GASPnet: Global Agreement to Synchronize Phases
This work addresses the visual binding problem for neural networks, offering a novel approach that combines neuroscience-inspired mechanisms with machine learning, though it appears incremental as it builds on prior work on global routing by agreement.
The authors tackled the visual binding problem in neural networks by proposing GASPnet, which incorporates angular phases and phase alignment via Kuramoto dynamics to enhance operations between neurons with similar phases. The results showed better accuracy than CNN networks on datasets with superimposed images, with improved robustness to noise and generalization abilities.
In recent years, Transformer architectures have revolutionized most fields of artificial intelligence, relying on an attentional mechanism based on the agreement between keys and queries to select and route information in the network. In previous work, we introduced a novel, brain-inspired architecture that leverages a similar implementation to achieve a global 'routing by agreement' mechanism. Such a system modulates the network's activity by matching each neuron's key with a single global query, pooled across the entire network. Acting as a global attentional system, this mechanism improves noise robustness over baseline levels but is insufficient for multi-classification tasks. Here, we improve on this work by proposing a novel mechanism that combines aspects of the Transformer attentional operations with a compelling neuroscience theory, namely, binding by synchrony. This theory proposes that the brain binds together features by synchronizing the temporal activity of neurons encoding those features. This allows the binding of features from the same object while efficiently disentangling those from distinct objects. We drew inspiration from this theory and incorporated angular phases into all layers of a convolutional network. After achieving phase alignment via Kuramoto dynamics, we use this approach to enhance operations between neurons with similar phases and suppresses those with opposite phases. We test the benefits of this mechanism on two datasets: one composed of pairs of digits and one composed of a combination of an MNIST item superimposed on a CIFAR-10 image. Our results reveal better accuracy than CNN networks, proving more robust to noise and with better generalization abilities. Overall, we propose a novel mechanism that addresses the visual binding problem in neural networks by leveraging the synergy between neuroscience and machine learning.