Bio-inspired divisive normalization improves object recognition performance in ANNs

Veerabadran, Vijay; Raina, Ritik; de, Virginia R.

doi:10.1167/jov.22.14.3592

Cited by 4 publications

(9 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These results demonstrate how the connectivity of inhibition balanced networks is shaped by their input statistics and explain the experience-dependent formation of extra-classical receptive fields. [70][71][72][73][74] Unlike previous models, [75][76][77][78][79] our networks are composed of excitatory and inhibitory neurons with fully plastic recurrent connectivity.…”

Section: Discussionmentioning

confidence: 99%

Synapse-type-specific competitive Hebbian learning forms functional recurrent networks

Eckmann

Gjorgjieva

2022

Preprint

View full text Add to dashboard Cite

Cortical networks exhibit complex stimulus-response patterns. Previous work has identified the balance between excitatory and inhibitory currents as a central component of cortical computations, but has not considered how the required synaptic connectivity emerges from biologically plausible plasticity rules. Using theory and modeling, we demonstrate how a wide range of cortical response properties can arise from Hebbian learning that is stabilized by the synapse-type-specific competition for synaptic resources. In fully plastic recurrent circuits, this competition enables the development and decorrelation of inhibition-balanced receptive fields. Networks develop an assembly structure with stronger connections between similarly tuned neurons and exhibit response normalization and surround suppression. These results demonstrate how neurons can self-organize into functional circuits and provide a foundational understanding of plasticity in recurrent networks.

show abstract

Section: Discussionmentioning

confidence: 99%

Synapse-type-specific competitive Hebbian learning forms functional recurrent networks

Eckmann

Gjorgjieva

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Surround suppression has been considered to have a number of beneficial roles in neural computation, for example, reducing coding redundancy and yielding more efficient neural codes [12, 14, 19, 21, 22, 24, 25, 27, 60, 61]. Some studies in machine learning noticed the lack of more sophisticated forms of brain-like divisive normalization in generic feedforward CNNs, and tried to integrate them into the network [47–51]. These studies found that incorporating divisive normalization in CNNs improves image classification in some limited cases, such as when the network is more shallow [49], when the dataset requires strong center-surround separation [49], or when the divisive normalization is combined with batch normalization [50].…”

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

“…Our findings extend the ability of generic CNNs as models of visual cortices. The mismatches we found may inspire future studies of contextual effects in deep neural networks with more sophisticated circuitry, including the role of divisive normalization [47][48][49][50][51]64], recurrent connections and feedback [65][66][67][68][69][70] Dense 1000-Softmax Table 1. CNN architectures used in this study.…”

Section: Discussionmentioning

confidence: 99%

“…The visualization method reveals a generalization of the idea of homogeneity to complex stimuli and provides a new experimental scheme that can be used in biological experiments. We also found mismatches to the biology, highlighting the limitations of the feedforward architectures, and identifying the need to further incorporate nonlinear computations and circuitry into deep neural networks [47][48][49][50][51][52].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Generalizing biological surround suppression based on center surround similarity via deep neural network models

DeForge

Schwartz

2023

Preprint

View full text Add to dashboard Cite

Sensory perception is dramatically influenced by the context. Models of contextual neural surround effects in vision have mostly accounted for Primary Visual Cortex (V1) data, via nonlinear computations such as divisive normalization. However, surround effects are not well understood within a hierarchy, for neurons with more complex stimulus selectivity beyond V1. We utilized feedforward deep neural networks and developed a gradient-based technique to visualize the most suppressive and excitatory surround. We found that deep neural networks exhibited a key signature of surround effects in V1, highlighting center stimuli that visually stand out from the surround and suppressing responses when the surround stimulus is similar to the center. Even when the center stimulus was altered, the most suppressive surround surprisingly followed. This ties to notions of efficient coding and salience perception, although the networks were trained to classify images. Through the visualization approach, we generalized previous understanding of surround effects to more complex stimuli, in ways that have not been revealed in visual cortices. Our results emerged without specialized nonlinear computations, but due to subtraction and the stacking of layers. We identified further successes including V2 surround data for textures that cannot be explained by divisive normalization models, along with mismatches to the biology that could not be explained by the feedforward deep neural networks. Our results provide a testable hypothesis of surround effects in higher visual cortices, and the visualization approach could be adopted in future biological experimental designs.

show abstract

Generalizing biological surround suppression based on center surround similarity via deep neural network models

Pan,

DeForge,

Schwartz

2023

PLoS Comput Biol

View full text Add to dashboard Cite

Sensory perception is dramatically influenced by the context. Models of contextual neural surround effects in vision have mostly accounted for Primary Visual Cortex (V1) data, via nonlinear computations such as divisive normalization. However, surround effects are not well understood within a hierarchy, for neurons with more complex stimulus selectivity beyond V1. We utilized feedforward deep convolutional neural networks and developed a gradient-based technique to visualize the most suppressive and excitatory surround. We found that deep neural networks exhibited a key signature of surround effects in V1, highlighting center stimuli that visually stand out from the surround and suppressing responses when the surround stimulus is similar to the center. We found that in some neurons, especially in late layers, when the center stimulus was altered, the most suppressive surround surprisingly can follow the change. Through the visualization approach, we generalized previous understanding of surround effects to more complex stimuli, in ways that have not been revealed in visual cortices. In contrast, the suppression based on center surround similarity was not observed in an untrained network. We identified further successes and mismatches of the feedforward CNNs to the biology. Our results provide a testable hypothesis of surround effects in higher visual cortices, and the visualization approach could be adopted in future biological experimental designs.

show abstract

Bio-inspired divisive normalization improves object recognition performance in ANNs

Cited by 4 publications

References 0 publications

Synapse-type-specific competitive Hebbian learning forms functional recurrent networks

Synapse-type-specific competitive Hebbian learning forms functional recurrent networks

Generalizing biological surround suppression based on center surround similarity via deep neural network models

Generalizing biological surround suppression based on center surround similarity via deep neural network models

Contact Info

Product

Resources

About