Zuoren Wang

Accumulating evidence suggests that cognitive factors such as rules and attention can modulate sensory processing through top-down influences. However, the neural mechanisms transforming cross-modal inputs into rule-guided perceptions remain unclear. Therefore, in this study, we recorded neuronal activity in the posterior parietal cortex (PPC) and primary auditory cortex (A1) of rats during an audiovisual rule-switching task to investigate how rules are dynamically encoded and modulate sensory processing. By tracking the dynamic neural encoding of internally estimated task rules on a trial-by-trial basis, our single-neuron analyses suggested that the PPC may proactively represent task rules before decision-making, whereas A1 may contribute to monitoring auditory-task performance by enhancing rule-related signals following errors. Both regions showed a preferential representation of conflicting rule combinations, suggesting a specialized mechanism for efficient conflict resolution. Accordingly, we dissociated stimulus-driven responses from rule-modulated responses and observed a within-trial transformation from sensory encoding to rule-weighted representations, progressing from physical features to behavioral relevance, although integration remained limited. Taken together, these results suggest that the PPC–A1 circuit dynamically encodes rules and reshapes sensory processing to implement structured computations for flexible behaviors.

Cooperative behaviors are widespread in nature and play fundamental roles throughout animal lifespans. However, the learning processes and strategy formation underlying cooperation remain largely unknown. Oxytocin (OXT) is a neuropeptide that regulates both social and non-social behaviors, yet its roles in modulating cooperative behaviors remain to be fully elucidated. To investigate these questions, we developed a temporal coordination-based cooperative paradigm for rats and manipulated the OXT system during training. Through progressive cooperative training, rats acquired cooperation and efficiently performed the task via social interaction. This learning process was accompanied by increased OXT release in social learning-associated brain regions, whereas OXT knockout impaired learning speed. Furthermore, rats displayed different cooperative strategies, with the communication-based strategy gradually increasing and becoming predominant during the cooperative training. Both OXT knockout and inhibition of OXT-ergic neurons in the paraventricular nucleus (PVN) of the hypothalamus during learning reduced the adoption of communication-based strategy. Together, our results demonstrate that OXT is necessary for both the acquisition and the strategy formation of cooperation in rats, thereby facilitating mutual rewards. These findings provide new insights into the neural mechanisms by which OXT regulates social behaviors and offer potential therapeutic implications for neuropsychiatric disorders involving social deficits, such as autism spectrum disorder and social anxiety disorder.

The functional role of glutamatergic (vGluT2) neurons in the pedunculopontine nucleus (PPN) in modulating motor activity remains controversial. Here, we demonstrated that the activity of vGluT2 neurons in the rostral PPN is correlated with locomotion and ipsilateral head-turning. Beyond these motor functions, we found that these rostral PPN-vGluT2 neurons remarkably respond to salient stimuli. Furthermore, we systematically traced the upstream and downstream projections of these neurons and identified two downstream projections from these neurons to the caudal pontine reticular nucleus/anterior gigantocellular reticular nucleus (PnC/GiA) and the zona incerta (ZI). Our findings indicate that the projections to the PnC/GiA inhibit movement, consistent with ‘pause-and-play’ behavior, whereas those to the ZI promote locomotion, and others respond to a new ‘pause-switch-play’ pattern. Collectively, these findings elucidate the multifaceted influence of the PPN on motor functions and provide a robust theoretical framework for understanding its physiological and potential therapeutic implications.

Despite the widespread adoption of Backpropagation algorithm-based Deep Neural Networks, the biological infeasibility of the BP algorithm could potentially limit the evolution of new DNN models. To find a biologically plausible algorithm to replace BP, we focus on the top-down mechanism inherent in the biological brain. Although top-down connections in the biological brain play crucial roles in high-level cognitive functions, their application to neural network learning remains unclear. This study proposes a two-level training framework designed to train a bottom-up network using a Top-Down Credit Assignment Network (TDCA-network). The TDCA-network serves as a substitute for the conventional loss function and the back-propagation algorithm, widely used in neural network training. We further introduce a brain-inspired credit diffusion mechanism, significantly reducing the TDCA-network's parameter complexity, thereby greatly accelerating training without compromising the network's performance.Our experiments involving non-convex function optimization, supervised learning, and reinforcement learning reveal that a well-trained TDCA-network outperforms back-propagation across various settings. The visualization of the update trajectories in the loss landscape indicates the TDCA-network's ability to bypass local minima where BP-based trajectories typically become trapped. The TDCA-network also excels in multi-task optimization, demonstrating robust generalizability across different datasets in supervised learning and unseen task settings in reinforcement learning. Moreover, the results indicate that the TDCA-network holds promising potential to train neural networks across diverse architectures.

Neural oscillations play a role in sensory processing by coordinating synchronized neuronal activity. Synchronization of gamma oscillations is engaged in local computation of feedforward signals and synchronization of alpha-beta oscillations is engaged in feedback processing over long-range areas. These spatially and spectrally segregated bi-directional signals may be integrated by a mechanism of cross-frequency coupling. Synchronization of neural oscillations has also been proposed as a mechanism for information integration across multiple sensory modalities. A transient stimulus or rhythmic stimulus from one modality may lead to phase alignment of ongoing neural oscillations in multiple sensory cortices, through a mechanism of cross-modal phase reset or cross-modal neural entrainment. Synchronized activities in multiple sensory cortices are more likely to boost stronger activities in downstream areas. Compared to synchronized oscillations, asynchronized oscillations may impede signal processing, and may contribute to sensory selection by setting the oscillations in the target-related cortex and the oscillations in the distractor-related cortex to opposite phases. Keywords Neural oscillation Multisensory processing Phase reset Neural entrainment Cross-frequency coupling Sensory selection Phase locking Spike phase

Lateral interaction in the biological brain is a key mechanism that underlies higher cognitive functions. Linear self‐organising map (SOM) introduces lateral interaction in a general form in which signals of any modality can be used. Some approaches directly incorporate SOM learning rules into neural networks, but incur complex operations and poor extendibility. The efficient way to implement lateral interaction in deep neural networks is not well established. The use of Laplacian Matrix‐based Smoothing (LS) regularisation is proposed for implementing lateral interaction in a concise form. The authors’ derivation and experiments show that lateral interaction implemented by SOM model is a special case of LS‐regulated k‐means, and they both show the topology‐preserving capability. The authors also verify that LS‐regularisation can be used in conjunction with the end‐to‐end training paradigm in deep auto‐encoders. Additionally, the benefits of LS‐regularisation in relaxing the requirement of parameter initialisation in various models and improving the classification performance of prototype classifiers are evaluated. Furthermore, the topologically ordered structure introduced by LS‐regularisation in feature extractor can improve the generalisation performance on classification tasks. Overall, LS‐regularisation is an effective and efficient way to implement lateral interaction and can be easily extended to different models.