Biased competition model

Lens adapted to this topic: What the model explains and the evidence supporting it

The mainstream account treats biased competition as a framework for selective attention in crowded visual scenes. Visual representations compete under limited processing capacity, while bottom-up salience and top-down goals or attention bias the competition. Neural, behavioral, and computational work uses the framework to explain selection and priority effects, including changes associated with arousal.

• Visual objects compete for limited cortical representation and processing, so selecting one object can reduce the relative processing of others.
• Competition is biased by both bottom-up properties such as salience and top-down factors such as attention, goals, and behavioral relevance.
• Computational analysis indicates that feedforward and feedback connections can implement biased competition when top-down attentional bias reaches suitable parameter ranges.
• The arousal-biased competition model predicts enhanced processing of salient stimuli alongside impaired processing of relatively less-salient stimuli; experiments tested this prediction using visual stimuli and arousing tones.

Lens adapted to this topic: Competing models and limits of a competition-only account

Dissenting interpretations do not necessarily reject attentional competition, but question whether it should be treated as a unique or sufficient theory of cortical function. One analysis argues that a particular biased-competition implementation is mathematically equivalent to linear predictive coding. Another examines cooperation as well as competition, suggesting that interactions among cortical areas may not be captured by a competition-only framing.

• A particular implementation of biased competition, where inhibition targets inputs to a cortical region, is mathematically equivalent to linear predictive coding.
• Because predictive coding and biased competition can share the same mathematical structure, their differences may be smaller than their labels suggest.
• A computational analysis explicitly studies both competition and cooperation among cortical areas, challenging an interpretation focused only on rivalry between representations.
• The model’s explanatory scope may depend on implementation details, including the strength of top-down bias and the network parameters that determine whether selection occurs.