arXiv subjects

Jibin Wu

Publications and source records attributed to Jibin Wu.

At least 19 recordsLinked to original sources

arXiv2026

CausalBN-Bench: A Comprehensive Benchmark for Causal Learning Capability of LLMs

The ability to understand causality significantly impacts the competence of large language models (LLMs) in output explanation and counterfactual reasoning, as causality reveals the underlying data distribution. However, the lack of a comprehensive benchmark currently limits the evaluation of LLMs' causal learning capabilities. To fill this gap, this paper develops CausalBN-Bench based on data from the causal research community, enabling comparative evaluations of LLMs against traditional causal learning algorithms. To provide a comprehensive investigation, we offer three tasks of varying difficulties, including correlation, causal skeleton, and causality identification. Evaluations of 19 leading LLMs reveal that, while closed-source LLMs show potential for simple causal relationships, they significantly lag behind traditional algorithms on larger-scale networks ($>50$ nodes). Specifically, LLMs struggle with collider structures but excel at chain structures, especially at long-chain causality analogous to Chain-of-Thought techniques. This supports the current prompt approaches while suggesting directions to enhance LLMs' causal reasoning capability. Furthermore, CausalBN-Bench incorporates background knowledge and training data into prompts to thoroughly unlock LLMs' text-comprehension ability during evaluation, whose findings indicate that, LLM understand causality through semantic associations with distinct entities, rather than directly from contextual information or numerical distributions.

cs.LG↗

arXiv2026

PMSN: A Parallel Multi-compartment Spiking Neuron for Multi-scale Temporal Processing

Spiking Neural Networks (SNNs) hold great potential to realize brain-inspired, energy-efficient computational systems. However, current SNNs still fall short in terms of multiscale temporal processing compared to their biological counterparts. This limitation has resulted in poor performance in many pattern recognition tasks with information that varies across different timescales. To address this issue, we put forward a novel spiking neuron model called Parallel Multi-compartment Spiking Neuron (PMSN). The PMSN emulates biological neurons by incorporating multiple interacting substructures and allows for flexible adjustment of the substructure counts to effectively represent temporal information across diverse timescales. Additionally, to address the computational burden associated with the increased complexity of the proposed model, we introduce two parallelization techniques that decouple the temporal dependencies of neuronal updates, enabling parallelized training across different time steps. Our experiments across a wide range of pattern recognition tasks demonstrate that PMSN outperforms state-of-the-art spiking neuron models in temporal processing capacity and training speed. Specifically, compared with the commonly used Leaky Integrate-and-Fire neuron, PMSN offers more than 10 times acceleration and a 30% accuracy improvement on Sequential CIFAR-10 dataset, while maintaining comparable computational cost. Our implementation on neuromorphic hardware further demonstrates the deployability of PMSN and highlights its favorable trade-off between effectiveness and efficiency. Therefore, the proposed PMSN presents a promising solution to harness the computational advantages of detailed biological neurons, enabling high-performance and efficient temporal processing on neuromorphic computing systems. Code is available at https://github.com/xychen-comp/PMSN.

cs.NE↗

arXiv2026

A Systematic Analysis of Hybrid Linear Attention

Transformers face quadratic complexity and memory issues with long sequences, prompting the adoption of linear attention mechanisms using fixed-size hidden states. However, linear models often suffer from limited recall performance, leading to hybrid architectures that combine linear and full attention layers. Despite extensive hybrid architecture research, the choice of linear attention component has not been deeply explored. We systematically evaluate various linear attention models across generations - vector recurrences to advanced gating mechanisms - both standalone and hybridized. To enable this comprehensive analysis, we trained and open-sourced 72 models: 36 at 340M parameters (20B tokens) and 36 at 1.3B parameters (100B tokens), covering six linear attention variants across five hybridization ratios. Benchmarking on standard language modeling and recall tasks reveals that superior standalone linear models do not necessarily excel in hybrids. While language modeling remains stable across linear-to-full attention ratios, recall significantly improves with increased full attention layers, particularly below a 3:1 ratio. Our study highlights selective gating, hierarchical recurrence, and controlled forgetting as critical for effective hybrid models. We recommend architectures such as HGRN-2 or GatedDeltaNet with a linear-to-full ratio between 3:1 and 6:1 to achieve Transformer-level recall efficiently. Our models are open-sourced at https://huggingface.co/collections/m-a-p/hybrid-linear-attention-research-686c488a63d609d2f20e2b1e.

cs.CL↗

arXiv2026

EEG Benchmarking Needs a Task Specification Layer: NeuroDoc for Rulebook-Guided, Executable Benchmark Construction

Electroencephalography (EEG) foundation models increasingly rely on multi-dataset training and evaluation, yet public EEG datasets still lack a shared task specification layer that can turn heterogeneous recordings into reusable benchmark units. Existing standards organize files, metadata, and provenance, but they do not specify EEG tasks under a common language and rulebook, leaving critical task semantics scattered across papers, code, and manual interpretation. We investigate whether heterogeneous public EEG datasets can be standardized through a structured task specification language paired with a shared rulebook. Our methodology represents each benchmark entry as a task document synchronized with an executable task kernel, with the rulebook defining task fields, evidence requirements, document-kernel alignment, review states, and machine-checkable constraints. Using this methodology, we release a community-reviewed EEG benchmark corpus centered on 53 completed and reviewed entries with 245 task definitions spanning diverse paradigms, and we introduce NeuroDoc and NeuroAudit as the operational support layer for rulebook-guided drafting, upgrading, review, amendment, and release management. We further examine whether the resulting benchmark units can be instantiated in a shared downstream setting across four EEG foundation model backbones, providing execution-based evidence for reusable, auditable, and executable EEG benchmarking infrastructure.

cs.LG↗

arXiv2026

Aligning Shared and Routed Experts for Cross-Subject EEG Generalization

Cross-subject EEG generalization is challenging due to substantial heterogeneity across subjects. Existing methods typically learn either a shared subject-invariant model or multiple subject-specialized experts, but these two paradigms fail in complementary ways: the former may over-reduce subject-specific discriminative signals, while the latter may under-reduce transferable structure. We show that their suitability depends on the reducibility cost of branch-specific functions to branch-invariant ones, and we further provide a theory-to-method mapping that instantiates alignment principles in cross-subject EEG learning. Based on this insight, we propose Shared-Routed Expert Alignment (SREA), a collaborative framework that couples a shared expert for reducible invariant functions with routed experts for irreducible subject-specific functions. SREA trains the shared branch with joint embedding over augmented temporal neighbors, the routed branch with prototype-based sparse routing and expert specialization, and both branches with numerically stable mutual-guided reweighting based on cross-branch learnability gaps. Experiments on seven public EEG benchmarks across different tasks show that SREA consistently outperforms state-of-the-art methods and EEG foundation models.

cs.CE↗

arXiv2026

ISTASTrack: Bridging ANN and SNN via ISTA Adapter for RGB-Event Tracking

RGB-Event tracking has become a promising trend in visual object tracking to leverage the complementary strengths of both RGB images and dynamic spike events for improved performance. However, existing artificial neural networks (ANNs) struggle to fully exploit the sparse and asynchronous nature of event streams. Recent efforts toward hybrid architectures combining ANNs and spiking neural networks (SNNs) have emerged as a promising solution in RGB-Event perception, yet effectively fusing features across heterogeneous paradigms remains a challenge. In this work, we propose ISTASTrack, the first transformer-based \textbf{A}NN-\textbf{S}NN hybrid \textbf{Track}er equipped with \textbf{ISTA} adapters for RGB-Event tracking. The two-branch model employs a vision transformer to extract spatial context from RGB inputs and a spiking transformer to capture spatio-temporal dynamics from event streams. To bridge the modality and paradigm gap between ANN and SNN features, we systematically design a model-based ISTA adapter for bidirectional feature interaction between the two branches, derived from sparse representation theory by unfolding the iterative shrinkage thresholding algorithm. Additionally, we incorporate a temporal downsampling attention module within the adapter to align multi-step SNN features with single-step ANN features in the latent space, improving temporal fusion. Experimental results on RGB-Event tracking benchmarks, such as FE240hz, VisEvent, COESOT, and FELT, have demonstrated that ISTASTrack achieves state-of-the-art performance while maintaining high energy efficiency, highlighting the effectiveness and practicality of hybrid ANN-SNN designs for robust visual tracking. The code is publicly available at https://github.com/lsying009/ISTASTrack.git.

cs.CV↗

arXiv2026

$\boldsymbol{f}$-OPD: Stabilizing Long-Horizon On-Policy Distillation with Freshness-Aware Control

Scaling on-policy distillation (OPD) for large language models (LLMs) confronts a fundamental tension: asynchronous execution is necessary for system efficiency, but structurally deviates from the ideal on-policy objective. To address this challenge, we theoretically decompose the objective discrepancy into rollout drift and supervision drift, capturing staleness in student rollout and teacher context, respectively. Building on this, we introduce a sample-level freshness score that quantifies the reliability of a buffered sample with respect to the on-policy objective. Guided by this signal, we further propose f-OPD, a novel framework that adaptively regulates stale-sample influence and constrains policy drift accumulated under asynchronous training. Across reasoning, tool-use, and coding-agent tasks of increasing interaction horizon, f-OPD consistently achieves task performance comparable to synchronous optimization while largely retaining the throughput advantages of asynchronous execution. Our results establish the first recipe for achieving a performance-efficiency trade-off in OPD, paving the way for long-horizon agentic post-training at scale.

cs.LG↗

arXiv2026

SpikingBrain: Spiking Brain-inspired Large Models

Mainstream Transformer-based large language models face major efficiency bottlenecks: training computation scales quadratically with sequence length, and inference memory grows linearly, limiting long-context processing. Building large models on non-NVIDIA platforms also poses challenges for stable and efficient training. To address this, we introduce SpikingBrain, a family of brain-inspired models designed for efficient long-context training and inference. SpikingBrain leverages the MetaX GPU cluster and focuses on three aspects: (1) Model Architecture: linear and hybrid-linear attention architectures with adaptive spiking neurons; (2) Algorithmic Optimizations: an efficient, conversion-based training pipeline and a dedicated spike coding framework; (3) System Engineering: customized training frameworks, operator libraries, and parallelism strategies tailored to MetaX hardware. Using these techniques, we develop two models: SpikingBrain-7B, a linear LLM, and SpikingBrain-76B, a hybrid-linear MoE LLM. These models demonstrate the feasibility of large-scale LLM development on non-NVIDIA platforms, and training remains stable for weeks on hundreds of MetaX GPUs with Model FLOPs Utilization at expected levels. SpikingBrain achieves performance comparable to open-source Transformer baselines while using only about 150B tokens for continual pre-training. Our models also significantly improve long-context efficiency and deliver inference with (partially) constant memory and event-driven spiking behavior. For example, SpikingBrain-7B attains over 100x speedup in Time to First Token for 4M-token sequences. Furthermore, the proposed spiking scheme achieves 69.15 percent sparsity, enabling low-power operation. Overall, this work demonstrates the potential of brain-inspired mechanisms to drive the next generation of efficient and scalable large model design.

cs.LG↗

arXiv2026

ReLaX: Reasoning with Latent Exploration for Large Reasoning Models

Reinforcement Learning with Verifiable Rewards (RLVR) has recently demonstrated remarkable potential in enhancing the reasoning capability of Large Reasoning Models (LRMs). However, RLVR often drives the policy toward over-determinism, resulting in ineffective exploration and premature policy convergence. While promoting token-level diversity has shown promise in mitigating entropy collapse, we argue that the latent dynamics underlying token generation encode a far richer computational structure for steering policy optimization toward a more effective exploration-exploitation tradeoff. To enable tractable analysis and intervention of the latent dynamics of LRMs, we leverage Koopman operator theory to obtain a linearized representation of their hidden state dynamics. This enables us to introduce Dynamic Spectral Dispersion (DSD), a new metric to quantify the heterogeneity of the model's latent dynamics, serving as a direct indicator of policy exploration. Building upon these foundations, we propose Reasoning with Latent eXploration (ReLaX), a framework that explicitly incorporates latent dynamics to regulate exploration and exploitation during policy optimization. Comprehensive experiments across a wide range of multimodal and text-only reasoning benchmarks show that ReLaX consistently incentivizes reasoning capability and outperforms existing token-level methods. Our project is available at https://github.com/ZhangShimin1/ReLaX.

cs.LG↗

arXiv2025

Diversity-Aware Policy Optimization for Large Language Model Reasoning

The reasoning capabilities of large language models (LLMs) have advanced rapidly, particularly following the release of DeepSeek R1, which has inspired a surge of research into data quality and reinforcement learning (RL) algorithms. Despite the pivotal role diversity plays in RL, its influence on LLM reasoning remains largely underexplored. To bridge this gap, this work presents a systematic investigation into the impact of diversity in RL-based training for LLM reasoning, and proposes a novel diversity-aware policy optimization method. Across evaluations on 12 LLMs, we observe a strong positive correlation between the solution diversity and Potential at k (a novel metric quantifying an LLM's reasoning potential) in high-performing models. This finding motivates our method to explicitly promote diversity during RL training. Specifically, we design a token-level diversity and reformulate it into a practical objective, then we selectively apply it to positive samples. Integrated into the R1-zero training framework, our method achieves a 3.5 percent average improvement across four mathematical reasoning benchmarks, while generating more diverse and robust solutions.

cs.LG↗

arXiv2025

HEAR: An EEG Foundation Model with Heterogeneous Electrode Adaptive Representation

Electroencephalography (EEG) is an essential technique for neuroscience research and brain-computer interface (BCI) applications. Recently, large-scale EEG foundation models have been developed, exhibiting robust generalization capabilities across diverse tasks and subjects. However, the heterogeneity of EEG devices not only hinders the widespread adoption of these models but also poses significant challenges to their further scaling and development. In this paper, we introduce HEAR, the first EEG foundation model explicitly designed to support heterogeneous EEG devices, accommodating varying electrode layouts and electrode counts. HEAR employs a learnable, coordinate-based spatial embedding to map electrodes with diverse layouts and varying counts into a unified representational space. This unified spatial representation is then processed by a novel spatially-guided transformer, which effectively captures spatiotemporal dependencies across electrodes. To support the development of HEAR, we construct a large-scale EEG dataset comprising 8,782 hours of data collected from over 150 distinct electrode layouts with up to 1,132 electrodes. Experimental results demonstrate that HEAR substantially outperforms existing EEG foundation models in supporting heterogeneous EEG devices and generalizing across diverse cognitive tasks and subjects.

eess.SP↗

arXiv2025

Efficient Training of Spiking Neural Networks by Spike-aware Data Pruning

Spiking neural networks (SNNs), recognized as an energy-efficient alternative to traditional artificial neural networks (ANNs), have advanced rapidly through the scaling of models and datasets. However, such scaling incurs considerable training overhead, posing challenges for researchers with limited computational resources and hindering the sustained development of SNNs. Data pruning is a promising strategy for accelerating training by retaining the most informative examples and discarding redundant ones, but it remains largely unexplored in SNNs. Directly applying ANN-based data pruning methods to SNNs fails to capture the intrinsic importance of examples and suffers from high gradient variance. To address these challenges, we propose a novel spike-aware data pruning (SADP) method. SADP reduces gradient variance by determining each example's selection probability to be proportional to its gradient norm, while avoiding the high cost of direct gradient computation through an efficient upper bound, termed spike-aware importance score. This score accounts for the influence of all-or-nothing spikes on the gradient norm and can be computed with negligible overhead. Extensive experiments across diverse datasets and architectures demonstrate that SADP consistently outperforms data pruning baselines and achieves training speedups close to the theoretical maxima at different pruning ratios. Notably, SADP reduces training time by 35% on ImageNet while maintaining accuracy comparable to that of full-data training. This work, therefore, establishes a data-centric paradigm for efficient SNN training and paves the way for scaling SNNs to larger models and datasets. The source code will be released publicly after the review process.

cs.NE↗

arXiv2025

IML-Spikeformer: Input-aware Multi-Level Spiking Transformer for Speech Processing

Spiking Neural Networks (SNNs), inspired by biological neural mechanisms, represent a promising neuromorphic computing paradigm that offers energy-efficient alternatives to traditional Artificial Neural Networks (ANNs). Despite proven effectiveness, SNN architectures have struggled to achieve competitive performance on large-scale speech processing tasks. Two key challenges hinder progress: (1) the high computational overhead during training caused by multi-timestep spike firing, and (2) the absence of large-scale SNN architectures tailored to speech processing tasks. To overcome the issues, we introduce Input-aware Multi-Level Spikeformer, i.e. IML-Spikeformer, a spiking Transformer architecture specifically designed for large-scale speech processing. Central to our design is the Input-aware Multi-Level Spike (IMLS) mechanism, which simulates multi-timestep spike firing within a single timestep using an adaptive, input-aware thresholding scheme. IML-Spikeformer further integrates a Re-parameterized Spiking Self-Attention (RepSSA) module with a Hierarchical Decay Mask (HDM), forming the HD-RepSSA module. This module enhances the precision of attention maps and enables modeling of multi-scale temporal dependencies in speech signals. Experiments demonstrate that IML-Spikeformer achieves word error rates of 6.0\% on AiShell-1 and 3.4\% on Librispeech-960, comparable to conventional ANN transformers while reducing theoretical inference energy consumption by 4.64$\times$ and 4.32$\times$ respectively. IML-Spikeformer marks an advance of scalable SNN architectures for large-scale speech processing in both task performance and energy efficiency. Our source code and model checkpoints are publicly available at github.com/Pooookeman/IML-Spikeformer.

cs.MM↗

arXiv2025

ZeCO: Zero Communication Overhead Sequence Parallelism for Linear Attention

Linear attention mechanisms deliver significant advantages for Large Language Models (LLMs) by providing linear computational complexity, enabling efficient processing of ultra-long sequences (e.g., 1M context). However, existing Sequence Parallelism (SP) methods, essential for distributing these workloads across devices, become the primary bottleneck due to substantial communication overhead. In this paper, we introduce ZeCO (Zero Communication Overhead) sequence parallelism for linear attention models, a new SP method designed to overcome these limitations and achieve end-to-end near-linear scalability for long sequence training. For example, training a model with a 1M sequence length across 64 devices using ZeCO takes roughly the same time as training with an 16k sequence on a single device. At the heart of ZeCO lies All-Scan, a new collective communication primitive. All-Scan provides each SP rank with precisely the initial operator state it requires while maintaining a minimal communication footprint, effectively eliminating communication overhead. Theoretically, we prove the optimaity of ZeCO, showing that it introduces only negligible time and space overhead. Empirically, we compare the communication costs of different sequence parallelism strategies and demonstrate that All-Scan achieves the fastest communication in SP scenarios. Specifically, on 256 GPUs with an 8M sequence length, ZeCO achieves a 60\% speedup compared to the current state-of-the-art (SOTA) SP method. We believe ZeCO establishes a clear path toward efficiently training next-generation LLMs on previously intractable sequence lengths.

cs.LG↗

arXiv2025

MSVIT: Improving Spiking Vision Transformer Using Multi-scale Attention Fusion

The combination of Spiking Neural Networks (SNNs) with Vision Transformer architectures has garnered significant attention due to their potential for energy-efficient and high-performance computing paradigms. However, a substantial performance gap still exists between SNN-based and ANN-based transformer architectures. While existing methods propose spiking self-attention mechanisms that are successfully combined with SNNs, the overall architectures proposed by these methods suffer from a bottleneck in effectively extracting features from different image scales. In this paper, we address this issue and propose MSVIT. This novel spike-driven Transformer architecture firstly uses multi-scale spiking attention (MSSA) to enhance the capabilities of spiking attention blocks. We validate our approach across various main datasets. The experimental results show that MSVIT outperforms existing SNN-based models, positioning itself as a state-of-the-art solution among SNN-transformer architectures. The codes are available at https://github.com/Nanhu-AI-Lab/MSViT.

cs.CV↗

arXiv2025

LLM Cannot Discover Causality, and Should Be Restricted to Non-Decisional Support in Causal Discovery

This paper critically re-evaluates LLMs' role in causal discovery and argues against their direct involvement in determining causal relationships. We demonstrate that LLMs' autoregressive, correlation-driven modeling inherently lacks the theoretical grounding for causal reasoning and introduces unreliability when used as priors in causal discovery algorithms. Through empirical studies, we expose the limitations of existing LLM-based methods and reveal that deliberate prompt engineering (e.g., injecting ground-truth knowledge) could overstate their performance, helping to explain the consistently favorable results reported in much of the current literature. Based on these findings, we strictly confined LLMs' role to a non-decisional auxiliary capacity: LLMs should not participate in determining the existence or directionality of causal relationships, but can assist the search process for causal graphs (e.g., LLM-based heuristic search). Experiments across various settings confirm that, by strictly isolating LLMs from causal decision-making, LLM-guided heuristic search can accelerate the convergence and outperform both traditional and LLM-based methods in causal structure learning. We conclude with a call for the community to shift focus from naively applying LLMs to developing specialized models and training method that respect the core principles of causal discovery.

cs.LG↗

arXiv2025

Spatio-Temporal Decoupled Learning for Spiking Neural Networks

Spiking neural networks (SNNs) have gained significant attention for their potential to enable energy-efficient artificial intelligence. However, effective and efficient training of SNNs remains an unresolved challenge. While backpropagation through time (BPTT) achieves high accuracy, it incurs substantial memory overhead. In contrast, biologically plausible local learning methods are more memory-efficient but struggle to match the accuracy of BPTT. To bridge this gap, we propose spatio-temporal decouple learning (STDL), a novel training framework that decouples the spatial and temporal dependencies to achieve both high accuracy and training efficiency for SNNs. Specifically, to achieve spatial decoupling, STDL partitions the network into smaller subnetworks, each of which is trained independently using an auxiliary network. To address the decreased synergy among subnetworks resulting from spatial decoupling, STDL constructs each subnetwork's auxiliary network by selecting the largest subset of layers from its subsequent network layers under a memory constraint. Furthermore, STDL decouples dependencies across time steps to enable efficient online learning. Extensive evaluations on seven static and event-based vision datasets demonstrate that STDL consistently outperforms local learning methods and achieves comparable accuracy to the BPTT method with considerably reduced GPU memory cost. Notably, STDL achieves 4x reduced GPU memory than BPTT on the ImageNet dataset. Therefore, this work opens up a promising avenue for memory-efficient SNN training. Code is available at https://github.com/ChenxiangMA/STDL.

cs.NE↗

arXiv2025

Neuromorphic Sequential Arena: A Benchmark for Neuromorphic Temporal Processing

Temporal processing is vital for extracting meaningful information from time-varying signals. Recent advancements in Spiking Neural Networks (SNNs) have shown immense promise in efficiently processing these signals. However, progress in this field has been impeded by the lack of effective and standardized benchmarks, which complicates the consistent measurement of technological advancements and limits the practical applicability of SNNs. To bridge this gap, we introduce the Neuromorphic Sequential Arena (NSA), a comprehensive benchmark that offers an effective, versatile, and application-oriented evaluation framework for neuromorphic temporal processing. The NSA includes seven real-world temporal processing tasks from a diverse range of application scenarios, each capturing rich temporal dynamics across multiple timescales. Utilizing NSA, we conduct extensive comparisons of recently introduced spiking neuron models and neural architectures, presenting comprehensive baselines in terms of task performance, training speed, memory usage, and energy efficiency. Our findings emphasize an urgent need for efficient SNN designs that can consistently deliver high performance across tasks with varying temporal complexities while maintaining low computational costs. NSA enables systematic tracking of advancements in neuromorphic algorithm research and paves the way for developing effective and efficient neuromorphic temporal processing systems.

cs.NE↗