Understanding How We Perceive Animal-Like Objects
Author Information
Author(s): Duyck Stefanie, Costantino Andrea I., Bracci Stefania, Op de Beeck Hans
Primary Institution: KU Leuven
Hypothesis
Can deep neural networks be trained to perceive zoomorphic objects as animals like humans do?
Conclusion
Explicit training with zoomorphic objects can induce a human-like 'Animal bias' in deep neural networks.
Supporting Evidence
- Deep neural networks can mimic human perception of zoomorphic objects when trained appropriately.
- The study highlights the importance of training data in shaping visual perception in both humans and machines.
- Findings suggest that the human visual system's biases can be replicated in artificial intelligence.
Takeaway
This study shows that when computers are trained to recognize animal-like objects, they can learn to see them like humans do.
Methodology
The study used deep neural networks trained on different datasets to analyze how zoomorphic objects are perceived.
Potential Biases
Potential biases in training data could affect the results.
Limitations
The findings may not generalize to all types of objects or visual processing tasks.
Statistical Information
P-Value
p<0.001
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website