260423T005839Z UID:CRIBB/2 DESCRIPTION:Title: Me tamers of neural networks reveal divergence from human perceptual systems< /a>\nby Jenelle Feather (MIT) as part of Computational Research in Boston and Beyond Seminar (CRIBB)\n\n\nAbstract\nArtificial neural networks now a chieve human-level performance on tasks such as image and speech recogniti on\, raising the question of whether they should be taken seriously as mod els of biological sensory systems. Such neural network models exhibit huma n-like patterns of behavior\, and their feature spaces reliably predict br ain activity. On the other hand\, neural network models can often be foole d by small adversarial perturbations that have no effect on humans. In thi s talk\, I will detail our work using “model metamers” to investigate similarities between neural networks and human sensory systems. Model meta mers are physically distinct stimuli that produce nearly the same response within a model\, and thus the same model prediction. Our results show tha t despite replicating aspects of human behavior and neural responses\, pre sent-day deep neural networks learn invariances that deviate markedly from those of biological sensory systems. Model metamers may help guide future model refinements to reduce or eliminate these discrepancies.\n\nhttps:// mit.zoom.us/j/96155042770 -- Meeting ID: 961 5504 2770\n LOCATION:https://researchseminars.org/talk/CRIBB/2/ END:VEVENT END:VCALENDAR