At long last, the pre-print to our MEG study + RIFT study and the final paper from my Ph.D with @olejensen.bsky.social We show that strong pre-search alpha oscillations are associated with faster responses in visual search www.biorxiv.org/content/10.1...
@thechbh.bsky.social #neuroskyence

statistically significant :) @camelphat.bsky.social

When presented with two stimuli simultaneously, the refractory dynamics result in the read-out alternating between the two letters (left). Adding inhibitory oscillations results in a phase code, whereby the inputs are sorted based on their salience (right).

We then introduced oscillatory dynamics based on refraction and pulses of inhibition (akin to alpha in visual cortex) into the network. The dynamics in the response to a single letter (A) show how the inhibitory oscillations clean up the dynamics (compare left & right panel).

We first trained a 2-layer NN to classify 3 letters presented in 4 quadrants. As expected, when presented with two stimuli simultaneously, the network produces a mixed representation of both stimuli. This indicates a computational bottleneck.

The notion that object recognition underlies serial processes goes back to early ideas by Broadbent (1953). The brain must find a way to segregate representations of simultaneous stimuli. One idea is that inhibitory oscillations organise these representations as a phase code.

We integrated refractory dynamics and pulses of inhibition into the hidden layers of an ANN & show that oscillations enable multiplexing of simultaneous stimuli. Proof-of-concept paper with Marco Idiart, Marcel van Gerven & @olejensen.bsky.social www.biorxiv.org/content/10.1...