Many machinery systems were engineered by humans based on principles that are observed in nature, for instance planes were highly inspired by the way birds fly. Similarly, biological brains are of major interest to discover the important principles underlying true general intelligence as they are the only existing proof that such a system can exist. Thus, modern AI, in particular deep learning, has been highly inspired by the brain’s structure and computations. Still, deep neural networks cannot extrapolate well to out-of-distribution events and are lagging far behind biological brains with respect to vision and other aspects of intelligence. My long-term goal is to address these limitations of AI and try to induce the right inductive biases from the brain into artificial networks to render better-generalizing networks, narrowing the gap between artificial and biological intelligence. Currently, I am working on the neural co-training of artificial networks on image classification and brain activity to bias computations in these networks towards brain like representations. Furthermore, I am investigating the inductive biases our co-training approach may induce in these models.
Publications
Lab members are shown in this color.
2021
|
Shahd Safarani, Arne Nix, Konstantin Willeke, Santiago A. Cadena, Kelli Restivo, George Denfield, Andreas S. Tolias, Fabian H. Sinz
Towards robust vision by multi-task learning on monkey visual cortex
NeurIPS (accepted)
|
|
Shahd Safarani, Arne Nix, Konstantin Willeke, Santiago A. Cadena, Kelli Restivo, George Denfield, Andreas S. Tolias, Fabian H. Sinz
Towards robust vision by multi-task learning on monkey visual cortex
ICLR 2021 Workshop: How Can Findings About The Brain Improve AI Systems?
|