Recent findings indicate significant variations in neuronal activity timescales across and within cortical areas, yet their impact on cognitive processing remains inadequately understood. This study explores the role of neurons with different timescales in information processing within the neural system, particularly during the execution of context-dependent working memory tasks. Especially, we hypothesized that neurons with varying timescales contribute distinctively to task performance by forming diverse representations of task-relevant information. To test this, the model was trained to perform a context-dependent working memory task with a machine-learning technique. Results revealed that slow timescale neurons maintained stable representations of contextual information throughout the trial, whereas fast timescale neurons responded transiently to immediate stimuli. This differentiation in neuronal function suggests a fundamental role for timescale diversity in supporting the neural system’s ability to integrate and process information dynamically. Our findings contribute to understanding how neural timescale diversity underpins cognitive flexibility and task-specific information processing, highlighting implications for both theoretical neuroscience and practical applications in designing artificial neural networks. ### Competing Interest Statement The authors have declared no competing interest.

Learning in neural systems occurs through change in synaptic connectivity that is driven by neural activity. Learning performance is influenced by both neural activity and the task to be learned. Experimental studies suggest a link between learning speed and variability in neural activity before learning. However, the theoretical basis of this relationship has remained unclear. In this work, using principles from the fluctuation-response relation in statistical physics, we derive two formulae that connect neural activity with learning speed. The first formula shows that learning speed is proportional to the variance of spontaneous neural activity and the neural response to input. The second formula, for small input, indicates that speed is proportional to the variances of spontaneous activity in both target and input directions. These formulae apply to various learning tasks governed by Hebbian or generalized learning rules. Numerical simulations confirm that these formulae are valid beyond their theoretical assumptions, even in cases where synaptic connectivity undergoes large changes. Our theory predicts that learning speed increases with the gain of neuronal activation functions and the number of pre-embedded memories, as both enhance the variance of spontaneous neural fluctuations. Additionally, the formulae reveal which input/output relationships are easier to learn, aligning with experimental data. Thus, our results provide a theoretical foundation for the quantitative relationship between pre-learning neural activity fluctuations and learning speed, offering insights into a range of empirical observations.

The frontal cortex-striatum circuit plays a pivotal role in adaptive goal-directed behaviors. However, it remains unclear how decision-related signals are mediated through cross-regional transmission between the medial frontal cortex and the striatum by neuronal ensembles in making decision based on outcomes of past action. Here, we analyzed neuronal ensemble activity obtained through simultaneous multiunit recordings in the secondary motor cortex (M2) and dorsal striatum (DS) in rats performing an outcome-based left- or-right choice task. By adopting tensor component analysis (TCA), a single-trial-based unsupervised dimensionality reduction approach, for concatenated ensembles of M2 and DS neurons, we identified distinct three spatiotemporal neural dynamics (TCA components) at the single-trial level specific to task- relevant variables. Choice-position-selective neural dynamics reflected the positions chosen and was correlated with the trial-to-trial fluctuation of behavioral variables. Intriguingly, choice-pattern-selective neural dynamics distinguished whether the incoming choice was a repetition or a switch from the previous choice before a response choice. Other neural dynamics was selective to outcome and increased within- trial activity following response. Our results demonstrate how the concatenated ensembles of M2 and DS process distinct features of decision-related signals at various points in time. Thereby, the M2 and DS collaboratively monitor action outcomes and determine the subsequent choice, whether to repeat or switch, for action selection.

The frontal cortex-striatum circuit plays a pivotal role in adaptive goal-directed behaviours. However, the mediation of decision-related signals through cross-regional transmission between the medial frontal cortex and the striatum by neuronal ensembles remains unclear. We analysed neuronal ensemble activity obtained through simultaneous multiunit recordings in the secondary motor cortex (M2) and dorsal striatum (DS) while the rats performed an outcome-based choice task. Tensor component analysis (TCA), an unsupervised dimensionality reduction approach at the single-trial level, was adopted for concatenated ensembles of M2 and DS neurons. We identified distinct three spatiotemporal neural dynamics (TCA components) at the single-trial level specific to task-relevant variables. Choice-position selective neural dynamics was correlated with the trial-to-trial fluctuation of behavioural variables. This analytical approach unveiled choice-pattern selective neural dynamics distinguishing whether the incoming choice was a repetition or switch from the previous choice. Other neural dynamics was selective to outcome. Choice-pattern selective within-trial activity increased before response choice, whereas outcome selective within-trial activity increased following response. These results suggest that the concatenated ensembles of M2 and DS process distinct features of decision-related signals at various points in time. The M2 and DS may collaboratively monitor action outcomes and determine the subsequent choice, whether to repeat or switch, for coordinated action selection. ### Competing Interest Statement The authors have declared no competing interest.

Excitatory/inhibitory (E/I) balance is significantly associated with cognitive function. Its imbalance impairs cognitive function, particularly in patients with psychiatric disorders. Recent physiological and modeling findings show that excitatory postsynaptic potentials (EPSPs) have a long-tailed distribution and contribute to the generation of spontaneous activity. Moreover, this spontaneous activity and its response to the external stimulus significantly alternate under the different E/I balance. However, the effects of the E/I balance under long-tailed EPSPs at the functional level remain unknown. Hence, to elucidate this relationship, we constructed a reservoir computing (RC) model generating the long-tailed distribution of EPSPs and investigated the effect of the E/I balance on the learning performance of RC in the memory capacity (MC) task, which measures how correctly delayed input signals can be reproduced. The results revealed that an appropriate E/I balance maximized the MC. This high MC was realized by recurrent spike propagation under long-tailed EPSPs. These findings contribute to the understanding of the effect of the E/I balance in physiologically relevant neural networks.

In neural information processing, inputs modulate neural dynamics to generate desired outputs. To unravel the dynamics and underlying neural connectivity enabling such input-output association, we propose an exactly solvable neural-network model with a connectivity matrix explicitly consisting of inputs and required outputs. An analytic form of the response under the input is derived, while three distinctive types of responses including chaotic dynamics are obtained as distinctive bifurcations against input strength, depending on the neural sensitivity and number of inputs. Optimal performance is achieved at the onset of chaos. The relevance of the results to cognitive dynamics is discussed.

In the cerebral cortex, excitatory postsynaptic potentials (EPSPs) exhibit a long-tailed distribution. Although EPSPs induce rich neural activity, their contributions to brain function remain unclear. Therefore, this study evaluated the effect of the dynamics induced by long-tailed synaptic weights by constructing a reservoir computing (RC) model and comparing the memory capacity and predictive accuracy for nonlinear time-series between RCs, with and without strong weights. The results revealed that strong weights significantly enhance the RC performance through gamma-band dynamic neural activity. This mechanism may support the cognitive processes in the actual brain network.

Transitions between metastable states are commonly observed in the neural system and underlie various cognitive functions such as working memory. In a previous study, we have developed a neural network model with the slow and fast populations, wherein simple Hebb-type learning enables stable and complex (e.g., non-Markov) transitions between neural states. This model is distinct from a network with asymmetric Hebbian connectivity and a network trained with supervised machine learning methods: the former generates simple Markov sequences. The latter generates complex but vulnerable sequences against perturbation and its learning methods are biologically implausible. By using our model, we propose and demonstrate a novel mechanism underlying stable working memories: sequentially stabilizing and destabilizing task-related states in the fast neural dynamics. The slow dynamics maintain a history of the applied inputs, e.g., context signals, and enable the task-related states to be stabilized in a context-dependent manner. We found that only a single (or a few) state(s) is stabilized in each epoch (i.e., a period in the presence of the context signal and a delayed period) in a working memory task, resulting in a robust performance against noise and change in a task protocol. These results suggest a simple mechanism underlying complex and stable processing in neural systems.

Collective dynamics of the neural population are involved in a variety of cognitive functions. How such neural dynamics are shaped through learning and how the learning performance is related to the property of the individual neurons are fundamental questions in neuroscience. Previous model studies answered these questions by using developing machine-learning techniques for training a recurrent neural network. However, these techniques are not biologically plausible. Does another type of learning method, for instance, a more biologically plausible learning method, shape the similar neural dynamics and the similar relation between the learning performance and the property of the individual neurons to those observed in the previous studies? In this study, we have used the recently proposed learning model with multiple timescales in the neural activity, which is more biologically plausible, and analyzed the neural dynamics and the relation regarding the sensitivity of neurons. As result, we have found that our model shapes similar neural dynamics and the relation. Further, the intermediate sensitivity of neurons that is optimal for the learning speed generates a variety of neural activity patterns in line with the experimental observations in the neural system. This result suggests that the neural system might develop the sensitivity of neural activities to optimize the learning speed through evolution.

Sequential transitions between metastable states are ubiquitously observed in the neural system and underlying various cognitive functions such as perception and decision making. Although a number of studies with asymmetric Hebbian connectivity have investigated how such sequences are generated, the focused sequences are simple Markov ones. On the other hand, fine recurrent neural networks trained with supervised machine learning methods can generate complex non-Markov sequences, but these sequences are vulnerable against perturbations and such learning methods are biologically implausible. How stable and complex sequences are generated in the neural system still remains unclear. We have developed a neural network with fast and slow dynamics, which are inspired by the hierarchy of timescales on neural activities in the cortex. The slow dynamics store the history of inputs and outputs and affect the fast dynamics depending on the stored history. We show that the learning rule that requires only local information can form the network generating the complex and robust sequences in the fast dynamics. The slow dynamics work as bifurcation parameters for the fast one, wherein they stabilize the next pattern of the sequence before the current pattern is destabilized depending on the previous patterns. This co-existence period leads to the stable transition between the current and the next pattern in the non-Markov sequence. We further find that timescale balance is critical to the co-existence period. Our study provides a novel mechanism generating robust complex sequences with multiple timescales. Considering the multiple timescales are widely observed, the mechanism advances our understanding of temporal processing in the neural system.

During the execution of working memory tasks, task-relevant information is processed by local circuits across multiple brain regions. How this multiarea computation is conducted by the brain remains largely unknown. To explore such mechanisms in spatial working memory, we constructed a neural network model involving parvalbumin-positive, somatostatin-positive, and vasoactive intestinal polypeptide-positive interneurons in the hippocampal CA1 and the superficial and deep layers of medial entorhinal cortex (MEC). Our model is based on a hypothesis that cholinergic modulations differently regulate information flows across CA1 and MEC at memory encoding, maintenance, and recall during delayed nonmatching-to-place tasks. In the model, theta oscillation coordinates the proper timing of interactions between these regions. Furthermore, the model predicts that MEC is engaged in decoding as well as encoding spatial memory, which we confirmed by experimental data analysis. Thus, our model accounts for the neurobiological characteristics of the cross-area information routing underlying working memory tasks.

The generation of robust sequential patterns that depend flexibly on the previous history of inputs and outputs is essential to temporal information processing with working memory in our neural system. We propose a neural network with two timescales, in which a sequence of fixed points of the fast dynamics is generated through bifurcations by slow dynamics as a control parameter. By adopting a simple, biologically plausible learning rule, the neural network can recall a complex context-dependent sequence. Considering multiple timescales experimentally observed in cortical areas, this study provides a general scheme to temporal processing in the brain.

Memories in neural system are shaped through the interplay of neural and learning dynamics under external inputs. By introducing a simple local learning rule to a neural network, we found that the memory capacity is drastically increased by sequentially repeating the learning steps of input-output mappings. The origin of this enhancement is attributed to the generation of a Psuedo-inverse correlation in the connectivity. This is associated with the emergence of spontaneous activity that intermittently exhibits neural patterns corresponding to embedded memories. Stablization of memories is achieved by a distinct bifurcation from the spontaneous activity under the application of each input.

In the brain, decision making is instantiated in dedicated neural circuits. However, there is considerable individual variability in decision-making behavior, particularly under uncertainty. The origins of decision variability within these conserved neural circuits are not known. Here we demonstrate in the rat medial frontal cortex (MFC) that individual variability is a consequence of altered stability in neuronal populations. In a sensory-guided choice task, rats trained on familiar stimuli were exposed to unfamiliar stimuli, resulting in variable choice responses across individuals. We created a recurrent network model to examine the source of variability in MFC neurons, and found that the landscape of neural population trajectories explained choice variability across different unfamiliar stimuli. We experimentally confirmed model predictions showing that trial-by-trial variability in neuronal activity indexes the landscape and predicts individual variation. These results show that neural stability is a critical component of the MFC neural dynamics that underpins individual variation in decision-making.

Decision making obeys common neural mechanisms, but there is considerable variability in individuals’ decision making behavior particularly under uncertainty. How individual differences arise within common decision making brain systems is not known. Here, we explored this question in the medial frontal cortex (MFC) of rats performing a sensory-guided choice task. When rats trained on familiar stimuli were exposed to unfamiliar stimuli, choice responses varied significantly across individuals. We examined how variability in MFC neural processing could mediate this individual difference and constructed a network model to replicate this. Our model suggested that susceptibility of neural trajectories is a crucial determinant of the observed choice variability. The model predicted that trial-by-trial variability of trajectories are correlated with the susceptibility, and hence also correlated with the individual difference. This prediction was confirmed by experiment. Thus, our results suggest that variability in neural dynamics in MFC networks underlies individual differences in decision making.