This seminar will begin at 4:00 pm at the Jerome L. Greene Science Center on Columbia University’s Manhattanville campus (L7-119). Light refreshments will be available starting at 3:45 pm.

This month's speakers:

Tal Golan, PhD (Kriegeskorte lab): "Testing deep neural network models of vision with controversial stimuli"

Deep neural networks (DNNs) are currently the best available models of visual object recognition, but different DNN models often make similar predictions. To efficiently compare different DNN models as hypotheses about human vision, we propose to use controversial stimuli: images optimized to induce disagreement between models. We piloted this approach on models trained to classify handwritten digits (MNIST). We optimized images to be classified as different digits by different models, and compared the model- and human- judgments of these synthetic controversial stimuli. We found that class-conditional generative models are considerably better than discriminative models at simulating how humans perform the MNIST task.

Jin Zhang, PhD (Zuker lab): "Sour sensing from the tongue to the brain"

The ability to sense sour provides an important sensory signal to prevent the ingestion of unripe, spoiled, or fermented foods.  Taste and somatosensory receptors in the oral cavity trigger aversive behaviors in response to acid stimuli.  Here we show that the proton channel OTOP1 is the sour taste receptor.  Next, we genetically identified the taste ganglion neurons mediating each of the five basic taste qualities and demonstrate that sour taste uses its own dedicated labeled line from taste receptor cells in the tongue to finely tuned taste neurons in the brain to trigger aversive behaviors.

This seminar is part of the Zuckerman Institute Postdoctoral Seminar series. For questions about this or future seminars, please contact [email protected].