Unraveling the complexity of rat object vision requires a full convolutional network and beyond

Patterns, Год журнала: 2025, Номер 6(2), С. 101149 - 101149

Опубликована: Янв. 18, 2025

Despite their prominence as model systems of visual functions, it remains unclear whether rodents are capable truly advanced processing information. Here, we used a convolutional neural network (CNN) to measure the computational complexity required account for rat object vision. We found that ability discriminate objects despite scaling, translation, and rotation was well accounted by CNN mid-level layers. However, tolerance displayed rats more severe image manipulations (occlusion reduction outlines) achieved only in final Moreover, deployed perceptual strategies were invariant than those CNN, they consistently relied on same set diagnostic features across transformations. These results reveal an unexpected level sophistication vision, while reinforcing intuition CNNs learn solutions marginally match biological systems.

Язык: Английский

---

Text-related functionality and dynamics of visual human pre-frontal activations revealed through neural network convergence

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Апрель 2, 2024

Summary The functional role of visual activations human pre-frontal cortex remains a deeply debated question. Its significance extends to fundamental issues localization and global theories consciousness. Here we addressed this question by comparing, dynamically, the potential parallels between relational structure prefrontal textual-trained deep neural networks (DNNs). frontal structures were revealed in intra-cranial recordings patients, conducted for clinical purposes, while patients viewed familiar images faces places. Our results reveal that were, surprisingly, predicted text not DNNs. Importantly, temporal dynamics these correlations showed striking differences, with rapid decline over time component, but persistent including significant image offset response component. point dynamic text-related function responses brain.

Язык: Английский

---

Brain–machine convergent evolution: Why finding parallels between brain and artificial systems is informative

Proceedings of the National Academy of Sciences, Год журнала: 2024, Номер 121(41)

Опубликована: Окт. 2, 2024

Central nervous system neurons manifest a rich diversity of selectivity profiles-whose precise role is still poorly understood. Following the striking success artificial networks, major debate has emerged concerning their usefulness in explaining neuronal properties. Here we propose that finding parallels between and networks informative precisely because these systems are so different from each other. Our argument based on an extension concept convergent evolution-well established biology-to domain systems. Applying this to areas levels cortical hierarchy can be powerful tool for elucidating functional well-known selectivities. Importantly, further demonstrate such uncover novel functionalities by showing grid cells entorhinal cortex modeled function as set basis functions lossy representation JPEG compression. Thus, contrary common intuition, here illustrate with provides insights, particularly those cases far removed realistic brain biology.

Язык: Английский

---

Multivariate analysis of brain activity patterns as a tool to understand predictive processes in speech perception

Language Cognition and Neuroscience, Год журнала: 2023, Номер 39(9), С. 1117 - 1133

Опубликована: Янв. 18, 2023

Speech perception is heavily influenced by our expectations about what will be said. In this review, we discuss the potential of multivariate analysis as a tool to understand neural mechanisms underlying predictive processes in speech perception. First, advantages approaches and they have added understanding processing from acoustic-phonetic form speech, over syllable identity syntax, its semantic content. Second, suggest that using techniques measure informational content across hierarchically organised speech-sensitive brain areas might enable us specify which prior knowledge sensory signals are combined. Specifically, approach allow decode how different priors, e.g. speaker's voice or topic current conversation, represented at stages incoming result differently represented.

Язык: Английский

---

Network Communications Flexibly Predict Visual Contents That Enhance Representations for Faster Visual Categorization

Journal of Neuroscience, Год журнала: 2023, Номер 43(29), С. 5391 - 5405

Опубликована: Июнь 27, 2023

Models of visual cognition generally assume that brain networks predict the contents a stimulus to facilitate its subsequent categorization. However, understanding prediction and categorization at network level has remained challenging, partly because we need reverse engineer their information processing mechanisms from dynamic neural signals. Here, used connectivity measures can isolate communications specific content reconstruct these in each individual participant ( N = 11, both sexes). Each was cued spatial location (left vs right) [low frequency (LSF) high (HSF)] predicted Gabor they then categorized. Using participant's concurrently measured MEG, reconstructed categorize LSF versus HSF for behavior. We found flexibly propagate top down temporal lateralized occipital cortex, depending on task demands, under supervisory control prefrontal cortex. When reach predictions enhance bottom-up representations stimulus, all way occipital-ventral-parietal premotor turn producing faster Importantly, are subsets (i.e., 55–75%) signal-to-signal typically between regions. Hence, our study isolates functional process cognitive functions. SIGNIFICANCE STATEMENT An enduring hypothesis states perception is influenced by sensory input but also top-down expectations. explanations according task-demands remain elusive. addressed them predictive experimental design isolating other communications. Our methods revealed Prediction Network communicates with explicit frontal control, an occipital-ventral-parietal-frontal Categorization represents more sharply shown leading framework results therefore shed new light activity.

Язык: Английский

---

Strength of predicted information content in the brain biases decision behavior

Current Biology, Год журнала: 2023, Номер 33(24), С. 5505 - 5514.e6

Опубликована: Дек. 1, 2023

Prediction-for-perception theories suggest that the brain predicts incoming stimuli to facilitate their categorization.1Smith F.W. Muckli L. Nonstimulated early visual areas carry information about surrounding context.Proc. Natl. Acad. Sci. USA. 2010; 107: 20099-20103Crossref PubMed Scopus (120) Google Scholar,2Uran C. Peter A. Lazar Barnes W. Klon-Lipok J. Shapcott K.A. Roese R. Fries P. Singer Vinck M. Predictive coding of natural images by V1 firing rates and rhythmic synchronization.Neuron. 2022; 110: 1240-1257.e8Abstract Full Text PDF (13) Scholar,3Clark Whatever next? brains, situated agents, future cognitive science.Behav. Brain 2013; 36: 181-204Crossref (2877) Scholar,4Friston K. The free-energy principle: a unified theory?.Nat. Rev. Neurosci. 11: 127-138Crossref (3852) Scholar,5Gilbert C.D. Sigman states: top-down influences in sensory processing.Neuron. 2007; 54: 677-696Abstract (627) Scholar,6Yuille Kersten D. Vision as Bayesian inference: analysis synthesis?.Trends Cogn. 2006; 10: 301-308Abstract (506) Scholar,7Glenberg A.M. What memory is for.Behav. 1997; 20: 1-19Crossref (1092) Scholar,8Ye Z. Shi Li Chen Xue G. Retrieval practice facilitates updating enhancing differentiating medial prefrontal cortex representations.eLife. 2020; 9e57023Crossref (16) Scholar,9De Lange F.P. Heilbron Kok How do expectations shape perception?.Trends 2018; 22: 764-779Abstract (376) Scholar,10Kok Jehee J.F.M. de Less more: expectation sharpens representations primary cortex.Neuron. 2012; 75: 265-270Abstract (441) Scholar,11Bar Kassam K.S. Ghuman A.S. Boshyan Schmid Dale Hämäläinen M.S. Marinkovic Schacter D.L. Rosen B.R. et al.Top-down facilitation recognition.Proc. 103: 449-454Crossref (1166) Scholar,12Stein T. Peelen M.V. Content-specific enhance stimulus detectability increasing perceptual sensitivity.J. Exp. Psychol. Gen. 2015; 144: 1089-1104Crossref Scholar,13Michalareas Vezoli Van Pelt S. Schoffelen J.M. Kennedy H. Alpha-beta gamma rhythms subserve feedback feedforward among human cortical areas.Neuron. 2016; 89: 384-397Abstract (417) Scholar,14Benedek Bergner Könen Fink Neubauer A.C. EEG alpha synchronization related processing convergent divergent thinking.Neuropsychologia. 2011; 49: 3505-3511Crossref (214) Scholar,15Lobier Palva High-alpha band across frontal, parietal mediates behavioral neuronal effects visuospatial attention.NeuroImage. 165: 222-237Crossref (82) Scholar,16Brandman Avancini Leticevscaia O. Auditory semantic cues decoding object category MEG.Cereb. Cortex. 30: 597-606Google Scholar,17Treder Charest I. Michelmann Martín-Buro M.C. Roux F. Carceller-Benito Ugalde-Canitrot Rollings D.T. Sawlani V. Chelvarajah al.The hippocampus switchboard between perception memory.Proc. 2021; 118e2114171118Crossref Scholar However, it remains unknown what contents these predictions are, which hinders mechanistic explanations. This because typical approaches cast an underconstrained contrast two categories18Linde-Domingo Treder Kerrén Wimber Evidence neural flow reversed reconstruction from memory.Nat. Commun. 2019; 179Crossref (56) Scholar,19Dijkstra N. Ambrogioni Vidaurre van Gerven Neural dynamics inference its reversal during imagery.eLife. 9e53588Crossref (29) Scholar,20Kok Mostert De Prior induce prestimulus templates.Proc. 2017; 114: 10473-10478Crossref (162) Scholar,21Lee S.H. Kravitz D.J. Baker C.I. Disentangling imagery real-world objects.NeuroImage. 59: 4064-4073Crossref (143) Scholar,22Hindy N.C. Ng F.Y. Turk-Browne N.B. Linking pattern completion predictive cortex.Nat. 19: 665-667Crossref Scholar,23Dijkstra Bosch S.E. M.A.J. Shared mechanisms imagery.Trends 23: 423-434Abstract (121) Scholar,24Kerrén Linde-Domingo Hanslmayr An optimal oscillatory phase for reactivation retrieval.Curr. Biol. 28: 3383-3392.e6Abstract (53) Scholar—e.g., faces versus cars, could lead features specific or both categories. Here, pinpoint thus brain, we identified enable different categorical perceptions same stimuli. We then trained multivariate classifiers discern, dynamic MEG responses, tied each perception. With auditory cueing design, reveal where, when, how reactivates (versus contrast) before shown. demonstrate have more direct influence (bias) on subsequent decision behavior participants than contrast. Specifically, are precisely localized (lateralized), specifically driven cues, strength presentation exerts greater bias individual participant later categorizes this stimulus. By characterizing processes, our findings provide new insights into brain's prediction

Язык: Английский

---

Whether pattern memory can be truly realized in deep neural network?

Research Square (Research Square), Год журнала: 2024, Номер unknown

Опубликована: Июль 24, 2024

The unknown boundary issue, between superior computational capability of deep neural networks (DNNs) and human cognitive ability, has becoming crucial foundational theoretical problem in AI evolution. Undoubtedly, DNN-empowered is increasingly surpassing intelligence handling general intelligent tasks. However, the absence DNN’s interpretability recurrent erratic behavior remain incontrovertible facts. Inspired by perceptual characteristics vision on optical illusions, we propose a novel working analysis framework for DNNs through innovative response visual illusion images, accompanied with fine adjustable sample image construction strategy. Our findings indicate that, although can infinitely approximate human-provided empirical standards pattern classification, object detection semantic segmentation, they are still unable to truly realize independent memorization. All super abilities purely come from their powerful classification performance similar known scenes. Above discovery establishes new foundation advancing artificial intelligence.

Язык: Английский

---

Pattern memory cannot be completely and truly realized in deep neural networks

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Дек. 30, 2024

The unknown boundary issue, between superior computational capability of deep neural networks (DNNs) and human cognitive ability, has becoming crucial foundational theoretical problem in AI evolution. Undoubtedly, DNN-empowered is increasingly surpassing intelligence handling general intelligent tasks. However, the absence DNN's interpretability recurrent erratic behavior remain incontrovertible facts. Inspired by perceptual characteristics vision on optical illusions, we propose a novel working analysis framework for DNNs through innovative response visual illusion images, accompanied with fine adjustable sample image construction strategy. Our findings indicate that, although can infinitely approximate human-provided empirical standards pattern classification, object detection semantic segmentation, they are still unable to truly realize independent memorization. All super abilities purely come from their powerful classification performance similar known scenes. Above discovery establishes new foundation advancing artificial intelligence.

Язык: Английский

---

Stimulus models test hypotheses in brains and DNNs

Trends in Cognitive Sciences, Год журнала: 2023, Номер 27(3), С. 216 - 217

Опубликована: Янв. 10, 2023

Язык: Английский

---

Advancing naturalistic affective science with deep learning

Опубликована: Янв. 11, 2023

People express their own emotions and perceive others’ via a variety of channels, including facial movements, body gestures, vocal prosody, language. Studying these channels affective behavior offers insight into both the experience perception emotion. Prior research has predominantly focused on studying individual in isolation using tightly controlled, non-naturalistic experiments. This approach limits our understanding emotion more naturalistic contexts where different information tend to interact. Traditional methods struggle address this limitation: manually annotating is time-consuming, making it infeasible do at large scale; selecting manipulating stimuli based hypotheses may neglect unanticipated features, potentially generating biased conclusions; common linear modeling approaches cannot fully capture complex, nonlinear, interactive nature real-life processes. In methodology review, we describe how deep learning can be applied challenges advance science. First, current practices explain why existing face revealing Second, introduce they tackle three main challenges: quantifying behaviors, stimuli, Finally, limitations methods, might avoided or mitigated. By detailing promise peril learning, review aims pave way for

Язык: Английский

---