REVIEW

published: 23 February 2018

doi: 10.3389/fnhum.2018.00061

Computational Neuropsychology
and Bayesian Inference
Thomas Parr 1* , Geraint Rees 1,2 and Karl J. Friston 1
1
Wellcome Trust Centre for Neuroimaging, Institute of Neurology, University College London, London, United Kingdom,
2
Institute of Cognitive Neuroscience, University College London, London, United Kingdom

Computational theories of brain function have become very influential in neuroscience.

They have facilitated the growth of formal approaches to disease, particularly in
psychiatric research. In this paper, we provide a narrative review of the body of
computational research addressing neuropsychological syndromes, and focus on
those that employ Bayesian frameworks. Bayesian approaches to understanding brain
function formulate perception and action as inferential processes. These inferences
combine ‘prior’ beliefs with a generative (predictive) model to explain the causes of
sensations. Under this view, neuropsychological deficits can be thought of as false
inferences that arise due to aberrant prior beliefs (that are poor fits to the real world).
This draws upon the notion of a Bayes optimal pathology – optimal inference with
suboptimal priors – and provides a means for computational phenotyping. In principle,
any given neuropsychological disorder could be characterized by the set of prior beliefs
that would make a patient’s behavior appear Bayes optimal. We start with an overview
of some key theoretical constructs and use these to motivate a form of computational
neuropsychology that relates anatomical structures in the brain to the computations they
perform. Throughout, we draw upon computational accounts of neuropsychological
syndromes. These are selected to emphasize the key features of a Bayesian approach,
Edited by:
and the possible types of pathological prior that may be present. They range from
Hauke R. Heekeren,
Freie Universität Berlin, Germany visual neglect through hallucinations to autism. Through these illustrative examples, we
Reviewed by: review the use of Bayesian approaches to understand the link between biology and
Francisco Barceló, computation that is at the heart of neuropsychology.
University of the Balearic Islands,
Spain Keywords: neuropsychology, active inference, predictive coding, computational phenotyping, precision
Gianfranco Spalletta,
Fondazione Santa Lucia (IRCCS), Italy
*Correspondence: INTRODUCTION
Thomas Parr
[email protected] The process of relating brain dysfunction to cognitive and behavioral deficits is complex.
Traditional lesion-deficit mapping has been vital in the development of modern neuropsychology
Received: 22 November 2017 but is confounded by several problems (Bates et al., 2003). The first is that there are statistical
Accepted: 05 February 2018
dependencies between lesions in different regions (Mah et al., 2014). These arise from, for example,
Published: 23 February 2018
the vascular anatomy of the brain. Such dependencies mean that regions commonly involved in
Citation:
stroke may be spuriously associated with a behavioral deficit (Husain and Nachev, 2007). The
Parr T, Rees G and Friston KJ (2018)
Computational Neuropsychology and
problem is further complicated by the distributed nature of brain networks (Valdez et al., 2015).
Bayesian Inference. Damage to one part of the brain may give rise to abnormal cognition indirectly – through its
Front. Hum. Neurosci. 12:61. influence over a distant region (Price et al., 2001; Carrera and Tononi, 2014). An understanding of
doi: 10.3389/fnhum.2018.00061 the contribution of a brain region to the network it participates in is crucial in forming an account

Frontiers in Human Neuroscience | www.frontiersin.org 1 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

of functional diaschisis of this form (Boes et al., 2015; Fornito phrased in terms of two questions: ‘what are the prior beliefs that
et al., 2015). Solutions that have been proposed to the above would have to be held to make this behavior optimal?’ and ‘what
problems include the use of multivariate methods (Karnath and are the biological substrates of these priors?’
Smith, 2014; Nachev, 2015) to account for dependencies, and the The notion of optimal pathology may seem counter-intuitive,
use of models of effective connectivity to assess network-level but we can draw upon another theorem, the good regulator
changes (Rocca et al., 2007; Grefkes et al., 2008; Abutalebi et al., theorem (Conant and Ashby, 1970), to highlight the difference
2009; Mintzopoulos et al., 2009) in response to lesions. between healthy and pathological behavior. This states that a
In this paper, we consider a complementary approach that brain (or any other system) is only able to effectively regulate its
has started to gain traction in psychiatric research (Adams environment if it is a good model of that environment. A brain
et al., 2013b, 2015; Corlett and Fletcher, 2014; Huys et al., 2016; that embodies a model with priors that diverge substantially
Schwartenbeck and Friston, 2016; Friston et al., 2017c). This from the world (i.e., body, ecological niche, culture, etc.) it is
is the use of models that relate the computations performed trying to regulate will fail at this task (Schwartenbeck et al.,
by the brain to measurable behaviors (Krakauer and Shadmehr, 2015). If pathological priors relate to the properties of the
2007; Mirza et al., 2016; Testolin and Zorzi, 2016; Iglesias et al., musculoskeletal system, we might expect motor disorders such
2017). Such models can be associated with process theories as tremors or paralysis (Friston et al., 2010; Adams et al., 2013a).
(Friston et al., 2017a) that map to neuroanatomy and physiology. If abnormal priors relate to perceptual systems, the results may
This complements the approaches outlined above, as it allows include sensory hallucinations (Fletcher and Frith, 2009; Adams
focal neuroanatomical lesions to be interpreted in terms of et al., 2013b) or anesthesia. In the following, we review some
their contribution to a network. Crucially, this approach ensures important concepts in Bayesian accounts of brain function.
that the relationship between brain structure and function is These include the notion of a generative model, the hierarchical
addressed within a conceptually rigorous framework – this is structure of such models, the representation of uncertainty in
essential for the construction of well-formed hypotheses for the brain, and the active nature of sensory perception. In doing
neuropsychological research (Nachev and Hacker, 2014). We so we will develop a taxonomy of pathological priors. While
focus here upon models that employ a conceptual framework this taxonomy concerns types of inferential deficit (and is not
based on Bayesian inference. a comprehensive review of neuropsychological syndromes), we
Bayesian inference is the process of forming beliefs about the draw upon examples of syndromes to illustrate these pathologies.
causes of sensory data. It relies upon the combination of prior We relate these to failures of neuromodulation and to the notion
beliefs about these causes, and beliefs about how these causes give of a ‘disconnection’ syndrome (Geschwind, 1965a; Catani and
rise to sensations. Using these two probabilities it is possible to Ffytche, 2005).
calculate the probability, given a sensation, of its cause. This is
known as a ‘posterior’ probability. This means that prior beliefs
are updated by a sensory experience to become posterior beliefs. THE GENERATIVE MODEL
These posteriors can then be used as the prior for the next
sensory experience. In short, Bayesian theories of brain function Bayesian Inference
propose the brain encodes beliefs about the causes of sensory Much work in theoretical neurobiology rests on the notion that
data, and that these beliefs are updated in response to new sensory the brain performs Bayesian inference (Knill and Pouget, 2004;
evidence. Doya, 2007; Friston, 2010; O’Reilly et al., 2012). In other words,
Our motivation for pursuing a Bayesian framework is that it the brain makes inferences about the (hidden or latent) causes
captures many different types of behavior, including apparently of sensory data. ‘Hidden’ variables are those that are not directly
suboptimal behaviors. According to an important result known observable and must be inferred. For example, the position
as the complete class theorem (Wald, 1947; Daunizeau et al., 2010), (hidden variable) of a lamp causes a pattern of photoreceptor
there is always a set of a prior belief that renders an observed activation (sensory data) in the retina. Bayesian inference can be
behavior Bayes optimal. This is fundamental for computational used to infer the probable position of the lamp from the retinal
neuropsychology as it means we can cast even pathological data. To do this, two probability distributions must be defined
behaviors as the result of processes that implement Bayesian (these are illustrated graphically in Figure 1A). These are the
inference (Schwartenbeck et al., 2015). In other words, we can prior probability of the causes, and a likelihood distribution that
assume that the brain makes use of a probabilistic model of its determines how the causes give rise to sensory data. Together,
environment to make inferences about the causes of sensory data these are referred to as a ‘generative model,’ as they describe the
(Knill and Pouget, 2004; Doya, 2007), and to act upon them processes by which data is (believed to be) generated. Bayesian
(Friston et al., 2012b). Another consequence of the theorem is inference uses a generative model to compute the probable causes
that computational models that are not (explicitly) motivated by of sensory data (Beal, 2003; Doya, 2007; Ghahramani, 2015).
Bayesian inference (Frank et al., 2004; O’Reilly, 2006) may be Many of the inferences that must be made by the brain relate to
written down in terms of Bayesian decision processes. Working causes that evolve through time. This means that the prior over
within this framework facilitates communication between these the trajectory of causes through time can be decomposed into a
models, and ensures they could all be used to phenotype patients prior for the initial state, and a series of transition probabilities
using a common currency (i.e., their prior beliefs). It follows that that account for sequences or dynamics (Figure 1B). These
the key challenges for computational neuropsychology can be dynamics can be subdivided into those that a subject has control

Frontiers in Human Neuroscience | www.frontiersin.org 2 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

FIGURE 1 | Generative models. These schematics graphically illustrate the structure of generative models. (A) The simplest model that permits Bayesian inference
involves a hidden state, s, that is equipped with a prior P(s). This hidden state generates observable data, o, through a process defined by the likelihood P(o|s)
(vertical arrow). (B) It is possible to equip such a model with dynamically changing hidden states. To do so, we must specify the probabilities of transitioning between
states P sτ+1 |sτ (horizontal arrows). (C) Transitions between states may be influenced by the course of action, π, that is pursued. (D) Hierarchical levels can be

added to the generative model (Friston et al., 2017d). This means that the processes that generate the hidden states can themselves be accommodated in the
inferences performed using the model.

over (Figure 1C), such as muscle length, and environmental and Howard, 1999). This phenomenon, known as Charles Bonnet
causes that they cannot directly influence. syndrome (Teunisse et al., 1996; Menon et al., 2003), can be
interpreted as a failure to constrain perceptual hypotheses with
Predictive Coding sensations (Reichert et al., 2013). In other words, there are
Predictive coding is a prominent theory describing how the no prediction errors to correct predictions. A similar line of
brain could perform Bayesian inference (Rao and Ballard, 1999; argument can be applied to phantom limbs (Frith et al., 2000;
Friston and Kiebel, 2009; Bastos et al., 2012). This relies upon the De Ridder et al., 2014). Following amputation, patients may
idea that the brain uses its generative model to form perceptual continue to experience ‘phantom’ sensory percepts from their
hypotheses (Gregory, 1980) and make predictions about sensory missing limb. The absence of corrective signals from amputated
data. The difference between this prediction and the incoming body parts means that any hypothesis held about the limb is
data is computed, and the ensuing prediction error is used unfalsifiable. In the next sections, we consider some of the
to refine hypotheses about the cause of the data. Under this important features of generative models, and their relationship
theory, the messages passed through neuronal signaling are to brain function.
either predictions, or prediction errors. There are other local
message passing schemes that can implement Bayesian inference
(Winn and Bishop, 2005; Yedidia et al., 2005; Dauwels, 2007; HIERARCHICAL MODELS
Friston et al., 2017b), particularly for categorical (as opposed
to continuous) inferences. Although we use the language of Cortical Architecture
predictive coding in the following, we note that our discussion An important feature of many generative models is hierarchy.
generalizes to other Bayesian belief propagation schemes. Hierarchical models assume that the hidden causes that generate
The notion that hypotheses are corrected by prediction errors sensory data are themselves generated from hidden causes at
makes sense of the kinds of neuropsychological pathologies that a higher level in the hierarchy (Figure 1D). As the hierarchy
result from the loss of sensory signals. For example, patients with is ascended, causes tend to become more abstract, and have
eye disease can experience complex visual hallucinations (Ffytche dynamics that play out over a longer time course (Kiebel et al.,

Frontiers in Human Neuroscience | www.frontiersin.org 3 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

2008, 2009). An intuitive example is the kind of generative

model required for reading (Friston et al., 2017d). While lower
levels may represent letters, higher levels represent words, then
sentences, then paragraphs.
There are several converging lines of evidence pointing to the
importance of hierarchy as a feature of brain organization. One
of these is the patterns of receptive fields in the cortex (Gallant
et al., 1993). In primary sensory cortices, cells tend to respond
to simple features such as oriented lines (Hubel and Wiesel,
1959). As we move further from sensory cortices, the complexity
of the stimulus required to elicit a response increases. Higher
areas become selective for contours (Desimone et al., 1985; von
der Heydt and Peterhans, 1989), shapes, and eventually objects
(Valdez et al., 2015). The sizes of receptive fields also increase
(Gross et al., 1972; Smith et al., 2001).
A second line of evidence is the change in temporal response
properties. Higher areas appear to respond to stimuli that
change over longer time courses than lower areas (Hasson et al.,
2008, 2015; Kiebel et al., 2008; Murray et al., 2014). This is FIGURE 2 | Hierarchy in the cortex. This schematic illustrates two key
features of cortical organization. The first is hierarchy, as defined by laminar
consistent with the structure of deep temporal generative models
specific projections. Projections from primary sensory areas, such as area VI,
(Friston et al., 2017d) (a sentence takes longer to read than to higher cortical areas typically arise from layer III of a cortical column, and
a word). A third line of evidence is the laminar specificity of target layer IV. These ascending connections are shown in red. In contrast,
inter-areal connections that corroborates the pattern implied by descending connections (in blue) originate in deep layers of the cortex and
electrophysiological responses (Felleman and Van Essen, 1991; project to both superficial and deep laminae. The second feature illustrated
here is the separation of visual processing into two, dorsal and ventral,
Shipp, 2007; Markov et al., 2013). As illustrated in Figure 2, streams. In terms of the functional anatomy implied by generative models in
cortical regions lower in the hierarchy project to layer IV of the brain, this segregation implies a factorization of beliefs about the location
the cortex in higher areas. These ‘ascending’ connections arise and identity of a visual object (i.e., knowing what an object is does not tell you
from layer III of the lower hierarchical region. ‘Descending’ where it is – and vice versa).
connections typically arise from deep layers of the cortex, and
target both deep and superficial layers of the cortical area lower
in the hierarchy. objects, can present with semantic deficits at different levels of
abstraction. For example, some neurological patients are able
Ascending and Descending Messages to distinguish between broad categories (fruits or vegetables)
The parallel between the hierarchical structure of generative but are unable to identify particular objects within a category
models and that of cortical organization has an interesting (Warrington, 1975). The preservation of the more abstract
consequence. It suggests that connections between cortical knowledge, with impairment of within-category semantics, is
regions at different hierarchical levels are the neurobiological taken as evidence for distinct hierarchical levels that can be
substrate of the likelihoods that map hidden causes to the sensory differentially impaired. This is endorsed by findings that some
data, or lower level causes, that they generate (Kiebel et al., 2008; patients have a category-specific agnosia (for example, a failure
Friston et al., 2017d). This is very important in understanding the to identify living but not inanimate stimuli) (Warrington and
computational nature of a ‘disconnection’ syndrome. It implies Shallice, 1984). A model that simulates these deficits relies
that the disruption of a white matter pathway corresponds upon a hierarchical structure that allows for specific categorical
to an abnormal prior belief about the form of the likelihood processing at higher levels to be lesioned while maintaining lower
distribution. This immediately allows us to think of neurological level processes (Humphreys and Forde, 2001). Notably, lesions
disconnection syndromes – such as visual agnosia, pure alexia, to this model were performed by modulating the connections
apraxia, and conduction aphasia (Catani and Ffytche, 2005) – in between hierarchical levels. This resonates well with the type of
probabilistic terms. We will address specific examples of these in computational ‘disconnection’ that predictive coding implicates
the next section, and a summary is presented in Table 1. Under in some psychiatric disorders (Friston et al., 2016a). We now turn
predictive coding, the signals carried by inter-areal connections to the probabilistic interpretation of such disconnections.
have a clear interpretation (Shipp et al., 2013; Shipp, 2016).
Descending connections carry the predictions derived from the
generative model about the causes or data at the lower level. SENSORY STREAMS AND
Ascending connections carry prediction error signals. DISCONNECTION SYNDROMES
It has been argued that deficits in semantic knowledge can
only be interpreted with reference to a hierarchically organized What and where?
set of representations in the brain. This argument rests on Figure 2 illustrates an additional feature common to cortical
observations that patients with agnosia, a failure to recognize architectures and inference methods. This is the factorization

Frontiers in Human Neuroscience | www.frontiersin.org 4 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

TABLE 1 | Bayesian computational neuropsychology.

Syndrome Abnormal prior Neurobiology Reference

Anosognosia Low exteroceptive or interoceptive Insula lesions Karnath et al., 2005; Fotopoulou
sensory precision et al., 2010; Fotopoulou, 2012
Failure of active inference Hemiplegia
Apraxia Disrupted likelihood (action to vision or Callosal disconnection Left Geschwind, 1965b
command to action consequences) frontoparietal disconnection
Autism High sensory precision ↑Cholinergic transmission? Dayan and Yu, 2006; Lawson et al.,
Secondary to high volatility ↑Noradrenergic transmission 2014; Marshall et al., 2016;
Lawson et al., 2017
Complex visual hallucinations Low sensory precision ↓Cholinergic transmission Collerton et al., 2005; Reichert
(Lewy body dementia, Charles Disrupted likelihood mapping Retino-geniculate disconnection et al., 2013; O’Callaghan et al.,
Bonnet syndrome) 2017
Conduction aphasia Disrupted likelihood mapping (speech Arcuate fasciculus disconnection Wernicke, 1969
to proprioceptive consequences)
Parkinson’s disease Low prior precision over policies ↓Dopaminergic transmission Friston et al., 2013
Visual agnosia Disrupted likelihood (‘what’ to Ventral visual stream disconnection Geschwind, 1965b
sensory data)
Visual neglect Disrupted likelihood (fixation to ‘what’) SLF II disconnection Karnath et al., 2002; Thiebaut de
mapping Schotten et al., 2005; Bartolomeo
Biased outcome prior Pulvinar lesion et al., 2007
Biased policy prior Putamen lesion

of beliefs about hidden causes into multiple streams. Bayesian that relate disorders of spatial processing to interhemispheric
inference often employs this device, known as a ‘mean-field’ (Kinsbourne, 1970) or intrahemispheric disruptions (Bartolomeo
assumption, which ‘carves’ posterior beliefs into the product et al., 2007; Bartolomeo, 2014). Any intrahemispheric lesion
of statistically independent factors (Beal, 2003; Friston and that induces a bias toward one side of space necessarily alters
Buzsáki, 2016). The factorization of visual hierarchies into the interhemispheric balance of activity (Parr and Friston,
ventral and dorsal ‘what’ and ‘where’ streams (Ungerleider and 2017b).
Mishkin, 1982; Ungerleider and Haxby, 1994) appears to be
an example of this. A closely related factorization separates the Disconnections and Likelihoods
dorsal and ventral attention networks (Corbetta and Shulman, The factorization of beliefs into distinct processing streams is not
2002). This factorization has important consequences for the limited to the visual system. Notably, theories of the neurobiology
representation of objects in space. Location is represented of speech propose a similar division into dorsal and ventral
bilaterally in the brain, with each side of space represented in streams (Hickok and Poeppel, 2007; Saur et al., 2008). The former
the contralateral hemisphere. As it is not necessary to know is thought to support articulatory components of speech, while
the location of an object to know its identity, it is possible the latter is involved in language comprehension. This mean-
to represent this information independently, and therefore field factorization accommodates the classical subdivision of
unilaterally (Parr and Friston, 2017a). It is notable that object aphasias into fluent (e.g., Wernicke’s aphasia) and non-fluent
recognition deficits tend to occur when patients experience (e.g., Broca’s aphasia) categories. The anatomy of these networks
damage to areas in the right hemisphere (Warrington and has been interpreted in terms of predictive coding (Hickok,
James, 1967, 1988; Warrington and Taylor, 1973). Lesions 2012a,b), and this interpretation allows us to illustrate the point
to contralateral (left hemispheric) homologs are more likely that disconnection syndromes are generally due to disruption
to give rise to difficulties in naming objects (Kirshner, of the likelihood mapping between two regions. We draw
2003). upon examples of aphasic and apraxic syndromes to make this
The bilateral representation of space has an important point.
consequence when we frame neuronal processing as probabilistic Conduction aphasia is the prototypical disconnection
inference. Following an inference that a stimulus is likely to syndrome (Wernicke, 1969), disconnecting Wernicke’s
be on one side of space, it must be the case that it is less area from Broca’s area. The former is found near the
likely to be on the contralateral side. If neuronal activities temporoparietal junction, and is thought to contribute to
in each hemisphere represent these probabilities, this induces language comprehension. The latter is in the inferior frontal lobe,
a form of interhemispheric competition (Vuilleumier et al., and is a key part of the dorsal language stream. Disconnection of
1996; Rushmore et al., 2006; Dietz et al., 2014). An important the two areas results in an inability to repeat spoken language.
role of commissural fiber pathways may be to enforce the This connection between these two areas, the arcuate fasciculus
normalization of probabilities across space [although some (Catani and Mesulam, 2008), could represent the likelihood
of these axons must represent likelihood mappings instead mapping from speech representations in Wernicke’s area to
(Glickstein and Berlucchi, 2008)]. This neatly unifies theories the articulatory proprioceptive data processed in Broca’s area

Frontiers in Human Neuroscience | www.frontiersin.org 5 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

1965b). Under this theory, deficits in imitation that accompany

this are due to disruption of axons that connect visual and
motor areas. These also travel in tracts from posterior to frontal
cortices.
Other disconnection syndromes include (Geschwind, 1965a;
Catani and Ffytche, 2005) visual agnosia, caused by disruption
of connections in the ventral visual stream, and visual neglect
(Doricchi and Tomaiuolo, 2003; Bartolomeo et al., 2007; He
et al., 2007; Ciaraffa et al., 2013). Neglect can be a consequence
FIGURE 3 | Dorsal and ventral streams. Here we depict a plausible mapping of frontoparietal disconnections (Figure 3, right), leading to an
of simple generative models to the dual streams of the language (Left) and impaired awareness of stimuli on the left despite intact early
attention (Right) networks. We highlight the likelihood mappings that visual processing (Rees et al., 2000). We consider the behavioral
correspond to white matter tracts implicated in disconnection syndromes. The manifestations of visual neglect in a later section. Before we do so,
number 1 in the blue circle on the left highlights the mapping from the left
temporoparietal region, which responds to spoken words (Howard et al.,
we turn from disconnections to a subtler form of computational
1992), to the inferior frontal gyrus, involved in the dorsal articulatory stream pathology.
(Hickok, 2012b). This region is well placed to deal with proprioceptive data
from the laryngeal and pharyngeal muscles (Simonyan and Horwitz, 2011).
The connection corresponds to the arcuate fasciculus and lesions give rise to UNCERTAINTY, PRECISION, AND
conduction aphasia. The number 2 indicates the mapping from dorsal frontal
regions that represent eye fixation locations to ventral regions associated with
AUTISM
target detection and identity. This corresponds to the second branch of the
superior longitudinal fasciculus. Lesions to this structure are implicated in Types of Uncertainty
visual neglect (Doricchi and Tomaiuolo, 2003; Thiebaut de Schotten et al., In predictive coding, the significance ascribed to a given
2005).
prediction error is determined by the precision of the mapping
from hidden causes to the data. If this mapping is very noisy,
the gain of the prediction error signal is turned down. A very
as in Figure 3 (left). While auditory data from the ventral precise relationship between causes and data leads to an increase
pathway may inform inferences about language, the failure to in this gain – it is this phenomenon that has been associated
translate these into proprioceptive predictions means that such with attention (Feldman and Friston, 2010). In other words,
predictions cannot be fulfilled by the brainstem motor system attention is the process of affording a greater weight to reliable
(Adams et al., 2013a). information.
The idea that a common generative model could generate both The generative models depicted in Figure 1 indicate that there
auditory and proprioceptive predictions, associated with speech, are multiple probability distributions that may be excessively
harmonizes well with theories of about the ‘mirror-neuron’ precise or imprecise (Parr and Friston, 2017c). One of these is
system (Di Pellegrino et al., 1992; Rizzolatti et al., 2001). These the sensory precision that relates to the likelihood. It is this that
neurons respond both to the performance of an action by weights sensory prediction errors in predictive coding (Friston
an individual, and when that individual observes the same and Kiebel, 2009; Feldman and Friston, 2010). Another source
action being performed by another. Similarly, Wernicke’s area of uncertainty relates to the dynamics of hidden causes. It
appears to be necessary for both language comprehension and may be that the mapping from the current hidden state to the
generation (Dronkers and Baldo, 2009) (but see Binder, 2015). next is very noisy, or volatile. Alternatively, these transitions
Anatomically, there is consistency between the mirror neuron may be very deterministic. A third source of uncertainty
system and the connectivity between the frontal and temporal relates to those states that a person has control over. It is
regions involved in speech. The former is often considered to possible for a person to hold beliefs about the course of
include Broca’s area and the superior temporal sulcus – adjacent action, or policy, that they will pursue with differing levels of
to Wernicke’s area (Frith and Frith, 1999; Keysers and Perrett, confidence.
2004). Beliefs about the degree of uncertainty in each of these
A common generative model for action observation and three distributions have been related to the transmission of
generation (Kilner et al., 2007) generalizes to include the notion acetylcholine (Dayan and Yu, 2001; Yu and Dayan, 2002;
of ‘conduction apraxia’ (Ochipa et al., 1994). As with conduction Moran et al., 2013), noradrenaline (Dayan and Yu, 2006), and
aphasia, this disorder involves a failure to repeat what another is dopamine (Friston et al., 2014) respectively (Marshall et al.,
doing. Instead of repeating spoken language, conduction apraxia 2016). The ascending neuromodulatory systems associated with
represents a deficit in mimicking motor behaviors. This implies these transmitters are depicted in Figure 4. The relationship
a disconnection between visual and motor regions (Goldenberg, between dopamine and the precision of prior beliefs about
2003; Catani and Ffytche, 2005). This must spare the route policies suggests that the difficulty initiating movements in
from language areas to motor areas. Other forms of apraxia Parkinson’s disease may be due to a high estimated uncertainty
have been considered to be disconnection syndromes in which about the course of action to pursue (Friston et al., 2013).
language areas are disconnected from motor regions, preventing A complementary perspective suggests that the role of dopamine
patients from obeying a verbal motor command (Geschwind, is to optimize sequences of actions into the future (O’Reilly

Frontiers in Human Neuroscience | www.frontiersin.org 6 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

The excessive dependence on sensory evidence has been

described in terms of an aberrant belief about the precision of
the likelihood distribution (Lawson et al., 2014). This account
additionally considers the source of this belief (Lawson et al.,
2017). It suggests that this may be understood in terms of an
aberrant prior belief about the volatility of the environment.
Volatility here means the ‘noisiness’ (or stochasticity) of the
transition probabilities that describe the dynamics of hidden
causes in the world. Highly volatile transitions prevent the
precise estimation of current states from the past, and result
in imprecise beliefs about hidden causes. In other words,
past beliefs become less informative when making inferences
about the present. Sensory prediction errors then elicit a
greater change in beliefs than they would do if a strong
FIGURE 4 | The anatomy of precision. The ascending neuromodulatory prior were in play. This theory of autism has been tested
systems carrying dopaminergic, cholinergic, and noradrenergic signals are empirically (Lawson et al., 2017), providing a convincing
shown (in a simplified form). Dopaminergic neurons have their cell-bodies in demonstration of computational neuropsychology in practice.
the ventral tegmental area (VTA) and the substantia nigra pars compacta
Using a Bayesian observer model (Mathys, 2012), it was shown
(SNc) – two nuclei in the midbrain. The medial forebrain bundle contains the
axons of these cells, and allows them to target neurons in the prefrontal that participants with autism overestimate the volatility of
cortex and the medium spiny neurons of the striatum. The nucleus basalis of their environment. Complementing this computational finding,
Meynert is found in the basal forebrain. This is the source of cholinergic pupillary responses, associated with central noradrenergic
projections to the cortex (Eckenstein et al., 1988). Axons originating here join activity (Koss, 1986), were found to be of a smaller magnitude
the cingulum. Neurons in the locus coeruleus project from the brainstem,
when participants encountered surprising stimuli compared to
through the dorsal noradrenergic bundle, and also join the cingulum to supply
the cortex with noradrenaline (Berridge and Waterhouse, 2003). neurotypical individuals.
A failure to properly balance the precision of sensory evidence,
in relation to prior beliefs, may be a ubiquitous theme in many
neuropsychiatric disorders. A potentially important aspect of this
and Frank, 2006). Deficient cholinergic signaling has been
imbalance is a failure to attenuate sensory precision during self-
implicated in the complex visual hallucinations associated
made acts. The attenuation of sensory precision is an important
with some neurodegenerative conditions (Collerton et al.,
aspect of movement and active sensing, because it allows us to
2005).
temporarily suspend attention to sensory evidence that we are
not moving (e.g., in the bradykinesia of Parkinson’s disease).
Precision and Autism In brief, a failure of sensory attenuation would have profound
One condition that has received considerable attention using consequences for self-generated movement, a sense of agency
Bayesian formulations is autism (Pellicano and Burr, 2012; and selfhood. We now consider the implications of Bayesian
Lawson et al., 2014). This condition usefully illustrates how pathologies for the active interrogation of the sensorium and its
aberrant prior beliefs about uncertainty can produce abnormal neuropsychology.
percepts. An influential treatment of the inferential deficits
in autism argues that the condition can be understood in
terms of weak prior beliefs (Pellicano and Burr, 2012). The ACTIVE INFERENCE AND VISUAL
consequence of this is that autistic individuals rely to a greater NEGLECT
extent upon current sensory data to make inferences about
hidden causes. This hypothesis is motivated by several empirical Active Sensing
observations, including the resistance of people with autism In the above, we have considered how hypotheses are evaluated
to sensory illusions (Happé, 1996; Simmons et al., 2009), and as if sensory data is passively presented to the brain. In reality,
their superior performance on tasks requiring the location perception is a much more active process of hypothesis testing
of low-level features in a complex image (Shah and Frith, (Krause, 2008; Yang et al., 2016a,b). Not only are hypotheses
1983). The susceptibility of the general population to sensory formed and refined, but experiments can be performed to
illusions is thought to be due to the exploitation of artificial confirm or refute them. Saccadic eye movements offer a good
scenarios that violate prior beliefs (Geisler and Kersten, 2002; example of this, as they turn vision from a passive to an active
Brown and Friston, 2012). For example, the perception of the process (Gibson, 1966; Ognibene and Baldassare, 2015; Parr
concave surface of a mask as a convex face is due to the, and Friston, 2017a). Each saccade can be thought of as an
normally accurate, prior (or ‘top-down’) belief that faces are experiment to adjudicate between plausible hypotheses about
convex (Gregory, 1970). Under this prior, the Bayes optimal the hidden causes that give rise to visual data (Friston et al.,
inference is a false inference (Weiss et al., 2002). If this prior 2012a; Mirza et al., 2016). As in science, the best experiments
belief is weakened, the optimal inference becomes the true are those that will bring about the greatest change in beliefs
inference. (Lindley, 1956; Friston et al., 2016b; Clark, 2017). A mathematical

Frontiers in Human Neuroscience | www.frontiersin.org 7 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

formulation of this imperative (Friston et al., 2015) suggests

that the form of the neuronal message passing required to
evaluate different (saccadic) policies maps well to the anatomy
of cortico-subcortical loops involving the basal ganglia (Friston
et al., 2017d). This is consistent with the known role of
this set of subcortical structures in action selection (Gurney
et al., 2001; Jahanshahi et al., 2015), and their anatomical
projections to oculomotor areas in the midbrain (Hikosaka
et al., 2000). To illustrate the importance of these points, we
consider visual neglect, a disorder in which active vision is
impaired.

Visual Neglect
A common neuropsychological syndrome, resulting from
damage to the right cerebral hemisphere, is visual neglect
(Halligan and Marshall, 1998). This is characterized by a failure
to attend to the left side of space. This rightward lateralization
may be a consequence of the mean-field factorization discussed
earlier. Although space is represented bilaterally in the brain, FIGURE 5 | The anatomy of visual neglect. Three lesions implicated in visual
there is no need for representations of identity to be bilateral. neglect are highlighted here. 1 – Disconnection of the second branch of the
right superior longitudinal fasciculus (a white matter tract that connects dorsal
This means that the relationships between location and identity
frontal with ventral parietal regions); 2 – Unilateral lesion to the right putamen;
should be asymmetrical, complementing the observation that 3 – Unilateral lesion to the right pulvinar (a thalamic nucleus). Note that lesion
visual neglect is very rarely the consequence of a left hemispheric 1 here is the same as lesion 2 in Figure 3.
lesion.
A behavioral manifestation of this disorder is a bias in
saccadic sampling (Husain et al., 2001; Fruhmann Berger et al.,
ventral parietal regions, we would expect there to be no
2008; Karnath and Rorden, 2012). Patients with neglect tend
change in beliefs following a saccade to the left. These
to perform saccades to locations on the right more frequently
would make for very poor ‘visual experiments’ (Lindley, 1956).
than to those on the left. There are several different sets
A very similar argument has been put forward for neglect
of prior beliefs that would make this behavior optimal. We
of personal space that emphasizes proprioceptive (rather than
will discuss three possibilities (Parr and Friston, 2017b), and
visual) consequences of action (Committeri et al., 2007). In
consider their biological bases (Figure 5). One is a prior
these circumstances, optimal behavior would require a greater
belief that proprioceptive data will be consistent with fixations
frequency of rightward saccades. Lesions to SLF II (Doricchi
on the right of space. The dorsal parietal lobe is known
and Tomaiuolo, 2003; Thiebaut de Schotten et al., 2005; Lunven
to contain the ‘parietal eye fields’ (Shipp, 2004), and it is
et al., 2015), and the regions it connects (Corbetta et al.,
plausible that an input to this region may specify such prior
2000; Corbetta and Shulman, 2002, 2011) are associated with
beliefs. A candidate structure is the dorsal pulvinar (Shipp,
neglect.
2003). This is a thalamic nucleus implicated in attentional
A third possibility is that the process of policy selection may be
processing (Ungerleider and Christensen, 1979; Robinson and
inherently biased. Above, we suggested that these computations
Petersen, 1992; Kanai et al., 2015). Crucially, lesions to this
may involve subcortical structures. The striatum, an input
structure have been implicated in neglect (Karnath et al.,
nucleus to the basal ganglia, is well known to be involved in
2002).
habit formation (Yin and Knowlton, 2006; Graybiel and Grafton,
A second possibility relates more directly to the question
2015). Habits may be formalized as a bias in prior beliefs about
of good experimental design. If a saccade is unlikely to
policy selection (FitzGerald et al., 2014). It is plausible that a
induce a change in current beliefs, then there is little value
lesion in the striatum might induce a similar behavioral bias
in performing it. One form that current beliefs take is the
toward saccades to rightward targets. One of the subcortical
likelihood distribution mapping ‘where I am looking’ to ‘what
structures most frequently implicated in visual neglect is the
I see’ (Mirza et al., 2016). As illustrated in Figure 3 (right)
putamen (Karnath et al., 2002), one of the constituent nuclei of
this likelihood distribution takes the form of a connection
the striatum. Such lesions may be interpretable as disrupting the
between dorsal frontal and ventral parietal regions (Parr
prior belief about policies.
and Friston, 2017a). To adjust beliefs about this mapping,
observations could induce a plastic change in synaptic strength
following each saccade (Friston et al., 2016b). If the white Anosognosia
matter tract connecting these areas is lesioned, it becomes The ideas outlined above, that movements can be thought of
impossible to update these beliefs. As such, if we were to as sensory experiments, are not limited to eye movements and
cut the second branch of the superior longitudinal fasciculus visual data. Plausibly, limb movements could be used to test
(SLF II) on the right, disconnecting dorsal frontal from hypotheses about proprioceptive (and visual) sensations. This

Frontiers in Human Neuroscience | www.frontiersin.org 8 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

has interesting consequences for a neuropsychological deficit asymmetry in the behaviors they elicit. It is also plausible that
known as anosognosia (Fotopoulou, 2012). This syndrome can functional disconnections might occur within a cortical
accompany hemiplegia, which prevents the performance of region. This would allow for disruption of transition
perceptual experiments using the paralyzed limb (Fotopoulou, probabilities. While not as vulnerable to vascular insult, other
2014). In addition to the failure to perform such an experiment, pathologies can cause changes in intrinsic cortical connectivity
patients must be able to ignore any discrepancy between (Cooray et al., 2015).
predicted movements and the contradictory sensory data Epistemic, foraging, behavior is vital for the evaluation of
suggesting the absence of a movement (Frith et al., 2000). beliefs about the world. Unusual patterns of sensorimotor
As this failure of monitoring movement trajectories can be sampling can be induced by abnormal beliefs about the
induced in healthy subjects (Fourneret and Jeannerod, 1998), motor experiments that best disambiguate between perceptual
it seems plausible that this could be exaggerated in the context hypotheses. These computations implicate subcortical structures,
of hemiplegia, through a dampening of exteroceptive sensory such as the basal ganglia. There are two ways that disruption
precision. of these computations may result in abnormal behavior. The
This explanation is not sufficient on its own, as anosognosia first is that prior beliefs about policies may be biased. This can
does not occur in all cases of hemiplegia. Lesion mapping be an indirect effect, through other beliefs, or a direct effect
has implicated the insula in the deficits observed in these due to dysfunction in basal ganglia networks. The second is
patients (Karnath et al., 2005; Fotopoulou et al., 2010). that an impairment in performing these experiments, due to
This is a region often associated with interoceptive inference paralysis, might impair the refutation of incorrect perceptual
(Barrett and Simmons, 2015) that has substantial efferent hypotheses. This may be compounded by a disconnection
connectivity to somatosensory cortex (Showers and Lauer, or a neuromodulatory failure, as has been proposed in
1961; Mesulam and Mufson, 1982). Damage to the insula anosognosia.
and surrounding regions might reflect a disconnection of the One further source of an aberrant priors, not discussed in
mapping from motor hypotheses to the interoceptive data the above, is neuronal loss. In neurodegenerative disorders,
that accounts for what it ‘feels like’ to move a limb. This is there may be a reduction in the number of neurons in a
consistent with evidence that the insula mediates inferences given brain area. This results in a smaller number of possible
about these kinds of sensations (Allen et al., 2016). A plausible activity patterns across these neurons and limits the number
hypothesis for the computational pathology in anosognosia of hypotheses they can represent. This means that disorders
is then that a failure of active inference is combined with in which neurons are lost may cause a shrinkage of the
a disconnection of the likelihood mapping between motor brain’s hypothesis space. In other words, the failure to form
control and its interoceptive (and exteroceptive) consequences accurate perceptual hypotheses in such conditions may be
(Fotopoulou et al., 2008). due to an attrition of the number of hypotheses that can
be entertained by the brain. An important future step in
Bayesian neuropsychology will be linking tissue pathology with
A (PROVISIONAL) TAXONOMY OF computation more directly. This may be one route toward
COMPUTATIONAL PATHOLOGY achieving this.

In the above, we have described the components of a

generative model required to perform Bayesian inference. We CONCLUSION
have reviewed some of the syndromes that may illustrate
deficits of one or more of these components. Broadly, the While Bayesian approaches are not in conflict with other
generative model constitutes beliefs about the hidden states, methods in computational neuroscience, they do offer a different
their dynamics, and the mechanisms by which sensory data (complementary) perspective that is often very useful. For
is generated from hidden states. Each of these beliefs can example, many traditional modeling approaches would not
be disrupted through an increase or decrease in precision, predict that disconnections in early sensory streams, such as
or through disconnections. Modulation of precision implicates the retino-geniculate system, could result in complex sensory
the ascending neuromodulatory systems. This modulation may hallucinations. Calling upon a hierarchical generative model
be important for a range of neuropsychiatric and functional that makes ‘top-down’ predictions about sensory data, clarifies
neurological disorders (Edwards et al., 2012). and provides insight into such issues. In the above we have
In addition to modulation of connectivity, disconnections discussed the features of the generative models that underwrite
can completely disrupt beliefs about the conditional probability perception and behavior. We have illustrated the importance
of one variable given another. The hierarchical architecture of these features through examples of their failures. These
of the cortex suggests that inter-areal white matter tracts, computational pathologies can be described in terms of abnormal
the most vulnerable to vascular or inflammatory lesions, prior beliefs, or in terms of their biological substrates. We
represent likelihood distributions (i.e., the probability of noted that aberrant priors about the structure of a likelihood
data, or a low-level cause, given a high-level cause). Drawing mapping relate to disconnection syndromes, ubiquitous in
upon the notion of a mean-field factorization, we noted neurology. Pathological beliefs about uncertainty may manifest
that such disconnections are likely to have a hemispheric as neuromodulatory disorders. The process of identifying the

Frontiers in Human Neuroscience | www.frontiersin.org 9 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

pathological priors that give rise to Bayes optimal behavior AUTHOR CONTRIBUTIONS
in patients is promising both scientifically and clinically. If
individual patients can be uniquely characterized by subject- All authors listed have made a substantial, direct and intellectual
specific priors, this facilitates a precision medicine approach contribution to the work, and approved it for publication.
grounded in computational phenotyping (Adams et al., 2016;
Schwartenbeck and Friston, 2016; Mirza et al., 2018). This also
allows for empirical evaluation of hypotheses about abnormal FUNDING
priors, by comparing quantitative, computational phenotypes
between clinical and healthy populations. Relating these priors TP was supported by the Rosetrees Trust (Award Number
to their biological substrates offers the further possibility of 173346). GR was supported by a Wellcome Trust Senior Clinical
treatments that target aberrant neurobiology in a patient specific Research Fellowship (100227). KF is a Wellcome Principal
manner. Research Fellow (Ref: 088130/Z/09/Z).

REFERENCES Catani, M., and Mesulam, M. (2008). The arcuate fasciculus and the disconnection
theme in language and aphasia: history and current state. Cortex 44, 953–961.
Abutalebi, J., Rosa, P. A. D., Tettamanti, M., Green, D. W., and Cappa, S. F. doi: 10.1016/j.cortex.2008.04.002
(2009). Bilingual aphasia and language control: a follow-up fMRI and intrinsic Ciaraffa, F., Castelli, G., Parati, E. A., Bartolomeo, P., and Bizzi, A. (2013). Visual
connectivity study. Brain Lang. 109, 141–156. doi: 10.1016/j.bandl.2009.03.003 neglect as a disconnection syndrome? A confirmatory case report. Neurocase
Adams, R. A., Bauer, M., Pinotsis, D., and Friston, K. J. (2016). Dynamic 19, 351–359. doi: 10.1080/13554794.2012.667130
causal modelling of eye movements during pursuit: confirming Clark, A. (2017). A nice surprise? Predictive processing and the active pursuit of
precision-encoding in V1 using MEG. Neuroimage 132, 175–189. doi: novelty. Phenomenol. Cogn. Sci. 1–14.
10.1016/j.neuroimage.2016.02.055 Collerton, D., Perry, E., and McKeith, I. (2005). Why people see things that
Adams, R. A., Huys, Q. J., and Roiser, J. P. (2015). Computational psychiatry: are not there: a novel perception and attention deficit model for recurrent
towards a mathematically informed understanding of mental illness. J. Neurol. complex visual hallucinations. Behav. Brain Sci. 28, 737–757. doi: 10.1017/
Neurosurg. Psychiatry 87, 53–63. doi: 10.1136/jnnp-2015-310737 S0140525X05000130
Adams, R. A., Shipp, S., and Friston, K. J. (2013a). Predictions not commands: Committeri, G., Pitzalis, S., Galati, G., Patria, F., Pelle, G., Sabatini, U., et al.
active inference in the motor system. Brain Struct. Funct. 218, 611–643. (2007). Neural bases of personal and extrapersonal neglect in humans. Brain
doi: 10.1007/s00429-012-0475-5 130, 431–441. doi: 10.1093/brain/awl265
Adams, R. A., Stephan, K. E., Brown, H. R., Frith, C. D., and Friston, K. J. (2013b). Conant, R. C., and Ashby, W. R. (1970). Every good regulator of a system must
The computational anatomy of psychosis. Front. Psychiatry 4:47. doi: 10.3389/ be a model of that system. Int. J. Syst. Sci. 1, 89–97. doi: 10.1080/002077270089
fpsyt.2013.00047 20220
Allen, M., Fardo, F., Dietz, M. J., Hillebrandt, H., Friston, K. J., Rees, G., et al. Cooray, G. K., Sengupta, B., Douglas, P., Englund, M., Wickstrom, R., and
(2016). Anterior insula coordinates hierarchical processing of tactile mismatch Friston, K. (2015). Characterising seizures in anti-NMDA-receptor encephalitis
responses. Neuroimage 127(Suppl. C), 34–43. doi: 10.1016/j.neuroimage.2015. with dynamic causal modelling. Neuroimage 118, 508–519. doi: 10.1016/j.
11.030 neuroimage.2015.05.064
Barrett, L. F., and Simmons, W. K. (2015). Interoceptive predictions in the brain. Corbetta, M., Kincade, J. M., Ollinger, J. M., McAvoy, M. P., and Shulman, G. L.
Nat. Rev. Neurosci. 16, 419–429. doi: 10.1038/nrn3950 (2000). Voluntary orienting is dissociated from target detection in human
Bartolomeo, P. (2014). “Spatially biased decisions: toward a dynamic interactive posterior parietal cortex. Nat. Neurosci. 3, 292–297. doi: 10.1038/73009
model of visual neglect,” in Cognitive Plasticity in Neurologic Disorders, eds J. I. Corbetta, M., and Shulman, G. L. (2002). Control of goal-directed and stimulus-
Tracy, B. Hampstead, and K. Sathian (Oxford: Oxford University Press), 299. driven attention in the brain. Nat. Rev. Neurosci. 3, 201–215. doi: 10.1038/
Bartolomeo, P., Thiebaut de Schotten, M., and Doricchi, F. (2007). Left nrn755
unilateral neglect as a disconnection syndrome. Cereb. Cortex 17, 2479–2490. Corbetta, M., and Shulman, G. L. (2011). Spatial neglect and attention networks.
doi: 10.1093/cercor/bhl181 Annu. Rev. Neurosci. 34, 569–599. doi: 10.1146/annurev-neuro-061010-113731
Bastos, A. M., Usrey, W. M., Adams, R. A., Mangun, G. R., Fries, P., and Friston, Corlett, P. R., and Fletcher, P. C. (2014). Computational psychiatry: a Rosetta
K. J. (2012). Canonical microcircuits for predictive coding. Neuron 76, 695–711. Stone linking the brain to mental illness. Lancet Psychiatry 1, 399–402.
doi: 10.1016/j.neuron.2012.10.038 doi: 10.1016/S2215-0366(14)70298-6
Bates, E., Wilson, S. M., Saygin, A. P., Dick, F., Sereno, M. I., Knight, R. T., Daunizeau, J., den Ouden, H. E. M., Pessiglione, M., Kiebel, S. J., Stephan, K. E.,
et al. (2003). Voxel-based lesion–symptom mapping. Nat. Neurosci. 6, 448–450. and Friston, K. J. (2010). Observing the observer (I): meta-bayesian models of
doi: 10.1038/nn1050 learning and decision-making. PLoS One 5:e15554. doi: 10.1371/journal.pone.
Beal, M. J. (2003). Variational Algorithms for Approximate Bayesian Inference. 0015554
London: University of London. Dauwels, J. (2007). “On variational message passing on factor graphs,” in
Berridge, C. W., and Waterhouse, B. D. (2003). The locus coeruleus–noradrenergic Proceedings of the IEEE International Symposium Information Theory
system: modulation of behavioral state and state-dependent cognitive processes. (ISIT), 24–29 June 2007, Nice, 2546–2550. doi: 10.1109/ISIT.2007.
Brain Res. Rev. 42, 33–84. doi: 10.1016/S0165-0173(03)00143-7 4557602
Binder, J. R. (2015). The Wernicke area: modern evidence and a reinterpretation. Dayan, P., and Yu, A. J. (2001). ACh, Uncertainty, and Cortical Inference. Montreal,
Neurology 85, 2170–2175. doi: 10.1212/WNL.0000000000002219 QC: NIPS.
Boes, A. D., Prasad, S., Liu, H., Liu, Q., Pascual-Leone, A., Caviness, V. S., et al. Dayan, P., and Yu, A. J. (2006). Phasic norepinephrine: a neural interrupt signal
(2015). Network localization of neurological symptoms from focal brain lesions. for unexpected events. Network 17, 335–350. doi: 10.1080/0954898060100
Brain 138, 3061–3075. doi: 10.1093/brain/awv228 4024
Brown, H., and Friston, K. J. (2012). Free-energy and illusions: the cornsweet effect. De Ridder, D., Vanneste, S., and Freeman, W. (2014). The Bayesian brain: phantom
Front. Psychol. 3:43. doi: 10.3389/fpsyg.2012.00043 percepts resolve sensory uncertainty. Neurosci. Biobehav. Rev. 44(Suppl. C),
Carrera, E., and Tononi, G. (2014). Diaschisis: past, present, future. Brain 137, 4–15. doi: 10.1016/j.neubiorev.2012.04.001
2408–2422. doi: 10.1093/brain/awu101 Desimone, R., Schein, S. J., Moran, J., and Ungerleider, L. G. (1985). Contour,
Catani, M., and Ffytche, D. H. (2005). The rises and falls of disconnection color and shape analysis beyond the striate cortex. Vision Res. 25, 441–452.
syndromes. Brain 128, 2224–2239. doi: 10.1093/brain/awh622 doi: 10.1016/0042-6989(85)90069-0

Frontiers in Human Neuroscience | www.frontiersin.org 10 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

Di Pellegrino, G., Fadiga, L., Fogassi, L., Gallese, V., and Rizzolatti, G. (1992). Friston, K., and Kiebel, S. (2009). Predictive coding under the free-energy principle.
Understanding motor events: a neurophysiological study. Exp. Brain Res. 91, Philos. Trans. R. Soc. B Biol. Sci. 364, 1211–1221. doi: 10.1098/rstb.2008.
176–180. doi: 10.1007/BF00230027 0300
Dietz, M. J., Friston, K. J., Mattingley, J. B., Roepstorff, A., and Garrido, M. I. (2014). Friston, K., Rigoli, F., Ognibene, D., Mathys, C., Fitzgerald, T., and Pezzulo, G.
Effective connectivity reveals right-hemisphere dominance in audiospatial (2015). Active inference and epistemic value. Cogn. Neurosci. 6, 187–214.
perception: implications for models of spatial neglect. J. Neurosci. 34, doi: 10.1080/17588928.2015.1020053
5003–5011. doi: 10.1523/JNEUROSCI.3765-13.2014 Friston, K., Samothrakis, S., and Montague, R. (2012b). Active inference and
Doricchi, F., and Tomaiuolo, F. (2003). The anatomy of neglect without agency: optimal control without cost functions. Biol. Cybern. 106, 523–541.
hemianopia: a key role for parietal–frontal disconnection? Neuroreport 14, Friston, K., Schwartenbeck, P., Fitzgerald, T., Moutoussis, M., Behrens, T., and
2239–2243. Dolan, R. (2013). The anatomy of choice: active inference and agency. Front.
Doya, K. (2007). Bayesian Brain: Probabilistic Approaches to Neural Coding. Hum. Neurosci. 7:598. doi: 10.3389/fnhum.2013.00598
Cambridge, MA: MIT Press. Friston, K., Schwartenbeck, P., FitzGerald, T., Moutoussis, M., Behrens, T.,
Dronkers, N. F., and Baldo, J. V. (2009). “Language: aphasia A2 - squire,” in and Dolan, R. J. (2014). The anatomy of choice: dopamine and decision-
Encyclopedia of Neuroscience, ed. R. Larry (Oxford: Academic Press), 343–348. making. Philos. Trans. R. Soc. B Biol. Sci. 369:20130481. doi: 10.1098/rstb.2013.
Eckenstein, F. P., Baughman, R. W., and Quinn, J. (1988). An anatomical study 0481
of cholinergic innervation in rat cerebral cortex. Neuroscience 25, 457–474. Friston, K. J., Daunizeau, J., Kilner, J., and Kiebel, S. J. (2010). Action and behavior:
doi: 10.1016/0306-4522(88)90251-5 a free-energy formulation. Biol. Cybern. 102, 227–260. doi: 10.1007/s00422-
Edwards, M. J., Adams, R. A., Brown, H., Pareés, I., and Friston, K. J. (2012). 010-0364-z
A Bayesian account of ‘hysteria’. Brain 135, 3495–3512. doi: 10.1093/brain/ Friston, K. J., FitzGerald, T., Rigoli, F., Schwartenbeck, P., and Pezzulo, G. (2017a).
aws129 Active inference: a process theory. Neural Comput. 29, 1–49. doi: 10.1162/
Feldman, H., and Friston, K. (2010). Attention, uncertainty, and free-energy. Front. NECO_a_00912
Hum. Neurosci. 4:215. doi: 10.3389/fnhum.2010.00215 Friston, K. J., Parr, T., and Vries, B. D. (2017b). The graphical brain: belief
Felleman, D. J., and Van Essen, D. C. (1991). Distributed hierarchical processing in propagation and active inference. Netw. Neurosci. 1, 381–414. doi: 10.1162/
the primate cerebral cortex. Cereb. Cortex 1, 1–47. doi: 10.1093/cercor/1.1.1 NETN_a_00018
Ffytche, D. H., and Howard, R. J. (1999). The perceptual consequences of visual Friston, K. J., Redish, A. D., and Gordon, J. A. (2017c). Computational nosology
loss: ‘positive’ pathologies of vision. Brain 122, 1247–1260. doi: 10.1093/brain/ and precision psychiatry. Comput. Psychiatry 1, 2–23. doi: 10.1162/CPSY_a_
122.7.1247 00001
FitzGerald, T. H. B., Dolan, R. J., and Friston, K. J. (2014). Model averaging, Friston, K. J., Rosch, R., Parr, T., Price, C., and Bowman, H. (2017d). Deep
optimal inference, and habit formation. Front. Hum. Neurosci. 8:457. temporal models and active inference. Neurosci. Biobehav. Rev. 77, 388–402.
doi: 10.3389/fnhum.2014.00457 doi: 10.1016/j.neubiorev.2017.04.009
Fletcher, P. C., and Frith, C. D. (2009). Perceiving is believing: a Bayesian approach Frith, C. D., Blakemore, S. J., and Wolpert, D. M. (2000). Abnormalities in
to explaining the positive symptoms of schizophrenia. Nat. Rev. Neurosci. 10, the awareness and control of action. Philos. Trans. R. Soc. B Biol. Sci. 355,
48–58. doi: 10.1038/nrn2536 1771–1788. doi: 10.1098/rstb.2000.0734
Fornito, A., Zalesky, A., and Breakspear, M. (2015). The connectomics of brain Frith, C. D., and Frith, U. (1999). Interacting minds—a biological basis. Science
disorders. Nat. Rev. Neurosci. 16, 159–172. doi: 10.1038/nrn3901 286, 1692–1695. doi: 10.1126/science.286.5445.1692
Fotopoulou, A. (2012). Illusions and delusions in anosognosia for hemiplegia: from Fruhmann Berger, M., Johannsen, L., and Karnath, H.-O. (2008). Time course
motor predictions to prior beliefs. Brain 135, 1344–1346. doi: 10.1093/brain/ of eye and head deviation in spatial neglect. Neuropsychology 22, 697–702.
aws094 doi: 10.1037/a0013351
Fotopoulou, A. (2014). Time to get rid of the ‘Modular’ in neuropsychology: a Gallant, J., Braun, J., and Van Essen, D. (1993). Selectivity for polar, hyperbolic,
unified theory of anosognosia as aberrant predictive coding. J. Neuropsychol. and Cartesian gratings in macaque visual cortex. Science 259, 100–103.
8, 1–19. doi: 10.1111/jnp.12010 doi: 10.1126/science.8418487
Fotopoulou, A., Pernigo, S., Maeda, R., Rudd, A., and Kopelman, M. A. (2010). Geisler, W. S., and Kersten, D. (2002). Illusions, perception and Bayes. Nat.
Implicit awareness in anosognosia for hemiplegia: unconscious interference Neurosci. 5, 508–510. doi: 10.1038/nn0602-508
without conscious re-representation. Brain 133, 3564–3577. doi: 10.1093/brain/ Geschwind, N. (1965a). Disconnexion syndromes in animals and man1. Brain 88,
awq233 237–294. doi: 10.1093/brain/88.2.237
Fotopoulou, A., Tsakiris, M., Haggard, P., Vagopoulou, A., Rudd, A., and Geschwind, N. (1965b). Disconnexion syndromes in animals and man. II. Brain 88,
Kopelman, M. (2008). The role of motor intention in motor awareness: an 585–644. doi: 10.1093/brain/88.3.585
experimental study on anosognosia for hemiplegia. Brain 131, 3432–3442. Ghahramani, Z. (2015). Probabilistic machine learning and artificial intelligence.
doi: 10.1093/brain/awn225 Nature 521, 452–459. doi: 10.1038/nature14541
Fourneret, P., and Jeannerod, M. (1998). Limited conscious monitoring of motor Gibson, J. J. (1966). The Senses Considered as Perceptual Systems. Boston, MA:
performance in normal subjects. Neuropsychologia 36, 1133–1140. doi: 10.1016/ Houghton Mifflin Company.
S0028-3932(98)00006-2 Glickstein, M., and Berlucchi, G. (2008). Classical disconnection studies of the
Frank, M. J., Seeberger, L. C., and O’Reilly, R. C. (2004). By carrot or by stick: corpus callosum. Cortex 44, 914–927. doi: 10.1016/j.cortex.2008.04.001
cognitive reinforcement learning in Parkinsonism. Science 306, 1940–1943. Goldenberg, G. (2003). Apraxia and beyond: life and work of Hugo Liepmann.
doi: 10.1126/science.1102941 Cortex 39, 509–524. doi: 10.1016/S0010-9452(08)70261-2
Friston, K. (2010). The free-energy principle: a unified brain theory? Nat. Rev. Graybiel, A. M., and Grafton, S. T. (2015). The striatum: where skills and habits
Neurosci. 11, 127–138. doi: 10.1038/nrn2787 meet. Cold Spring Harb. Perspect. Biol. 7:a021691. doi: 10.1101/cshperspect.
Friston, K., Adams, R. A., Perrinet, L., and Breakspear, M. (2012a). Perceptions as a021691
hypotheses: saccades as experiments. Front. Psychol. 3:151. doi: 10.3389/fpsyg. Grefkes, C., Nowak, D. A., Eickhoff, S. B., Dafotakis, M., Küst, J., Karbe, H., et al.
2012.00151 (2008). Cortical connectivity after subcortical stroke assessed with functional
Friston, K., Brown, H. R., Siemerkus, J., and Stephan, K. E. (2016a). The magnetic resonance imaging. Ann. Neurol. 63, 236–246. doi: 10.1002/ana.
dysconnection hypothesis 2016). Schizophr. Res. 176, 83–94. doi: 10.1016/j. 21228
schres.2016.07.014 Gregory, R. L. (1970). The Intelligent Eye. London: Weidenfeld and Nicolson.
Friston, K., and Buzsáki, G. (2016). The functional anatomy of time: what and when Gregory, R. L. (1980). Perceptions as hypotheses. Philos. Trans. R. Soc. Lond. B Biol.
in the brain. Trends Cogn. Sci. 20, 500–511. doi: 10.1016/j.tics.2016.05.001 Sci. 290, 181–197. doi: 10.1098/rstb.1980.0090
Friston, K., FitzGerald, T., Rigoli, F., Schwartenbeck, P., O’Doherty, J., and Gross, C. G., Rocha-Miranda, C. E., and Bender, D. B. (1972). Visual properties of
Pezzulo, G. (2016b). Active inference and learning. Neurosci. Biobehav. Rev. 68, neurons in inferotemporal cortex of the Macaque. J. Neurophysiol. 35, 96–111.
862–879. doi: 10.1016/j.neubiorev.2016.06.022 doi: 10.1152/jn.1972.35.1.96

Frontiers in Human Neuroscience | www.frontiersin.org 11 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

Gurney, K., Prescott, T. J., and Redgrave, P. (2001). A computational model of Kiebel, S. J., Daunizeau, J., and Friston, K. J. (2008). A hierarchy of time-scales and
action selection in the basal ganglia. I. A new functional anatomy. Biol. Cybern. the brain. PLoS Comput. Biol. 4:e1000209. doi: 10.1371/journal.pcbi.1000209
84, 401–410. doi: 10.1007/PL00007984 Kiebel, S. J., Daunizeau, J., and Friston, K. J. (2009). Perception and hierarchical
Halligan, P. W., and Marshall, J. C. (1998). Neglect of awareness. Conscious. Cogn. dynamics. Front. Neuroinform. 3:20. doi: 10.3389/neuro.11.020.2009
7, 356–380. doi: 10.1006/ccog.1998.0362 Kilner, J. M., Friston, K. J., and Frith, C. D. (2007). Predictive coding: an account of
Happé, F. G. E. (1996). Studying weak central coherence at low levels: children with the mirror neuron system. Cogn. Process. 8, 159–166. doi: 10.1007/s10339-007-
autism do not succumb to visual illusions. A research note. J. Child Psychol. 0170-2
Psychiatry 37, 873–877. doi: 10.1111/j.1469-7610.1996.tb01483.x Kinsbourne, M. (1970). A model for the mechanism of unilateral neglect of space.
Hasson, U., Chen, J., and Honey, C. J. (2015). Hierarchical process memory: Trans. Am. Neurol. Assoc. 95, 143–146.
memory as an integral component of information processing. Trends Cogn. Sci. Kirshner, H. S. (2003). “Chapter 140 - speech and language disorders A2 - Samuels,
19, 304–313. doi: 10.1016/j.tics.2015.04.006 Martin A,” in Office Practice of Neurology, 2nd Edn, ed. S. K. Feske (Philadelphia,
Hasson, U., Yang, E., Vallines, I., Heeger, D. J., and Rubin, N. (2008). A hierarchy PA: Churchill Livingstone), 890–895.
of temporal receptive windows in human cortex. J. Neurosci 28, 2539–2550. Knill, D. C., and Pouget, A. (2004). The Bayesian brain: the role of uncertainty in
doi: 10.1523/JNEUROSCI.5487-07.2008 neural coding and computation. Trends Neurosci. 27, 712–719. doi: 10.1016/j.
He, B. J., Snyder, A. Z., Vincent, J. L., Epstein, A., Shulman, G. L., and tins.2004.10.007
Corbetta, M. (2007). Breakdown of functional connectivity in frontoparietal Koss, M. C. (1986). Pupillary dilation as an index of central nervous system α2-
networks underlies behavioral deficits in spatial neglect. Neuron 53, 905–918. adrenoceptor activation. J. Pharmacol. Methods 15, 1–19. doi: 10.1016/0160-
doi: 10.1016/j.neuron.2007.02.013 5402(86)90002-1
Hickok, G. (2012a). Computational neuroanatomy of speech production. Nat. Rev. Krakauer, J. W., and Shadmehr, R. (2007). Towards a computational
Neurosci. 13, 135–145. doi: 10.1038/nrn3158 neuropsychology of action. Prog. Brain Res. 165, 383–394.
Hickok, G. (2012b). The cortical organization of speech processing: feedback doi: 10.1016/S0079-6123(06)65024-3
control and predictive coding the context of a dual-stream model. J. Commun. Krause, A. (2008). Optimizing Sensing: Theory and Applications. Pittsburgh, PA:
Disord. 45, 393–402. doi: 10.1016/j.jcomdis.2012.06.004 Carnegie Mellon University.
Hickok, G., and Poeppel, D. (2007). The cortical organization of speech processing. Lawson, R. P., Mathys, C., and Rees, G. (2017). Adults with autism overestimate
Nat. Rev. Neurosci. 8, 393–402. doi: 10.1038/nrn2113 the volatility of the sensory environment. Nat. Neurosci. 20, 1293–1299.
Hikosaka, O., Takikawa, Y., and Kawagoe, R. (2000). Role of the basal ganglia in doi: 10.1038/nn.4615
the control of purposive saccadic eye movements. Physiol. Rev. 80, 953–978. Lawson, R. P., Rees, G., and Friston, K. J. (2014). An aberrant precision account of
doi: 10.1152/physrev.2000.80.3.953 autism. Front. Hum. Neurosci. 8:302. doi: 10.3389/fnhum.2014.00302
Howard, D., Patterson, K., Wise, R., Brown, W. D., Friston, K., Weiller, C., Lindley, D. V. (1956). On a measure of the information provided by an experiment.
et al. (1992). The cortical localization of the lexicons: positron emission Ann. Math. Stat. 27, 986–1005. doi: 10.1214/aoms/1177728069
tomography evidence. Brain 115, 1769–1782. doi: 10.1093/brain/115.6. Lunven, M., Thiebaut De Schotten, M., Bourlon, C., Duret, C., Migliaccio, R.,
1769 Rode, G., et al. (2015). White matter lesional predictors of chronic visual
Hubel, D. H., and Wiesel, T. N. (1959). Receptive fields of single neurones in the neglect: a longitudinal study. Brain 138(Pt 3), 746–760. doi: 10.1093/brain/
cat’s striate cortex. J. Physiol. 148, 574–591. doi: 10.1113/jphysiol.1959.sp006308 awu389
Humphreys, G. W., and Forde, E. M. (2001). Hierarchies, similarity, and Mah, Y.-H., Husain, M., Rees, G., and Nachev, P. (2014). Human brain lesion-
interactivity in object recognition: “category-specific” neuropsychological deficit inference remapped. Brain 137, 2522–2531. doi: 10.1093/brain/awu164
deficits. Behav. Brain Sci. 24, 453–476. Markov, N. T., Vezoli, J., Chameau, P., Falchier, A., Quilodran, R., Huissoud, C.,
Husain, M., Mannan, S., Hodgson, T., Wojciulik, E., Driver, J., and Kennard, C. et al. (2013). Anatomy of hierarchy: feedforward and feedback pathways in
(2001). Impaired spatial working memory across saccades contributes to macaque visual cortex. J. Comp. Neurol. 522, 225–259. doi: 10.1002/cne.23458
abnormal search in parietal neglect. Brain 124(Pt 5), 941–952. doi: 10.1093/ Marshall, L., Mathys, C., Ruge, D., de Berker, A. O., Dayan, P., Stephan, K. E.,
brain/124.5.941 et al. (2016). Pharmacological fingerprints of contextual uncertainty. PLoS Biol.
Husain, M., and Nachev, P. (2007). Space and the parietal cortex. Trends Cogn. Sci. 14:e1002575. doi: 10.1371/journal.pbio.1002575
11, 30–36. doi: 10.1016/j.tics.2006.10.011 Mathys, C. D. (2012). Hierarchical Gaussian Filtering. Doctoral dissertation, ETH
Huys, Q. J., Maia, T. V., and Frank, M. J. (2016). Computational psychiatry as a Zurich, Zürich.
bridge from neuroscience to clinical applications. Nat. Neurosci. 19, 404–413. Menon, G. J., Rahman, I., Menon, S. J., and Dutton, G. N. (2003). Complex visual
doi: 10.1038/nn.4238 hallucinations in the visually impaired: the Charles bonnet syndrome. Surv.
Iglesias, S., Tomiello, S., Schneebeli, M., and Stephan, K. E. (2017). Models of Ophthalmol. 48, 58–72. doi: 10.1016/S0039-6257(02)00414-9
neuromodulation for computational psychiatry. Wiley Interdiscip. Rev. Cogn. Mesulam, M. M., and Mufson, E. J. (1982). Insula of the old world monkey. III:
Sci. 8:e1420. doi: 10.1002/wcs.1420 efferent cortical output and comments on function. J. Comp. Neurol. 212, 38–52.
Jahanshahi, M., Obeso, I., Rothwell, J. C., and Obeso, J. A. (2015). A fronto-striato- doi: 10.1002/cne.902120104
subthalamic-pallidal network for goal-directed and habitual inhibition. Nat. Mintzopoulos, D., Astrakas, L. G., Khanicheh, A., Konstas, A. A., Singhal, A.,
Rev. Neurosci. 16, 719–732. doi: 10.1038/nrn4038 Moskowitz, M. A., et al. (2009). Connectivity alterations assessed by
Kanai, R., Komura, Y., Shipp, S., and Friston, K. (2015). Cerebral hierarchies: combining fMRI and MR-compatible hand robots in chronic stroke.
predictive processing, precision and the pulvinar. Philos. Trans. R. Soc. B Biol. Neuroimage 47(Suppl. 2), T90–T97. doi: 10.1016/j.neuroimage.2009.
Sci. 370:20140169. doi: 10.1098/rstb.2014.0169 03.007
Karnath, H.-O., Baier, B., and Nägele, T. (2005). Awareness of the functioning of Mirza, M. B., Adams, R. A., Mathys, C., and Friston, K. J. (2018). Human visual
one’s own limbs mediated by the insular cortex? J. Neurosci. 25, 7134–7138. exploration reduces uncertainty about the sensed world. PLoS One 13:e0190429.
doi: 10.1523/JNEUROSCI.1590-05.2005 doi: 10.1371/journal.pone.0190429
Karnath, H. O., Himmelbach, M., and Rorden, C. (2002). The subcortical anatomy Mirza, M. B., Adams, R. A., Mathys, C. D., and Friston, K. J. (2016). Scene
of human spatial neglect: putamen, caudate nucleus and pulvinar. Brain 125, construction, visual foraging, and active inference. Front. Comput. Neurosci.
350–360. doi: 10.1093/brain/awf032 10:56. doi: 10.3389/fncom.2016.00056
Karnath, H.-O., and Rorden, C. (2012). The anatomy of spatial neglect. Moran, R. J., Campo, P., Symmonds, M., Stephan, K. E., Dolan, R. J., and
Neuropsychologia 50, 1010–1017. doi: 10.1016/j.neuropsychologia.2011.06.027 Friston, K. J. (2013). Free energy, precision and learning: the role of cholinergic
Karnath, H.-O., and Smith, D. V. (2014). The next step in modern brain lesion neuromodulation. J. Neurosci. 33, 8227–8236. doi: 10.1523/JNEUROSCI.4255-
analysis: multivariate pattern analysis. Brain 137, 2405–2407. doi: 10.1093/ 12.2013
brain/awu180 Murray, J. D., Bernacchia, A., Freedman, D. J., Romo, R., Wallis, J. D., Cai, X., et al.
Keysers, C., and Perrett, D. I. (2004). Demystifying social cognition: a Hebbian (2014). A hierarchy of intrinsic timescales across primate cortex. Nat. Neurosci.
perspective. Trends Cogn. Sci. 8, 501–507. doi: 10.1016/j.tics.2004.09.005 17, 1661–1663. doi: 10.1038/nn.3862

Frontiers in Human Neuroscience | www.frontiersin.org 12 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

Nachev, P. (2015). The first step in modern lesion-deficit analysis. Brain Shah, A., and Frith, U. (1983). An islet of ability in autistic children: a research
138(Pt 6):e354. doi: 10.1093/brain/awu275 note. J. Child Psychol. Psychiatry 24, 613–620. doi: 10.1111/j.1469-7610.1983.
Nachev, P., and Hacker, P. (2014). The neural antecedents to voluntary action: tb00137.x
a conceptual analysis. Cogn. Neurosci. 5, 193–208. doi: 10.1080/17588928.2014. Shipp, S. (2003). The functional logic of cortico-pulvinar connections.
934215 Philos. Trans. R. Soc. B Biol. Sci. 358, 1605–1624. doi: 10.1098/rstb.2002.
O’Callaghan, C., Hall, J. M., Tomassini, A., Muller, A. J., Walpola, I. C., Moustafa, 1213
A. A., et al. (2017). Visual hallucinations are characterized by impaired sensory Shipp, S. (2004). The brain circuitry of attention. Trends Cogn. Sci. 8, 223–230.
evidence accumulation: insights from hierarchical drift diffusion modeling in doi: 10.1016/j.tics.2004.03.004
Parkinson’s disease. Biol. Psychiatry Cogn. Neurosci. Neuroimaging 2, 680–688. Shipp, S. (2007). Structure and function of the cerebral cortex. Curr. Biol. 17,
doi: 10.1016/j.bpsc.2017.04.007 R443–R449. doi: 10.1016/j.cub.2007.03.044
Ochipa, C., Rothi, L. J., and Heilman, K. M. (1994). Conduction apraxia. J. Neurol. Shipp, S. (2016). Neural elements for predictive coding. Front. Psychol. 7:1792.
Neurosurg. Psychiatry 57, 1241–1244. doi: 10.1136/jnnp.57.10.1241 doi: 10.3389/fpsyg.2016.01792
Ognibene, D., and Baldassare, G. (2015). Ecological active vision: four bioinspired Shipp, S., Adams, R. A., and Friston, K. J. (2013). Reflections on agranular
principles to integrate bottom–up and adaptive top–down attention tested architecture: predictive coding in the motor cortex. Trends Neurosci. 36,
with a simple camera-arm robot. IEEE Trans. Auton. Ment. Dev. 7, 3–25. 706–716. doi: 10.1016/j.tins.2013.09.004
doi: 10.1109/TAMD.2014.2341351 Showers, M. J., and Lauer, E. W. (1961). Somatovisceral motor patterns
O’Reilly, J. X., Jbabdi, S., and Behrens, T. E. (2012). How can a Bayesian approach in the insula. J. Comp. Neurol. 117, 107–115. doi: 10.1002/cne.90117
inform neuroscience? Eur. J. Neurosci. 35, 1169–1179. doi: 10.1111/j.1460-9568. 0109
2012.08010.x Simmons, D. R., Robertson, A. E., McKay, L. S., Toal, E., McAleer, P., and Pollick,
O’Reilly, R. C. (2006). Biologically based computational models of high-level F. E. (2009). Vision in autism spectrum disorders. Vision Res. 49, 2705–2739.
cognition. Science 314, 91–94. doi: 10.1126/science.1127242 doi: 10.1016/j.visres.2009.08.005
O’Reilly, R. C., and Frank, M. J. (2006). Making working memory work: a Simonyan, K., and Horwitz, B. (2011). Laryngeal motor cortex and control of
computational model of learning in the prefrontal cortex and basal ganglia. speech in humans. Neuroscientist 17, 197–208. doi: 10.1177/107385841038
Neural Comput. 18, 283–328. doi: 10.1162/089976606775093909 6727
Parr, T., and Friston, K. J. (2017a). The active construction of the visual world. Smith, A. T., Singh, K. D., Williams, A. L., and Greenlee, M. W. (2001). Estimating
Neuropsychologia 104, 92–101. doi: 10.1016/j.neuropsychologia.2017.08.003 receptive field size from fMRI data in human striate and extrastriate visual
Parr, T., and Friston, K. J. (2017b). The computational anatomy of visual neglect. cortex. Cereb. Cortex 11, 1182–1190. doi: 10.1093/cercor/11.12.1182
Cereb. Cortex 28, 777–790. doi: 10.1093/cercor/bhx316 Testolin, A., and Zorzi, M. (2016). Probabilistic models and generative neural
Parr, T., and Friston, K. J. (2017c). Uncertainty, epistemics and active inference. networks: towards an unified framework for modeling normal and impaired
J. R. Soc. Interface 14:20170376. doi: 10.1098/rsif.2017.0376 neurocognitive functions. Front. Comput. Neurosci. 10:73. doi: 10.3389/fncom.
Pellicano, E., and Burr, D. (2012). When the world becomes ‘too real’: a Bayesian 2016.00073
explanation of autistic perception. Trends Cogn. Sci. 16, 504–510. doi: 10.1016/ Teunisse, R. J., Zitman, F. G., Cruysberg, J. R., Hoefnagels, W. H., and
j.tics.2012.08.009 Verbeek, A. L. (1996). Visual hallucinations in psychologically normal people:
Price, C., Warburton, E., Moore, C., Frackowiak, R., and Friston, K. (2001). Charles Bonnet’s syndrome. Lancet 347, 794–797. doi: 10.1016/S0140-6736(96)
Dynamic diaschisis: anatomically remote and context-sensitive human brain 90869-7
lesions. J. Cogn. Neurosci. 13, 419–429. doi: 10.1162/08989290152001853 Thiebaut de Schotten, M., Urbanski, M., Duffau, H., Volle, E., Lévy, R., Dubois, B.,
Rao, R. P., and Ballard, D. H. (1999). Predictive coding in the visual cortex: et al. (2005). Direct evidence for a parietal-frontal pathway subserving
a functional interpretation of some extra-classical receptive-field effects. Nat. spatial awareness in humans. Science 309, 2226–2228. doi: 10.1126/science.
Neurosci. 2, 79–87. doi: 10.1038/4580 1116251
Rees, G., Wojciulik, E., Clarke, K., Husain, M., Frith, C., and Driver, J. (2000). Ungerleider, L. G., and Christensen, C. A. (1979). Pulvinar lesions in monkeys
Unconscious activation of visual cortex in the damaged right hemisphere of a produce abnormal scanning of a complex visual array. Neuropsychologia 17,
parietal patient with extinction. Brain 123, 1624–1633. doi: 10.1093/brain/123. 493–501. doi: 10.1016/0028-3932(79)90056-3
8.1624 Ungerleider, L. G., and Haxby, J. V. (1994). ‘What’ and ‘where’ in the
Reichert, D. P., Seriès, P., and Storkey, A. J. (2013). Charles bonnet syndrome: human brain. Curr. Opin. Neurobiol. 4, 157–165. doi: 10.1016/0959-4388(94)
evidence for a generative model in the cortex? PLoS Comput. Biol. 9:e1003134. 90066-3
doi: 10.1371/journal.pcbi.1003134 Ungerleider, L. G., and Mishkin, M. (1982). “Two cortical visual systems,” in
Rizzolatti, G., Fogassi, L., and Gallese, V. (2001). Neurophysiological mechanisms Analysis of Visual Behavior, eds D. Ingle, M. A. Goodale, and J. W. Mansfield
underlying the understanding and imitation of action. Nat. Rev. Neurosci. 2, (Cambridge, MA: MIT Press), 549–586.
661–670. doi: 10.1038/35090060 Valdez, A. B., Papesh, M. H., Treiman, D. M., Smith, K. A., Goldinger, S. D.,
Robinson, D. L., and Petersen, S. E. (1992). The pulvinar and visual salience. Trends and Steinmetz, P. N. (2015). Distributed representation of visual objects by
Neurosci. 15, 127–132. doi: 10.1016/0166-2236(92)90354-B single neurons in the human brain. J. Neurosci. 35, 5180–5186. doi: 10.1523/
Rocca, M. A., Pagani, E., Absinta, M., Valsasina, P., Falini, A., Scotti, G., JNEUROSCI.1958-14.2015
et al. (2007). Altered functional and structural connectivities in patients with von der Heydt, R., and Peterhans, E. (1989). Mechanisms of contour perception
MS: a 3-T study. Neurology 69, 2136–2145. doi: 10.1212/01.wnl.0000295504. in monkey visual cortex. I. Lines of pattern discontinuity. J. Neurosci. 9,
92020.ca 1731–1748.
Rushmore, R. J., Valero-Cabre, A., Lomber, S. G., Hilgetag, C. C., and Payne, B. R. Vuilleumier, P., Hester, D., Assal, G., and Regli, F. (1996). Unilateral spatial neglect
(2006). Functional circuitry underlying visual neglect. Brain 129, 1803–1821. recovery after sequential strokes. Neurology 46, 184–189. doi: 10.1212/WNL.46.
doi: 10.1093/brain/awl140 1.184
Saur, D., Kreher, B. W., Schnell, S., Kümmerer, D., Kellmeyer, P., Vry, M.-S., et al. Wald, A. (1947). An essentially complete class of admissible decision functions.
(2008). Ventral and dorsal pathways for language. Proc. Natl. Acad. Sci. U.S.A. Ann. Math. Stat. 18, 549–555. doi: 10.1214/aoms/1177730345
105, 18035–18040. doi: 10.1073/pnas.0805234105 Warrington, E. K. (1975). The selective impairment of semantic memory. Q. J. Exp.
Schwartenbeck, P., FitzGerald, T. H. B., Mathys, C., Dolan, R., Wurst, F., Psychol. 27, 635–657. doi: 10.1080/14640747508400525
Kronbichler, M., et al. (2015). Optimal inference with suboptimal models: Warrington, E. K., and James, M. (1967). Disorders of visual perception in patients
addiction and active Bayesian inference. Med. Hypotheses 84, 109–117. with localised cerebral lesions. Neuropsychologia 5, 253–266. doi: 10.1016/0028-
doi: 10.1016/j.mehy.2014.12.007 3932(67)90040-1
Schwartenbeck, P., and Friston, K. (2016). Computational phenotyping in Warrington, E. K., and James, M. (1988). Visual apperceptive agnosia: a clinico-
psychiatry: a worked example. eNeuro 3:ENEURO.0049-16.2016. doi: 10.1523/ anatomical study of three cases. Cortex 24, 13–32. doi: 10.1016/S0010-9452(88)
ENEURO.0049-16.2016 80014-5

Frontiers in Human Neuroscience | www.frontiersin.org 13 February 2018 | Volume 12 | Article 61

Parr et al. Computational Neuropsychology

Warrington, E. K., and Shallice, T. (1984). Category specific semantic impairments. Yedidia, J. S., Freeman, W. T., and Weiss, Y. (2005). Constructing free-energy
Brain 107, 829–853. doi: 10.1093/brain/107.3.829 approximations and generalized belief propagation algorithms. IEEE Trans.
Warrington, E. K., and Taylor, A. M. (1973). The contribution of the right parietal Inform. Theory 51, 2282–2312. doi: 10.1109/TIT.2005.850085
lobe to object recognition. Cortex 9, 152–164. doi: 10.1016/S0010-9452(73) Yin, H. H., and Knowlton, B. J. (2006). The role of the basal ganglia in habit
80024-3 formation. Nat. Rev. Neurosci. 7, 464–476. doi: 10.1038/nrn1919
Weiss, Y., Simoncelli, E. P., and Adelson, E. H. (2002). Motion illusions as optimal Yu, A. J., and Dayan, P. (2002). Acetylcholine in cortical inference. Neural Netw.
percepts. Nat. Neurosci. 5, 598–604. doi: 10.1038/nn0602-858 15, 719–730. doi: 10.1016/S0893-6080(02)00058-8
Wernicke, C. (1969). “The symptom complex of aphasia,” in Proceedings of the
Boston Colloquium for the Philosophy of Science 1966/1968, eds R. S. Cohen Conflict of Interest Statement: The authors declare that the research was
and M. W. Wartofsky (Dordrecht: Springer), 34–97. doi: 10.1007/978-94-010- conducted in the absence of any commercial or financial relationships that could
3378-7_2 be construed as a potential conflict of interest.
Winn, J., and Bishop, C. M. (2005). Variational message passing. J. Mach. Learn.
Res. 6, 661–694. Copyright © 2018 Parr, Rees and Friston. This is an open-access article distributed
Yang, S. C.-H., Lengyel, M., and Wolpert, D. M. (2016a). Active sensing in the under the terms of the Creative Commons Attribution License (CC BY). The use,
categorization of visual patterns. eLife 5:e12215. doi: 10.7554/eLife.12215 distribution or reproduction in other forums is permitted, provided the original
Yang, S. C.-H., Wolpert, D. M., and Lengyel, M. (2016b). Theoretical perspectives author(s) and the copyright owner are credited and that the original publication
on active sensing. Curr. Opin. Behav. Sci. 11, 100–108. doi: 10.1016/j.cobeha. in this journal is cited, in accordance with accepted academic practice. No use,
2016.06.009 distribution or reproduction is permitted which does not comply with these terms.

Frontiers in Human Neuroscience | www.frontiersin.org 14 February 2018 | Volume 12 | Article 61