Natural constraints explain working memory capacity limitations in sensory-cognitive models

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: March 30, 2023

Abstract The limited capacity of the brain to retain information in working memory has been well-known and studied for decades, yet root this limitation remains unclear. Here we built sensory-cognitive neural network models that perform tasks using raw visual stimuli. Contrary intuitions stems from or cognitive constraints, found pre-training sensory region our with natural images imposes sufficient constraints on exhibit a wide range human-like behaviors designed probe capacity. Examining mechanisms model reveals mainly arises bottom-up manner. Our offer principled functionally grounded explanation without parameter fitting behavioral data much hyperparameter tuning. This work highlights importance developing realistic processing even when investigating other high-level phenomena.

Language: Английский

---

The Secondary Motor Cortex-striatum Circuit Contributes to Suppressing Inappropriate Responses in Perceptual Decision Behavior

Neuroscience Bulletin, Journal Year: 2023, Volume and Issue: 39(10), P. 1544 - 1560

Published: May 31, 2023

The secondary motor cortex (M2) encodes choice-related information and plays an important role in cue-guided actions. M2 neurons innervate the dorsal striatum (DS), which also contributes to decision-making behavior, yet how modulates signals DS influence perceptual is unclear. Using mice performing a visual Go/No-Go task, we showed that inactivating projections impaired performance by increasing false alarm (FA) rate reward-irrelevant No-Go stimulus. choice signal of correlated with behavioral performance, inactivation projecting reduced DS. By measuring manipulating responses direct or indirect pathway striatal defined inputs, found exhibited shorter response latency stimulus, their early increased FA rate. These results demonstrate M2-to-DS crucial for suppressing inappropriate decision behavior.

Language: Английский

---

Visual experience induces 4–8 Hz synchrony between V1 and higher-order visual areas

Cell Reports, Journal Year: 2023, Volume and Issue: 42(12), P. 113482 - 113482

Published: Nov. 22, 2023

Visual perceptual experience induces persistent 4-8 Hz oscillations in the mouse primary visual cortex (V1), encoding familiarity. Recent studies suggest that higher-order areas (HVAs) are functionally specialized and segregated into information streams processing distinct features. However, whether memories processed stored within is not understood. We report here V1 lateromedial (LM), but anterolateral, become more phase synchronized after entrainment of stimulus maximally responses LM. Directed analysis reveals changes top-down functional connectivity between HVAs. Optogenetic inactivation LM reduces post-stimulus oscillation peaks impairs discrimination behavior. Our results demonstrate familiarity-evoked specific for features present corresponding HVAs, where they may be used inter-areal communication with during memory-related behaviors.

Language: Английский

---

Wide-field imaging in behaving mice as a tool to study cognitive function

Neurophotonics, Journal Year: 2024, Volume and Issue: 11(03)

Published: Feb. 19, 2024

Cognitive functions are mediated through coordinated and dynamic neuronal responses that involve many different areas across the brain. Therefore, it is of high interest to simultaneously record activity from as brain possible while subject performs a cognitive behavioral task. One emerging tools achieve mesoscopic field view wide-field imaging cortex-wide dynamics in mice. Wide-field cost-effective, user-friendly, enables obtaining signals mice performing complex demanding tasks. Importantly, offers an unbiased observation sheds light on overlooked cortical regions highlights parallel processing circuits. Recent studies have shown multi-area patterns, rather than just single area, involved encoding functions. The optical properties enable signals, such layer-specific, inhibitory subtypes, or neuromodulation signals. Here, I review main advantages mice, recent literature, discuss future directions field. It expected behaving will continue gain popularity aid understanding mesoscale underlying function.

Language: Английский

---

Predictive learning shapes the representational geometry of the human brain

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: March 7, 2024

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

Language: Английский

---

Localized and global representation of prior value, sensory evidence, and choice in male mouse cerebral cortex

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: May 22, 2024

Abstract Adaptive behavior requires integrating prior knowledge of action outcomes and sensory evidence for making decisions while maintaining future actions. As outcome- sensory-based are often tested separately, it is unclear how these processes integrated in the brain. In a tone frequency discrimination task with two sound durations asymmetric reward blocks, we found that neurons medial prefrontal cortex male mice represented additive combination expectations choices. The inputs choices were selectively decoded from auditory irrespective priors secondary motor cortex, respectively, suggesting localized computations variables required within single trials. contrast, all recorded regions values needed to be maintained across We propose global different time scales cerebral cortex.

Language: Английский

---

Neural interactions in the human frontal cortex dissociate reward and punishment learning

eLife, Journal Year: 2024, Volume and Issue: 12

Published: June 28, 2024

How human prefrontal and insular regions interact while maximizing rewards minimizing punishments is unknown. Capitalizing on intracranial recordings, we demonstrate that the functional specificity toward reward or punishment learning better disentangled by interactions compared to local representations. Prefrontal cortices display non-selective neural populations punishments. Non-selective responses, however, give rise context-specific interareal interactions. We identify a subsystem with redundant between orbitofrontal ventromedial cortices, driving role of latter. In addition, find dorsolateral insula. Finally, switching mediated synergistic two subsystems. These results provide unifying explanation distributed cortical representations supporting learning.

Language: Английский

---

A computational mechanism of cue-stimulus integration for pain in the brain

Science Advances, Journal Year: 2024, Volume and Issue: 10(37)

Published: Sept. 11, 2024

The brain integrates information from pain-predictive cues and noxious inputs to construct the pain experience. Although previous studies have identified neural encodings of individual components, how they are integrated remains elusive. Here, using a cue-induced task, we examined temporal functional magnetic resonance imaging activities within state space, where axes represent voxel activities. By analyzing features these at large-scale network level, demonstrated that overall networks preserve both cue stimulus in their respective subspaces space. However, only higher-order networks, including limbic default mode could reconstruct pattern participants’ reported by linear summation subspace activities, providing evidence for integration information. These results suggest hierarchical organization processing components elucidate mechanism underlying our perception.

Language: Английский

---

Predictive learning shapes the representational geometry of the human brain

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: Nov. 8, 2024

Predictive coding theories propose that the brain constantly updates internal models to minimize prediction errors and optimize sensory processing. However, neural mechanisms link error encoding optimization of representations remain unclear. Here, we provide evidence how predictive learning shapes representational geometry human brain. We recorded magnetoencephalography (MEG) in humans listening acoustic sequences with different levels regularity. found aligns its match statistical structure inputs, by clustering temporally contiguous predictable stimuli. Crucially, magnitude this shift correlates synergistic a network high-level areas. Our findings suggest that, response regularities environment, large-scale interactions engaged processing modulate content areas enhance

Language: Английский

---

Distributed representations of prediction error signals across the cortical hierarchy are synergistic

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 13, 2023

Abstract An important question concerning inter-areal communication in the cortex is whether these interactions are synergistic, i.e. brain signals can either share common information (redundancy) or they encode complementary that only available when both considered together (synergy). Here, we dissociated cortical sharing from those encoding during prediction error processing. To this end, computed co-information, an information-theoretical measure distinguishes redundant synergistic among signals. We analyzed auditory and frontal electrocorticography (ECoG) five awake marmosets performing two distinct oddball tasks investigated to what extent event-related potentials (ERP) broadband (BB) dynamics encoded In tasks, observed multiple patterns of synergy across entire hierarchy with dynamics. The conveyed by ERPs BB was highly even at lower stages cortex, as well between regions. Using a brain-constrained neural network, simulated spatio-temporal redundancy experimental results further demonstrated emergence regions requires presence strong, long-distance, feedback feedforward connections. These indicate distributed representations be synergistic.

Language: Английский

---