Incorporating cycle-consistency loss into a generative adversarial network creates a high-performance and robust 'aligner' of neural population activity, permitting a fixed intracortical brain computer interface to be used for months without recalibration and without requiring inference of a latent manifold.

DeCalciOn is a low-cost open-source hardware system for real-time in vivo calcium imaging that offers capabilities for online decoding of neural population activity and delivery of short latency closed-loop feedback in freely behaving animals.

Single-peaked tuning curves found in early sensory areas are more optimized for quick decoding than accuracy, while multi-peaked tuning curves (e.g. grid cells) give higher accuracy but only at longer time scales.

I-Spin enables researchers to decode motor unit firings from surface and intramuscular electromyographic signals in real time.

A brain–computer interface for real-time identification of transient neural activity patterns enables causal inference of the role of these patterns in cognition through closed-loop manipulation.

Declined coding of sensory information in the brain reflects human vigilance decrements and allows prediction of behavioural errors.

Time-resolved decoding of EEG reveals the nature of representations of upcoming actions and how the planned action is selected from working memory.

The dynamics of neural responses during visual perception are best explained by a joint feedforward and recurrent architecture, which both maintains and broadcasts input features over time.

Electrophysiological and pharmacological data provide evidence that the medial prefrontal cortex is essential for rats to learn a timing task, becoming unnecessary later, while the dorsal striatum does not play a role initially, taking over when animals become proficient.