Seungdae Baek

Hard-wired Gabor filters enable environment-agnostic object recognition without overfitting.

This paper shows that deep neural networks can spontaneously develop invariant object detection without training, through random feedforward projections.

This study shows that number-selective neurons can emerge in untrained deep neural networks, mimicking innate number sense observed in animals.

The paper shows that sparse long-range connections in visual cortex create small-world networks that enhance object recognition while minimizing wiring costs.

This study shows face-selective neurons can emerge spontaneously in deep neural networks without training, exhibiting viewpoint-invariant properties similar to those in primates.