Guobin Shen

Xiang He, Xiangxi Liu, Yang Li et al.

The audio-visual event localization task requires identifying concurrent visual and auditory events from unconstrained videos within a network model, locating them, and classifying their category. The efficient extraction and integration of audio and visual modal information have always been challenging in this field. In this paper, we introduce CACE-Net, which differs from most existing methods that solely use audio signals to guide visual information. We propose an audio-visual co-guidance attention mechanism that allows for adaptive bi-directional cross-modal attentional guidance between audio and visual information, thus reducing inconsistencies between modalities. Moreover, we have observed that existing methods have difficulty distinguishing between similar background and event and lack the fine-grained features for event classification. Consequently, we employ background-event contrast enhancement to increase the discrimination of fused feature and fine-tuned pre-trained model to extract more refined and discernible features from complex multimodal inputs. Specifically, we have enhanced the model's ability to discern subtle differences between event and background and improved the accuracy of event classification in our model. Experiments on the AVE dataset demonstrate that CACE-Net sets a new benchmark in the audio-visual event localization task, proving the effectiveness of our proposed methods in handling complex multimodal learning and event localization in unconstrained videos. Code is available at https://github.com/Brain-Cog-Lab/CACE-Net.

Xiang He, Dongcheng Zhao, Yang Li et al.

Spiking neural networks (SNNs) are rich in spatio-temporal dynamics and are suitable for processing event-based neuromorphic data. However, event-based datasets are usually less annotated than static datasets. This small data scale makes SNNs prone to overfitting and limits their performance. In order to improve the generalization ability of SNNs on event-based datasets, we use static images to assist SNN training on event data. In this paper, we first discuss the domain mismatch problem encountered when directly transferring networks trained on static datasets to event data. We argue that the inconsistency of feature distributions becomes a major factor hindering the effective transfer of knowledge from static images to event data. To address this problem, we propose solutions in terms of two aspects: feature distribution and training strategy. Firstly, we propose a knowledge transfer loss, which consists of domain alignment loss and spatio-temporal regularization. The domain alignment loss learns domain-invariant spatial features by reducing the marginal distribution distance between the static image and the event data. Spatio-temporal regularization provides dynamically learnable coefficients for domain alignment loss by using the output features of the event data at each time step as a regularization term. In addition, we propose a sliding training strategy, which gradually replaces static image inputs probabilistically with event data, resulting in a smoother and more stable training for the network. We validate our method on neuromorphic datasets, including N-Caltech101, CEP-DVS, and N-Omniglot. The experimental results show that our proposed method achieves better performance on all datasets compared to the current state-of-the-art methods. Code is available at https://github.com/Brain-Cog-Lab/Transfer-for-DVS.

Sicheng Shen, Mingyang Lv, Han Shen et al.

The safety of large language models (LLMs) has increasingly emerged as a fundamental aspect of their development. Existing safety alignment for LLMs is predominantly achieved through post-training methods, which are computationally expensive and often fail to generalize well across different models. A small number of lightweight alignment approaches either rely heavily on prior-computed safety injections or depend excessively on the model's own capabilities, resulting in limited generalization and degraded efficiency and usability during generation. In this work, we propose a safety-aware decoding method that requires only low-cost training of an expert model and employs a single neuron as a gating mechanism. By effectively balancing the model's intrinsic capabilities with external guidance, our approach simultaneously preserves utility and enhances output safety. It demonstrates clear advantages in training overhead and generalization across model scales, offering a new perspective on lightweight alignment for the safe and practical deployment of large language models. Code: https://github.com/Beijing-AISI/NGSD.

Yonghao Yu, Dongcheng Zhao, Guobin Shen et al.

The hierarchical architecture has become a mainstream design paradigm for Vision Transformers (ViTs), with Patch Merging serving as the pivotal component that transforms a columnar architecture into a hierarchical one. Drawing inspiration from the brain's ability to integrate global and local information for comprehensive visual understanding, we propose Stepwise Patch Merging (SPM), which enhances the subsequent attention mechanism's ability to 'see' better. SPM consists of Multi-Scale Aggregation (MSA) and Guided Local Enhancement (GLE) striking a proper balance between long-range dependency modeling and local feature enhancement. Extensive experiments conducted on benchmark datasets, including ImageNet-1K, COCO, and ADE20K, demonstrate that SPM significantly improves the performance of various models, particularly in dense prediction tasks such as object detection and semantic segmentation. Meanwhile, experiments show that combining SPM with different backbones can further improve performance. The code has been released at https://github.com/Yonghao-Yu/StepwisePatchMerging.

Guobin Shen, Dongcheng Zhao, Yi Zeng

Spiking Neural Networks (SNNs) use discrete spike sequences to transmit information, which significantly mimics the information transmission of the brain. Although this binarized form of representation dramatically enhances the energy efficiency and robustness of SNNs, it also leaves a large gap between the performance of SNNs and Artificial Neural Networks based on real values. There are many different spike patterns in the brain, and the dynamic synergy of these spike patterns greatly enriches the representation capability. Inspired by spike patterns in biological neurons, this paper introduces the dynamic Burst pattern and designs the Leaky Integrate and Fire or Burst (LIFB) neuron that can make a trade-off between short-time performance and dynamic temporal performance from the perspective of network information capacity. LIFB neuron exhibits three modes, resting, Regular spike, and Burst spike. The burst density of the neuron can be adaptively adjusted, which significantly enriches the characterization capability. We also propose a decoupling method that can losslessly decouple LIFB neurons into equivalent LIF neurons, which demonstrates that LIFB neurons can be efficiently implemented on neuromorphic hardware. We conducted experiments on the static datasets CIFAR10, CIFAR100, and ImageNet, which showed that we greatly improved the performance of the SNNs while significantly reducing the network latency. We also conducted experiments on neuromorphic datasets DVS-CIFAR10 and NCALTECH101 and showed that we achieved state-of-the-art with a small network structure.

Yiting Dong, Yang Li, Dongcheng Zhao et al.

The prevalence of violence in daily life poses significant threats to individuals' physical and mental well-being. Using surveillance cameras in public spaces has proven effective in proactively deterring and preventing such incidents. However, concerns regarding privacy invasion have emerged due to their widespread deployment. To address the problem, we leverage Dynamic Vision Sensors (DVS) cameras to detect violent incidents and preserve privacy since it captures pixel brightness variations instead of static imagery. We introduce the Bullying10K dataset, encompassing various actions, complex movements, and occlusions from real-life scenarios. It provides three benchmarks for evaluating different tasks: action recognition, temporal action localization, and pose estimation. With 10,000 event segments, totaling 12 billion events and 255 GB of data, Bullying10K contributes significantly by balancing violence detection and personal privacy persevering. And it also poses a challenge to the neuromorphic dataset. It will serve as a valuable resource for training and developing privacy-protecting video systems. The Bullying10K opens new possibilities for innovative approaches in these domains.

Bing Han, Feifei Zhao, Yi Zeng et al.

Children possess the ability to learn multiple cognitive tasks sequentially, which is a major challenge toward the long-term goal of artificial general intelligence. Existing continual learning frameworks are usually applicable to Deep Neural Networks (DNNs) and lack the exploration on more brain-inspired, energy-efficient Spiking Neural Networks (SNNs). Drawing on continual learning mechanisms during child growth and development, we propose Dynamic Structure Development of Spiking Neural Networks (DSD-SNN) for efficient and adaptive continual learning. When learning a sequence of tasks, the DSD-SNN dynamically assigns and grows new neurons to new tasks and prunes redundant neurons, thereby increasing memory capacity and reducing computational overhead. In addition, the overlapping shared structure helps to quickly leverage all acquired knowledge to new tasks, empowering a single network capable of supporting multiple incremental tasks (without the separate sub-network mask for each task). We validate the effectiveness of the proposed model on multiple class incremental learning and task incremental learning benchmarks. Extensive experiments demonstrated that our model could significantly improve performance, learning speed and memory capacity, and reduce computational overhead. Besides, our DSD-SNN model achieves comparable performance with the DNNs-based methods, and significantly outperforms the state-of-the-art (SOTA) performance for existing SNNs-based continual learning methods.

Bing Han, Feifei Zhao, Yi Zeng et al.

Developmental plasticity plays a prominent role in shaping the brain's structure during ongoing learning in response to dynamically changing environments. However, the existing network compression methods for deep artificial neural networks (ANNs) and spiking neural networks (SNNs) draw little inspiration from brain's developmental plasticity mechanisms, thus limiting their ability to learn efficiently, rapidly, and accurately. This paper proposed a developmental plasticity-inspired adaptive pruning (DPAP) method, with inspiration from the adaptive developmental pruning of dendritic spines, synapses, and neurons according to the ``use it or lose it, gradually decay" principle. The proposed DPAP model considers multiple biologically realistic mechanisms (such as dendritic spine dynamic plasticity, activity-dependent neural spiking trace, and local synaptic plasticity), with additional adaptive pruning strategy, so that the network structure can be dynamically optimized during learning without any pre-training and retraining. Extensive comparative experiments show consistent and remarkable performance and speed boost with the extremely compressed networks on a diverse set of benchmark tasks for deep ANNs and SNNs, especially the spatio-temporal joint pruning of SNNs in neuromorphic datasets. This work explores how developmental plasticity enables complex deep networks to gradually evolve into brain-like efficient and compact structures, eventually achieving state-of-the-art (SOTA) performance for biologically realistic SNNs.

Guobin Shen, Dongcheng Zhao, Aorigele Bao et al.

Human beings often experience stress, which can significantly influence their performance. This study explores whether Large Language Models (LLMs) exhibit stress responses similar to those of humans and whether their performance fluctuates under different stress-inducing prompts. To investigate this, we developed a novel set of prompts, termed StressPrompt, designed to induce varying levels of stress. These prompts were derived from established psychological frameworks and carefully calibrated based on ratings from human participants. We then applied these prompts to several LLMs to assess their responses across a range of tasks, including instruction-following, complex reasoning, and emotional intelligence. The findings suggest that LLMs, like humans, perform optimally under moderate stress, consistent with the Yerkes-Dodson law. Notably, their performance declines under both low and high-stress conditions. Our analysis further revealed that these StressPrompts significantly alter the internal states of LLMs, leading to changes in their neural representations that mirror human responses to stress. This research provides critical insights into the operational robustness and flexibility of LLMs, demonstrating the importance of designing AI systems capable of maintaining high performance in real-world scenarios where stress is prevalent, such as in customer service, healthcare, and emergency response contexts. Moreover, this study contributes to the broader AI research community by offering a new perspective on how LLMs handle different scenarios and their similarities to human cognition.

Jihang Wang, Dongcheng Zhao, Guobin Shen et al.

Privacy protection is a crucial issue in machine learning algorithms, and the current privacy protection is combined with traditional artificial neural networks based on real values. Spiking neural network (SNN), the new generation of artificial neural networks, plays a crucial role in many fields. Therefore, research on the privacy protection of SNN is urgently needed. This paper combines the differential privacy(DP) algorithm with SNN and proposes a differentially private spiking neural network (DPSNN). The SNN uses discrete spike sequences to transmit information, combined with the gradient noise introduced by DP so that SNN maintains strong privacy protection. At the same time, to make SNN maintain high performance while obtaining high privacy protection, we propose the temporal enhanced pooling (TEP) method. It fully integrates the temporal information of SNN into the spatial information transfer, which enables SNN to perform better information transfer. We conduct experiments on static and neuromorphic datasets, and the experimental results show that our algorithm still maintains high performance while providing strong privacy protection.

Wenxuan Pan, Feifei Zhao, Guobin Shen et al.

Spiking Neural Networks (SNNs) have received considerable attention not only for their superiority in energy efficiency with discrete signal processing but also for their natural suitability to integrate multi-scale biological plasticity. However, most SNNs directly adopt the structure of the well-established Deep Neural Networks (DNNs), and rarely automatically design Neural Architecture Search (NAS) for SNNs. The neural motifs topology, modular regional structure and global cross-brain region connection of the human brain are the product of natural evolution and can serve as a perfect reference for designing brain-inspired SNN architecture. In this paper, we propose a Multi-Scale Evolutionary Neural Architecture Search (MSE-NAS) for SNN, simultaneously considering micro-, meso- and macro-scale brain topologies as the evolutionary search space. MSE-NAS evolves individual neuron operation, self-organized integration of multiple circuit motifs, and global connectivity across motifs through a brain-inspired indirect evaluation function, Representational Dissimilarity Matrices (RDMs). This training-free fitness function could greatly reduce computational consumption and NAS's time, and its task-independent property enables the searched SNNs to exhibit excellent transferability on multiple datasets. Furthermore, MSE-NAS show robustness against the training method and noise. Extensive experiments demonstrate that the proposed algorithm achieves state-of-the-art (SOTA) performance with shorter simulation steps on static datasets (CIFAR10, CIFAR100) and neuromorphic datasets (CIFAR10-DVS and DVS128-Gesture). The thorough analysis also illustrates the significant performance improvement and consistent bio-interpretability deriving from the topological evolution at different scales and the RDMs fitness function.

Dachuan Lin, Guobin Shen, Zihao Yang et al.

Safety evaluation of large language models (LLMs) increasingly relies on LLM-as-a-Judge frameworks, but the high cost of frontier models limits scalability. We propose a cost-efficient multi-agent judging framework that employs Small Language Models (SLMs) through structured debates among critic, defender, and judge agents. To rigorously assess safety judgments, we construct HAJailBench, a large-scale human-annotated jailbreak benchmark comprising 12,000 adversarial interactions across diverse attack methods and target models. The dataset provides fine-grained, expert-labeled ground truth for evaluating both safety robustness and judge reliability. Our SLM-based framework achieves agreement comparable to GPT-4o judges on HAJailBench while substantially reducing inference cost. Ablation results show that three rounds of debate yield the optimal balance between accuracy and efficiency. These findings demonstrate that structured, value-aligned debate enables SLMs to capture semantic nuances of jailbreak attacks and that HAJailBench offers a reliable foundation for scalable LLM safety evaluation.

Guobin Shen, Chenxiao Zhao, Xiang Cheng et al.

Training stability remains a central challenge in reinforcement learning (RL) for large language models (LLMs). Policy staleness, asynchronous training, and mismatches between training and inference engines all cause the behavior policy to diverge from the current policy, risking training collapse. Importance sampling provides a principled correction for this distribution shift but suffers from high variance; existing remedies such as token-level clipping and sequence-level normalization lack a unified theoretical foundation. We propose Variational sEquence-level Soft Policy Optimization (VESPO). By incorporating variance reduction into a variational formulation over proposal distributions, VESPO derives a closed-form reshaping kernel that operates directly on sequence-level importance weights without length normalization. Experiments on mathematical reasoning benchmarks show that VESPO maintains stable training under staleness ratios up to 64x and fully asynchronous execution, and delivers consistent gains across both dense and Mixture-of-Experts models. Code is available at https://github.com/FloyedShen/VESPO

Sicheng Shen, Dongcheng Zhao, Guobin Shen et al.

Spiking Neural Networks (SNNs), as the third generation of neural networks, have gained prominence for their biological plausibility and computational efficiency, especially in processing diverse datasets. The integration of attention mechanisms, inspired by advancements in neural network architectures, has led to the development of Spiking Transformers. These have shown promise in enhancing SNNs' capabilities, particularly in the realms of both static and neuromorphic datasets. Despite their progress, a discernible gap exists in these systems, specifically in the Spiking Self Attention (SSA) mechanism's effectiveness in leveraging the temporal processing potential of SNNs. To address this, we introduce the Temporal Interaction Module (TIM), a novel, convolution-based enhancement designed to augment the temporal data processing abilities within SNN architectures. TIM's integration into existing SNN frameworks is seamless and efficient, requiring minimal additional parameters while significantly boosting their temporal information handling capabilities. Through rigorous experimentation, TIM has demonstrated its effectiveness in exploiting temporal information, leading to state-of-the-art performance across various neuromorphic datasets. The code is available at https://github.com/BrainCog-X/Brain-Cog/tree/main/examples/TIM.

Xiang He, Dongcheng Zhao, Yiting Dong et al.

Humans interpret and perceive the world by integrating sensory information from multiple modalities, such as vision and hearing. Spiking Neural Networks (SNNs), as brain-inspired computational models, exhibit unique advantages in emulating the brain's information processing mechanisms. However, existing SNN models primarily focus on unimodal processing and lack efficient cross-modal information fusion, thereby limiting their effectiveness in real-world multimodal scenarios. To address this challenge, we propose a semantic-alignment cross-modal residual learning (S-CMRL) framework, a Transformer-based multimodal SNN architecture designed for effective audio-visual integration. S-CMRL leverages a spatiotemporal spiking attention mechanism to extract complementary features across modalities, and incorporates a cross-modal residual learning strategy to enhance feature integration. Additionally, a semantic alignment optimization mechanism is introduced to align cross-modal features within a shared semantic space, improving their consistency and complementarity. Extensive experiments on three benchmark datasets CREMA-D, UrbanSound8K-AV, and MNISTDVS-NTIDIGITS demonstrate that S-CMRL significantly outperforms existing multimodal SNN methods, achieving the state-of-the-art performance. The code is publicly available at https://github.com/Brain-Cog-Lab/S-CMRL.

Guobin Shen, Xiang Cheng, Chenxiao Zhao et al.

On-policy self-distillation, where a student is pulled toward a copy of itself conditioned on privileged context (e.g., a verified solution or feedback), offers a promising direction for advancing reasoning capability without a stronger external teacher. Yet in math reasoning the gains are inconsistent, even when the same approach succeeds elsewhere. A pointwise mutual information analysis traces the failure to the privileged context itself: it inflates the teacher's confidence on tokens already implied by the solution (structural connectives, verifiable claims) and deflates it on deliberation tokens ("Wait", "Let", "Maybe") that drive multi-step search. We propose Anti-Self-Distillation (AntiSD), which ascends a divergence between student and teacher rather than descending it: this reverses the per-token sign and yields a naturally bounded advantage in one step. An entropy-triggered gate disables the term once the teacher entropy collapses, completing a drop-in replacement for default self-distillation. Across five models from 4B to 30B parameters on math reasoning benchmarks, AntiSD reaches the GRPO baseline's accuracy in 2 to 10x fewer training steps and improves final accuracy by up to 11.5 points. AntiSD opens a path to scalable self-improvement, where a language model bootstraps its own reasoning through its training signal.

Guobin Shen, Lei Huang, Xiang Cheng et al.

On-policy self-distillation has emerged as a promising paradigm for post-training language models, in which the model conditions on environment feedback to serve as its own teacher, providing dense token-level rewards without external teacher models or step-level annotations. Despite its empirical success, what this reward actually measures and what kind of credit it assigns remain unclear. Under a posterior-compatibility interpretation of feedback conditioning, standard in the implicit-reward literature, we show that the self-distillation token reward is a Bayesian filtering increment whose trajectory sum is exactly the pointwise mutual information between the response and the feedback given the input. This pMI can be raised by input-specific reasoning or by input-generic shortcuts, so we further decompose the teacher log-probability along the input axis. Based on this analysis, we propose CREDIT (Contrastive REward from DIsTillation), which isolates the input-specific component with a batch-contrastive baseline. At the sequence level, CREDIT is a teacher-side surrogate for a contrastive pMI objective that also penalizes responses remaining likely under unrelated inputs. Across coding, scientific reasoning, and tool-use benchmarks on two model families, CREDIT delivers the strongest aggregate performance at negligible additional compute.

Ping Wu, Guobin Shen, Dongcheng Zhao et al.

Ensuring that Large Language Models (LLMs) align with mainstream human values and ethical norms is crucial for the safe and sustainable development of AI. Current value evaluation and alignment are constrained by Western cultural bias and incomplete domestic frameworks reliant on non-native rules; furthermore, the lack of scalable, rule-driven scenario generation methods makes evaluations costly and inadequate across diverse cultural contexts. To address these challenges, we propose a hierarchical value framework grounded in core Chinese values, encompassing three main dimensions, 12 core values, and 50 derived values. Based on this framework, we construct a large-scale Chinese Values Corpus (CVC) containing over 250,000 value rules enhanced and expanded through human annotation. Experimental results show that CVC-guided scenarios outperform direct generation ones in value boundaries and content diversity. In the evaluation across six sensitive themes (e.g., surrogacy, suicide), seven mainstream LLMs preferred CVC-generated options in over 70.5% of cases, while five Chinese human annotators showed an 87.5% alignment with CVC, confirming its universality, cultural relevance, and strong alignment with Chinese values. Additionally, we construct 400,000 rule-based moral dilemma scenarios that objectively capture nuanced distinctions in conflicting value prioritization across 17 LLMs. Our work establishes a culturally-adaptive benchmarking framework for comprehensive value evaluation and alignment, representing Chinese characteristics. All data are available at https://huggingface.co/datasets/Beijing-AISI/CVC, and the code is available at https://github.com/Beijing-AISI/CVC.

Lei Huang, Xiang Cheng, Chenxiao Zhao et al.

Large language models (LLMs) typically receive diverse natural language (NL) feedback through interaction with the environment. However, current reinforcement learning (RL) algorithms rely solely on scalar rewards, leaving the rich information in NL feedback underutilized and leading to inefficient exploration. In this work, we propose GOLF, an RL framework that explicitly exploits group-level language feedback to guide targeted exploration through actionable refinements. GOLF aggregates two complementary feedback sources: (i) external critiques that pinpoint errors or propose targeted fixes, and (ii) intra-group attempts that supply alternative partial ideas and diverse failure patterns. These group-level feedbacks are aggregated to produce high-quality refinements, which are adaptively injected into training as off-policy scaffolds to provide targeted guidance in sparse-reward regions. Meanwhile, GOLF jointly optimizes generation and refinement within a unified RL loop, creating a virtuous cycle that continuously improves both capabilities. Experiments on both verifiable and non-verifiable benchmarks show that GOLF achieves superior performance and exploration efficiency, achieving 2.2$\times$ improvements in sample efficiency compared to RL methods trained solely on scalar rewards. Code is available at https://github.com/LuckyyySTA/GOLF.

Huatao Xu, Yan Zhang, Wei Gao et al.

This paper presents the first nationwide deployment of human activity recognition (HAR) technology in the on-demand food delivery industry. We successfully adapted the state-of-the-art LIMU-BERT foundation model to the delivery platform. Spanning three phases over two years, the deployment progresses from a feasibility study in Yangzhou City to nationwide adoption involving 500,000 couriers across 367 cities in China. The adoption enables a series of downstream applications, and large-scale tests demonstrate its significant operational and economic benefits, showcasing the transformative potential of HAR technology in real-world applications. Additionally, we share lessons learned from this deployment and open-source our LIMU-BERT pretrained with millions of hours of sensor data.

Zhuoxiao Li, Wenzong Ma, Taoyu Wu et al.

Recent advances in Neural Radiance Fields and 3D Gaussian Splatting have demonstrated strong potential for large-scale UAV-based 3D reconstruction tasks by fitting the appearance of images. However, real-world large-scale captures are often based on multi-temporal data capture, where illumination inconsistencies across different times of day can significantly lead to color artifacts, geometric inaccuracies, and inconsistent appearance. Due to the lack of UAV datasets that systematically capture the same areas under varying illumination conditions, this challenge remains largely underexplored. To fill this gap, we introduceSkyLume, a large-scale, real-world UAV dataset specifically designed for studying illumination robust 3D reconstruction in urban scene modeling: (1) We collect data from 10 urban regions data comprising more than 100k high resolution UAV images (four oblique views and nadir), where each region is captured at three periods of the day to systematically isolate illumination changes. (2) To support precise evaluation of geometry and appearance, we provide per-scene LiDAR scans and accurate 3D ground-truth for assessing depth, surface normals, and reconstruction quality under varying illumination. (3) For the inverse rendering task, we introduce the Temporal Consistency Coefficient (TCC), a metric that measuress cross-time albedo stability and directly evaluates the robustness of the disentanglement of light and material. We aim for this resource to serve as a foundation that advances research and real-world evaluation in large-scale inverse rendering, geometry reconstruction, and novel view synthesis.

Ze Wang, Guobin Shen, Michael Thaler

Humans increasingly delegate decisions to language models, yet whether these systems reproduce or reshape human patterns of discrimination remains unclear. Here we run a large-scale study to analyse whether language models use demographic information in hiring decisions. We show, across 27 models and 177 occupations, that language models give female and Black candidates hiring advantages relative to otherwise-comparable male and white candidates, while giving disabled candidates disadvantages. The differences are meaningful in magnitude: the role of race, gender, and disability status is comparable to six months to one year of additional education. Post-training alignment is the primary driver: relative to matched pre-trained models, alignment amplifies advantages for female and Black candidates by 325% and 330%, and disadvantages for disabled candidates by 171%. Compared with previous human correspondence studies, language models reverse the direction of racial discrimination, attenuate the disability penalty, and amplify the female advantage by 190%. Alignment changes how models use qualification signals: alignment increases returns to skills and work experience overall, but relatively more so for female and Black candidates. Meanwhile, the absence of qualification signals harms marginalised groups more, particularly for disabled candidates, differences that may explain the asymmetry of alignment effects across groups we observe.

Guobin Shen, Dongcheng Zhao, Yiting Dong et al.

Within the complex neuroarchitecture of the brain, astrocytes play crucial roles in development, structure, and metabolism. These cells regulate neural activity through tripartite synapses, directly impacting cognitive processes such as learning and memory. Despite the growing recognition of astrocytes' significance, traditional Spiking Neural Network (SNN) models remain predominantly neuron-centric, overlooking the profound influence of astrocytes on neural dynamics. Inspired by these biological insights, we have developed an Astrocyte-Modulated Spiking Unit (AM-SU), an innovative framework that integrates neuron-astrocyte interactions into the computational paradigm, demonstrating wide applicability across various hardware platforms. Our Astrocyte-Modulated Spiking Neural Network (AstroSNN) exhibits exceptional performance in tasks involving memory retention and natural language generation, particularly in handling long-term dependencies and complex linguistic structures. The design of AstroSNN not only enhances its biological authenticity but also introduces novel computational dynamics, enabling more effective processing of complex temporal dependencies. Furthermore, AstroSNN shows low latency, high throughput, and reduced memory usage in practical applications, making it highly suitable for resource-constrained environments. By successfully integrating astrocytic dynamics into intelligent neural networks, our work narrows the gap between biological plausibility and neural modeling, laying the groundwork for future biologically-inspired neural computing research that includes both neurons and astrocytes.

Yi Zeng, Feifei Zhao, Yuxuan Zhao et al.

The question "Can machines think?" and the Turing Test to assess whether machines could achieve human-level intelligence is one of the roots of AI. With the philosophical argument "I think, therefore I am", this paper challenge the idea of a "thinking machine" supported by current AIs since there is no sense of self in them. Current artificial intelligence is only seemingly intelligent information processing and does not truly understand or be subjectively aware of oneself and perceive the world with the self as human intelligence does. In this paper, we introduce a Brain-inspired and Self-based Artificial Intelligence (BriSe AI) paradigm. This BriSe AI paradigm is dedicated to coordinating various cognitive functions and learning strategies in a self-organized manner to build human-level AI models and robotic applications. Specifically, BriSe AI emphasizes the crucial role of the Self in shaping the future AI, rooted with a practical hierarchical Self framework, including Perception and Learning, Bodily Self, Autonomous Self, Social Self, and Conceptual Self. The hierarchical framework of the Self highlights self-based environment perception, self-bodily modeling, autonomous interaction with the environment, social interaction and collaboration with others, and even more abstract understanding of the Self. Furthermore, the positive mutual promotion and support among multiple levels of Self, as well as between Self and learning, enhance the BriSe AI's conscious understanding of information and flexible adaptation to complex environments, serving as a driving force propelling BriSe AI towards real Artificial General Intelligence.

Guobin Shen, Dongcheng Zhao, Linghao Feng et al.

Large language models (LLMs) have achieved remarkable capabilities but remain vulnerable to adversarial prompts known as jailbreaks, which can bypass safety alignment and elicit harmful outputs. Despite growing efforts in LLM safety research, existing evaluations are often fragmented, focused on isolated attack or defense techniques, and lack systematic, reproducible analysis. In this work, we introduce PandaGuard, a unified and modular framework that models LLM jailbreak safety as a multi-agent system comprising attackers, defenders, and judges. Our framework implements 19 attack methods and 12 defense mechanisms, along with multiple judgment strategies, all within a flexible plugin architecture supporting diverse LLM interfaces, multiple interaction modes, and configuration-driven experimentation that enhances reproducibility and practical deployment. Built on this framework, we develop PandaBench, a comprehensive benchmark that evaluates the interactions between these attack/defense methods across 49 LLMs and various judgment approaches, requiring over 3 billion tokens to execute. Our extensive evaluation reveals key insights into model vulnerabilities, defense cost-performance trade-offs, and judge consistency. We find that no single defense is optimal across all dimensions and that judge disagreement introduces nontrivial variance in safety assessments. We release the code, configurations, and evaluation results to support transparent and reproducible research in LLM safety.

Guobin Shen, Dongcheng Zhao, Haibo Tong et al.

Ensuring Large Language Model (LLM) safety remains challenging due to the absence of universal standards and reliable content validators, making it difficult to obtain effective training signals. We discover that aligned models already possess robust internal safety beliefs: they consistently produce high-confidence refusals to harmful requests while exhibiting high entropy when generating potentially dangerous content. This entropy gap reveals an untapped signal--models intrinsically "know" when to refuse. We introduce Safety Instincts Reinforcement Learning (SIRL), which transforms this internal confidence into a self-generated reward signal, eliminating dependence on external validators or human annotations. SIRL teaches models to trust their safety instincts by reinforcing low-entropy refusal behaviors. Evaluated on Llama and Qwen models, SIRL maintains 89%+ Defense Success Rates (DSRs) against 20+ jailbreak methods, from static prompts to adaptive attacks. Using only 15,000 unlabeled prompts, SIRL surpasses resource-intensive supervised methods while preserving performance on mathematics, coding, and conversation benchmarks. Our work demonstrates that effective alignment can emerge from within, paving the way for more autonomous and robust AI safety mechanisms that scale without extensive human oversight.

Haibo Tong, Dongcheng Zhao, Guobin Shen et al.

The remarkable capabilities of Large Language Models (LLMs) have raised significant safety concerns, particularly regarding "jailbreak" attacks that exploit adversarial prompts to bypass safety alignment mechanisms. Existing defense research primarily focuses on single-turn attacks, whereas multi-turn jailbreak attacks progressively break through safeguards through by concealing malicious intent and tactical manipulation, ultimately rendering conventional single-turn defenses ineffective. To address this critical challenge, we propose the Bidirectional Intention Inference Defense (BIID). The method integrates forward request-based intention inference with backward response-based intention retrospection, establishing a bidirectional synergy mechanism to detect risks concealed within seemingly benign inputs, thereby constructing a more robust guardrails that effectively prevents harmful content generation. The proposed method undergoes systematic evaluation compared with a no-defense baseline and seven representative defense methods across three LLMs and two safety benchmarks under 10 different attack methods. Experimental results demonstrate that the proposed method significantly reduces the Attack Success Rate (ASR) across both single-turn and multi-turn jailbreak attempts, outperforming all existing baseline methods while effectively maintaining practical utility. Notably, comparative experiments across three multi-turn safety datasets further validate the proposed model's significant advantages over other defense approaches.

Yao Liang, Dongcheng Zhao, Feifei Zhao et al.

The alignment of large language models (LLMs) with human values is critical for their safe and effective deployment across diverse user populations. However, existing benchmarks often neglect cultural and demographic diversity, leading to limited understanding of how value alignment generalizes globally. In this work, we introduce MVPBench, a novel benchmark that systematically evaluates LLMs' alignment with multi-dimensional human value preferences across 75 countries. MVPBench contains 24,020 high-quality instances annotated with fine-grained value labels, personalized questions, and rich demographic metadata, making it the most comprehensive resource of its kind to date. Using MVPBench, we conduct an in-depth analysis of several state-of-the-art LLMs, revealing substantial disparities in alignment performance across geographic and demographic lines. We further demonstrate that lightweight fine-tuning methods, such as Low-Rank Adaptation (LoRA) and Direct Preference Optimization (DPO), can significantly enhance value alignment in both in-domain and out-of-domain settings. Our findings underscore the necessity for population-aware alignment evaluation and provide actionable insights for building culturally adaptive and value-sensitive LLMs. MVPBench serves as a practical foundation for future research on global alignment, personalized value modeling, and equitable AI development.

Bing Han, Feifei Zhao, Dongcheng Zhao et al.

Fine-tuning as service injects domain-specific knowledge into large language models (LLMs), while challenging the original alignment mechanisms and introducing safety risks. A series of defense strategies have been proposed for the alignment, fine-tuning, and post-fine-tuning phases, where most post-fine-tuning defenses rely on coarse-grained safety layer mapping. These methods lack a comprehensive consideration of both safety layers and fine-grained neurons, limiting their ability to efficiently balance safety and utility. To address this, we propose the Fine-Grained Safety Neurons (FGSN) with Training-Free Continual Projection method to reduce the fine-tuning safety risks. FGSN inherently integrates the multi-scale interactions between safety layers and neurons, localizing sparser and more precise fine-grained safety neurons while minimizing interference with downstream task neurons. We then project the safety neuron parameters onto safety directions, improving model safety while aligning more closely with human preferences. Extensive experiments across multiple fine-tuned LLM models demonstrate that our method significantly reduce harmfulness scores and attack success rates with minimal parameter modifications, while preserving the model's utility. Furthermore, by introducing a task-specific, multi-dimensional heterogeneous safety neuron cluster optimization mechanism, we achieve continual defense and generalization capability against unforeseen emerging safety concerns.

Guobin Shen, Dongcheng Zhao, Yi Zeng

The spiking neural network (SNN) mimics the information processing operation in the human brain, represents and transmits information in spike trains containing wealthy spatial and temporal information, and shows superior performance on many cognitive tasks. In addition, the event-driven information processing enables the energy-efficient implementation on neuromorphic chips. The success of deep learning is inseparable from backpropagation. Due to the discrete information transmission, directly applying the backpropagation to the training of the SNN still has a performance gap compared with the traditional deep neural networks. Also, a large simulation time is required to achieve better performance, which results in high latency. To address the problems, we propose a biological plausible spatial adjustment, which rethinks the relationship between membrane potential and spikes and realizes a reasonable adjustment of gradients to different time steps. And it precisely controls the backpropagation of the error along the spatial dimension. Secondly, we propose a biologically plausible temporal adjustment making the error propagate across the spikes in the temporal dimension, which overcomes the problem of the temporal dependency within a single spike period of the traditional spiking neurons. We have verified our algorithm on several datasets, and the experimental results have shown that our algorithm greatly reduces the network latency and energy consumption while also improving network performance. We have achieved state-of-the-art performance on the neuromorphic datasets N-MNIST, DVS-Gesture, and DVS-CIFAR10. For the static datasets MNIST and CIFAR10, we have surpassed most of the traditional SNN backpropagation training algorithm and achieved relatively superior performance.