The neuroconnectionist research programme

Doerig, Adrien; Sommers, Rowan P.; Seeliger, Katja; Richards, Blake A; Ismael, Jenann; Lindsay, Grace W.; Kording, Konrad; Konkle, Talia; Gerven, Marcel van; Kriegeskorte, Nikolaus; Kietzmann, Tim C.

doi:10.1038/s41583-023-00705-w

Cited by 114 publications

(60 citation statements)

References 267 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Bowers et al's long list of cited studies presenting shortcomings of particular models neither demonstrates the failure of the ANN modeling framework in general nor a lack of openness of the field to falsifications of ANN models. Instead, their list of citations rather impressively illustrates the opposite: That the emerging ANN research program (referred to as "neuroconnectionism" in Doerig et al, 2022) is progressive in the sense of Lakatos: It generates a rich variety of falsifiable hypotheses (expressed in the language of ANNs) and advances through model comparison (Doerig et al, 2022). Each shortcoming drives improvement.…”

Section: Introductionmentioning

confidence: 99%

Deep problems with neural network models of human vision

Bowers¹,

Dujmović²,

Montero³

et al. 2022

Behav Brain Sci

View full text Add to dashboard Cite

Deep neural networks (DNNs) have had extraordinary successes in classifying photographic images of objects and are often described as the best models of biological vision. This conclusion is largely based on three sets of findings: (1) DNNs are more accurate than any other model in classifying images taken from various datasets, (2) DNNs do the best job in predicting the pattern of human errors in classifying objects taken from various behavioral datasets, and (3) DNNs do the best job in predicting brain signals in response to images taken from various brain datasets (e.g., single cell responses or fMRI data). However, these behavioral and brain datasets do not test hypotheses regarding what features are contributing to good predictions and we show that the predictions may be mediated by DNNs that share little overlap with biological vision. More problematically, we show that DNNs account for almost no results from psychological research. This contradicts the common claim that DNNs are good, let alone the best, models of human object recognition. We argue that theorists interested in developing biologically plausible models of human vision need to direct their attention to explaining psychological findings. More generally, theorists need to build models that explain the results of experiments that manipulate independent variables designed to test hypotheses rather than compete on making the best predictions. We conclude by briefly summarizing various promising modelling approaches that focus on psychological data.

show abstract

Section: Introductionmentioning

confidence: 99%

Deep problems with neural network models of human vision

Bowers¹,

Dujmović²,

Montero³

et al. 2022

Behav Brain Sci

View full text Add to dashboard Cite

show abstract

“…Initially random in high-dimensional space, the hidden layers' representations become organized to distinguish class instances effectively through learning. This process mirrors the brain's processing strategy and suggests shared computational principles between natural and artificial systems (58)(59)(60). Moreover, both systems might cluster similar sensory data to optimize energy use, adhering to environmental and computational limits, thus minimizing extraneous exploration of the sensory state space in favor of more streamlined information processing (61)(62)(63)(64).…”

Section: Discussionmentioning

confidence: 82%

Predictive learning shapes the representational geometry of the human brain

Greco,

Moser,

Preissl

et al. 2024

Preprint

View full text Add to dashboard Cite

Predictive coding theories propose that the brain constantly updates its internal models of the world to minimize prediction errors and optimize sensory processing. However, the neural mechanisms that link the encoding of prediction errors and optimization of sensory representations remain unclear. Here, we provide direct evidence how predictive learning shapes the representational geometry of the human brain. We recorded magnetoencephalography (MEG) in human participants listening to acoustic sequences with different levels of regularity. Representational similarity analysis revealed how, through learning, the brain aligned its representational geometry to match the statistical structure of the sensory inputs, by clustering the representations of temporally contiguous and predictable stimuli. Crucially, we found that in sensory areas the magnitude of the representational shift correlated with the encoding strength of prediction errors. Furthermore, using partial information decomposition we found that, prediction errors were processed by a synergistic network of high-level associative and sensory areas. Importantly, the strength of synergistic encoding of precition errors predicted the magnitude of representational alignment during learning. Our findings provide evidence that large-scale neural interactions engaged in predictive processing modulate the representational content of sensory areas, which may enhance the efficiency of perceptual processing in response to the statistical regularities of the environment.

show abstract

“…In both of these cases, seemingly complex neural representations can be described with more elementary mathematical algorithms. That said, the fact that the modern incarnation of connectionist networks, deep learning models, provides increasingly rich explanations of neurocognitive phenomena (Churchland & Sejnowski, 2016; Cichy & Kaiser, 2019; Doerig et al, 2023; Kriegeskorte, 2015; Peters & Kriegeskorte, 2021; Saxe et al, 2021; Yamins & DiCarlo, 2016) attests that the correspondence between the neural and algorithmic levels may be more than just an analogy (notwithstanding various objections: Arshavsky, 2023; Bowers et al, 2022; Pang et al, 2023). It thus remains to be determined how deep into the brain the controllosphere metaphor can be pressed.…”

Section: Discussionmentioning

confidence: 99%

The controllosphere: The neural origin of cognitive effort.

Holroyd

2024

Psychological Review

View full text Add to dashboard Cite

The neuroconnectionist research programme

Cited by 114 publications

References 267 publications

Deep problems with neural network models of human vision

Deep problems with neural network models of human vision

Predictive learning shapes the representational geometry of the human brain

The controllosphere: The neural origin of cognitive effort.

Contact Info

Product

Resources

About