decoded neurofeedback
we use decoded neurofeedback (DecNef) to cause human subjects to mainpulate their own brain activity totally unconsciously, thereby causing “inception-like” implantation of ideas. DecNef pairs machine learning decoding of multivoxel BOLD signals (blood flow in the brain) to identify and induce patterns of activity associated with various neural representations. current projects aim to use DecNef to examine and improve connectivity among brain regions that are involved in perception and the computation of uncertainty. we couple this with development of new connectivity measures based on mutual information and probability theory. these projects are done in cooperation with researchers in the Decoded Neurofeedback Department at ATR in Kyoto, Japan.
neural networks
we use leaky competing integration neural network models to examine suboptimalities in uncertainty computations. while much research has examined how the brian processes probabilistic information to arrive at perceptual inferences, less is known about the explicit computation of uncertainty in this process. through neural network modeling we aim to uncover biologicially plausible mechanisms that cause the patterns of uncertainty computation observed in human and animal models.
we also use deep learning and other “brain-inspired” (ha) neural networks to study information processing in the brain. Mehdi and Fan are working on projects using these techniques right now.
neuroimaging
we use functional magnetic resonance imaging (fMRI) and intracranial electrocorticography (ECoG) to access the complex, multivariate representations of perceptual variables and uncertainty in all areas of human cortex. future work will add EEG to this set of neuroimaging approaches.
we recently used ECoG to look at the computational and neuroanatomical correlates of perceptual decisions (inferences) versus uncertainty about those decisions (confidence/metacognition). we used machine learning and computational modeling to decode the information from electrodes placed directly on the surface of the cortex in humans. we found that, in humans, perceptual decisions optimally rely on information supporting all possible choices, but that uncertainty overly relies on evidence supporting whatever decision was made, i.e. a “confirmation bias” even in perception. this project is published in Nature Human Behaviour; here’s a pdf (for personal use only) because some universities don’t have access to NHB.
psychophysics + computational models
we try to understand the decision policies and inference processes the brain uses to transform noisy, incomplete incoming information into meaningful decisions and ultimately actions. previously we have used this approach to study whether the computations underlying metacognition are optimal (they probably aren’t, at least using the traditional cost functions). this research has been published in eLife, Cortex, and some other places. right now we are trying to develop tools and paradigms to use this approach in studying the function of consciousness, which policies describe metacognitive evaluations, and how these elements may change with certain diseases or disorders.
brain stimulation
transcranial direct current stimulation (tDCS) applies a weak electrical current to target areas of the brain, while transcranial magnetic stimulation (TMS) uses powerful electromagnets to manipulate neural activity. these techniques have been used to study perception, decision making, and problem solving, and have clinical applications as well. our work looks at how these techniques can modulate conscious perception and metacognition by understanding the neural consequences of stimulation through computational modeling.
we’ve recently collaborated with Tony Ro to use TMS to look for blindsight in normal observers. We found that TMS to visual cortex doesn’t actually make a target invisible (as lots of people may think), but it does induce suboptimal introspection/metacognition that might be akin to real, neurological cases of blindsight. Projects with noninvasive brain stimulation are ongoing. This paper is published in Cortex.