Active sensing with predictive coding and uncertainty minimization.

We present an end-to-end architecture for embodied exploration inspired by two biological computations: predictive coding and uncertainty minimization. The architecture can be applied to any exploration setting in a task-independent and intrinsically driven manner. We first demonstrate our approach in a maze navigation task and show that it can discover the underlying transition distributions and spatial features of the environment. Second, we apply our model to a more complex active vision task, whereby an agent actively samples its visual environment to gather information. We show that our model builds unsupervised representations through exploration that allow it to efficiently categorize visual scenes. We further show that using these representations for downstream classification leads to superior data efficiency and learning speed compared to other baselines while maintaining lower parameter complexity. Finally, the modular structure of our model facilitates interpretability, allowing us to probe its internal mechanisms and representations during exploration.

Changes in Local Network Activity Approximated by Reverse Spike-Triggered Local Field Potentials Predict the Focus of Attention.

The effects of visual spatial attention on neuronal firing rates have been well characterized for neurons throughout the visual processing hierarchy. Interestingly, the mechanisms by which attention generates more or fewer spikes in response to a visual stimulus remain unknown. One possibility is that attention boosts the likelihood that synaptic inputs to a neuron result in spikes. We performed a novel analysis to measure local field potentials (LFPs) just prior to spikes, or reverse spike-triggered LFP "wavelets," for neurons recorded in primary visual cortex (V1) of monkeys performing a contrast change detection task requiring covert shifts in visual spatial attention. We used dimensionality reduction to define LFP wavelet shapes with single numerical values, and we found that LFP wavelet shape changes correlated with changes in neuronal firing rate. We then tested whether a simple classifier could predict monkeys' focus of attention from LFP wavelet shape. LFP wavelet shapes sampled in discrete windows were predictive of the locus of attention for some neuronal types. These findings suggest that LFP wavelets are a useful proxy for local network activity influencing spike generation, and changes in LFP wavelet shape are predictive of the focus of attention.