Deep supervised, but not unsupervised, models may explain IT cortical representation.
Khaligh-Razavi SM, Kriegeskorte N. · 2014
Read Deep supervised, but not unsupervised, models may explain IT cortical representation. by Khaligh-Razavi SM, Kriegeskorte N. (2014) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc4222664.txt.
Description
Seyed-Mahdi Khaligh-Razavi and Nikolaus Kriegeskorte tested computer vision models against the brain in PLoS Computational Biology in 2014. The inferior temporal cortex is where primates recognize objects, and its activity patterns cluster objects by kind. Many computer models claim to mimic that area, with little direct comparison.
The team compared a shelf of models, from classic neuroscience designs to a deep neural network, against brain data from human scans and monkey cell recordings for the same images. Models that categorized images better also matched the brain's pattern geometry better. Unsupervised models, though, missed the brain's strong clustering by category. The deep network, trained with supervision on over a million labelled images, came closest to the brain while still falling short, hinting that category training shapes brain-like representations.
Catalog record
| CREATOR | Khaligh-Razavi SM, Kriegeskorte N. |
|---|---|
| YEAR | 2014 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 130 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |