The human brain can achieve a remarkable ability to recognize visual patterns in an Invariant, selective and fast manner. The human visual system is quite powerful. It has an exquisite selectivity that allows us to distinguish among very similar objects. In addition, iy also allows Robustness (Invariance) to transformations.

In the model (Ventral stream pathway and architecture proposed by Poggio's group) shown in the image below, why the model is selective in S layers and invariant in C layers ? I didn't understand well why the S layer achieves selectivity and C layer achieves invariance. Please can someone explain to me in details why the model is selective in S and Invariance in C layers?

And who know the HMAX model, the model is called "Hierarchical Model and X". why X? what was meant by "X" ?

enter image description here This figure is from Serre et al.'s A quantitative theory of immediate visual recognition. Prog Brain Res. 2007.

Any help will be very very appreciated.

  • $\begingroup$ Here's a free version of the paper. I'm still working through it, myself. My impression is that you basically can always rely on objects being made up of edges with orientations, so they're fixed (invariant) inputs, but each object is an emergent combination of the fixed inputs, so they need selectivity to distinguish particular emergent shapes from other emergent shapes at higher abstraction levels, but still be able to recognize the object in different perspectives or configurations so S can't be too invariant. $\endgroup$ Jan 1, 2014 at 21:49
  • $\begingroup$ Thank you very much for your explanation. Are you working in HMAX model ? if yes, so I can help you :) $\endgroup$
    – Christina
    Jan 1, 2014 at 23:10
  • $\begingroup$ No, it was just an interesting question so I wanted to read about it. Not finding a lot of time lately though. $\endgroup$ Jan 3, 2014 at 4:43
  • $\begingroup$ You can see my response below :) $\endgroup$
    – Christina
    Jan 3, 2014 at 17:33

1 Answer 1


The question is resolved :

The name HMAX ( Hierarchical Model And X ) was coined by “Mike Tarr”, who wrote the “News and Views” accompanying the paper in Nature Neuroscience.

What was meant by Hierarchical?

The model has a hierarchical architecture : it contains different stages (layers).

-- Increase in receptive field size and complexity in unit feature preferences along the

What was meant by X?

I think it is the pooling operation (it refers to the MAX).

Selectivity in S layer

The activity of each S unit is highest if all the afferents have the specified values.

Invariance in C layer

In contrast to S units, the C cells pool over their afferents, and the response is high if one of the afferents is high. This increases invariance if the afferents are tuned to the same feature, but at different positions or scales.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.