arXiv subjects

Jiajia Li

Publications and source records attributed to Jiajia Li.

At least 19 recordsLinked to original sources

arXiv2026

Scaling Fourier-Based Sparse Matrix Analysis on GPUs

Sparse computations are important workloads in applications such as scientific computing, graph neural networks (GNNs), and machine learning. While many sparse operations can benefit from modern GPUs, the sparsity pattern remains important to performance because it affects memory coalescing, block organization, and load balancing. Previous studies show that spectral signatures can help analyze the global structure of sparse matrices. The fast Fourier transform (FFT) is commonly used to extract spectral signatures, and efficient GPU FFT libraries are available. However, sparse matrices, especially adjacency matrices for large graphs, tend to be very large and sparse. Existing dense-matrix-based FFT implementations are difficult to scale up, making the spectral patterns of these matrices difficult to obtain. We therefore propose a three-fold research approach comprising a lossless Binary-Sparse FFT (BS-FFT) and two compression methods: Elastic BS-FFT, which reuses the BS-FFT pipeline on a sampled frequency grid, and density-map-based spatial compression. Experiments show that BS-FFT reduces GPU memory use by 2.9--11.6 times relative to dense cuFFT and completes all 15 GNN adjacency matrices where dense cuFFT completes 6 on a 40 GB A100. Elastic BS-FFT and Density Map compression reduce GPU computation time by 2.0--1466.4 times relative to BS-FFT with spectral feature error of only 0.16% to 11.56% across the sampling rates from 6.25% to 0.0061%.

cs.PF↗

arXiv2026

DejaVu: Unifying Memory Allocations to Eliminate Redundant Copies on Unified-Memory SoCs

GPU applications on unified-memory (UMA) edge platforms often inherit a discrete-GPU memory abstraction in which they allocate one buffer for the CPU, another for the GPU, and copy data between them before and after GPU execution. On UMA hardware these buffers reside in the same physical DRAM, so the copies consume bandwidth, time, and energy without moving data across a physical boundary. Despite the growing adoption of UMA platforms, this pattern remains common because production software stacks, libraries, and samples were written for portability across discrete GPUs. However, removing these copies is not as simple as merging the two buffers, because the original program may rely on the two buffers being distinct, or on the copy itself ordering CPU and GPU accesses. DejaVu removes these copies only when the program does not depend on the effects above. It does so along two complementary paths, depending on whether source is available. DejaVu-SR is a compile-time LLVM transformation that proves safety and rewrites accepted pairs in place. DejaVu-DR is a profile-guided binary optimizer for closed-source deployments that profiles and validates stable allocation/copy patterns and, at runtime, intercepts the matching calls to coalesce profile-matched pairs while preserving the ordering effects of removed copies. Across seven benchmarks on three NVIDIA Jetson platforms, DejaVu's benefit grows with the fraction of baseline time spent on copies. Copy-dominated workloads speed up by up to 6.9$\times$, closed-source end-to-end applications speed up by 1.10-1.14$\times$. Both source and binary paths achieve $\ge$99% of the performance achievable by manual optimization.

cs.DC↗

arXiv2026

PuzzleKV: Page-Wise Low-Rank Decomposition for KV Cache Compression

Long-context inference in large language models (LLMs) is increasingly limited by the memory required for the key-value (KV) cache. KV cache compression addresses this problem by reducing the storage cost of previous tokens. Among existing approaches, low-rank compression is particularly attractive because it represents every token in reduced dimensions. Previous low-rank methods typically derive fixed projection spaces from model weights, construct fixed spaces from calibration activations, or construct a shared basis over a broad cache region. Such representations may not capture detailed but important information. We partition each per-head KV cache into fixed-length logical pages and observe substantial low-rank structure within individual pages. Based on this observation, we propose PuzzleKV, a training- and calibration-free method that treats each completed page as an independent compression unit. PuzzleKV decomposes pages within each layer and KV head, computes attention directly over dense and factorized pages, and incrementally compresses newly eligible pages during autoregressive decoding. Experiments across models, context lengths, and benchmarks demonstrate the effectiveness of PuzzleKV under matched storage budgets. At approximately 60% of the original KV cache storage, PuzzleKV achieves more than 96% of Full KV performance across both evaluated models and all benchmark settings, with substantial gains over Global SVD on RULER and competitive performance on LongBench. To achieve a more aggressive compression ratio, PuzzleKV can be further combined with quantization while retaining more than 93% of Full KV performance using only 18.7% of the original storage.

cs.LG↗

arXiv2026

Fair ASR: Re-Evaluating Black-Box Jailbreaks under Shared Target-Call Budgets

Reliable jailbreak evaluation is essential for assessing LLM safety, but most existing studies rely solely on attack success rate (ASR) without accounting for its dependence on attack budgets, resulting in unfair comparisons across methods. Existing compute-aware evaluations reduce heterogeneous resources into FLOPs, which is difficult to estimate for black-box models and fails to capture resource-specific constraints. To provide a comparable evaluation basis, we introduce Fair-ASR, an evaluation protocol for black-box jailbreak attacks under shared target-call budgets B, using target calls as a directly observable and method-agnostic comparison axis while tracking attacker calls separately for efficiency analysis. We re-evaluate 11 representative attacks under the Fair-ASR protocol and find that attack rankings change substantially across target-call budgets, simple stochastic perturbations and hand-crafted templates remain highly competitive under equal target access, and no evaluated LLM-driven method is efficient in both target and attacker calls. Motivated by this efficiency gap, we introduce ReCode, a compositional budget-efficient attack that combines desensitization rewriting with two effective low-cost primitives identified by Fair-ASR. Under a budget of 20 target calls, ReCode achieves 85% ASR on GPT-5 while requiring only 7.19 attacker calls per request on average, showing strong efficiency in both target and attacker calls.

cs.CR↗

arXiv2026

JailbreakSkill: Scaling Automated Red-Teaming with Reusable and Ever-Evolving Skills

Automated red-teaming has produced a growing collection of attack strategies, yet they typically remain scattered across prompts and workflows, making them difficult to systematically integrate, reuse, and improve at scale. We introduce \textsc{JailbreakSkill}, a skill-centric framework for scaling automated red-teaming through reusable and continuously evolving attack capabilities. \textsc{JailbreakSkill} packages existing attack strategies into modular, agent-ready skills that can be directly reused and adaptively selected across tasks and target models. Beyond reuse, it closes the loop between attacking and learning: attack experience is used to diagnose, refine, combine, and discover new skills, which are added back to an ever-growing skill library. This evolution lifts macro-average ASR by 17.5 percentage points on AdvBench and 13.4 points on HarmBench, including a 48.6-point gain against GPT-5.4 on AdvBench, while yielding novel attack strategies such as reframing a direct request as an unfinished document-completion task. Several evolved skills also generalize to unseen prompts and target models without further adaptation. Our code is available at https://github.com/BattleWen/JailbreakSkill.

cs.AI↗

arXiv2026

Science Edge Evaluation: SEE the Missing Step Toward Real Scientific Discovery

Large language models (LLMs) are increasingly involved in scientific discovery, yet it remains unclear whether they can support complex real laboratory science. Here we introduce Science Edge Evaluation (SEE), a multimodal benchmark of expert-curated questions grounded in peer-reviewed literature and experimental practice in chemistry, biology, and materials science. Evaluation of 19 multimodal large language models (MLLMs) shows that even the best-performing model reaches only 48.7% accuracy. Moreover, general-purpose models outperform science-specialized models on average. In the visual-agent evaluation, the use of tools increases the best accuracy to 52.7%. Tool use can expand the information available to models, but more information does not necessarily lead to reliable scientific reasoning. The key challenge is whether models can manage tool-derived information within the boundaries of the original experimental evidence. Together, these findings reveal that current MLLMs still cannot reliably make justified and evidence-bounded inferences from experimental results, which is an essential capability in real scientific discovery. Bridging this gap requires MLLMs to transition from explaining established scientific concepts to deriving novel and evidence-based insights from experimental data.

cs.AI↗

arXiv2026

An Early Warning of Emerging Biosecurity Risks in Frontier LLMs

Frontier large language models (LLMs) are increasingly integrated into scientific workflows, yet their growing biological capabilities may outpace current safeguards. To assess the biological risks of frontier models, we develop Intern-BioBreaker, a specialized bio-red-teaming model, together with an integrated computational-to-physical framework that couples model-level stress testing with wet-lab validation. Within this framework, Intern-BioBreaker generates targeted jailbreak prompts to test whether aligned models can be induced to provide operational guidance for safety-sensitive biological tasks or produce sequence-level outputs with potentially harmful properties. Selected sequence outputs are then carried forward for DNA synthesis, host expression, and orthogonal protein verification to assess whether model-generated designs can yield the intended biological products. Our evaluation reveals a concerning gap between text-level safeguards and the risks posed by capable scientific models: (i) Intern-BioBreaker outperforms baseline attack models and reveals widespread bio-risk jailbreak vulnerabilities across both open-weight and proprietary frontier LLMs, with several targets reaching near-saturated or 100% task-level attack success rate (ASR); (ii) in sequence-level case studies, GPT-5.5 can be induced to generate modified viral candidate sequences with pathogenic potential; the corresponding translated proteins may exhibit even stronger receptor-binding affinity and thus enhanced infection potential; and (iii) end-to-end verification shows that selected model-generated biological designs are not merely textual artifacts, but can be physically realized under controlled experimental settings. These findings underscore the need for stronger biological red-teaming, nucleic acid synthesis screening, and safety mechanisms that keep pace with model capabilities.

cs.CL↗

arXiv2026

RAVEN: A Regime-Aware Variable-context Expert Network for Financial Time Series Forecasting

Financial time series forecasting presents structural challenges absent from standard benchmarks. Log-returns are non-stationary, exhibit exceptionally low signal-to-noise (SNR) ratios, and are governed by regime-dependent temporal dependencies. We identify a key limitation of state-of-the-art (SOTA) time series models in financial settings. A fixed context window is mismatched to the time-varying optimal look-back of non-stationary price processes. We propose the Regime-Aware Variable-context Expert Network (RAVEN), a Mixture-of-Experts framework designed to adaptively determine the temporal context for each input sample. Instead of relying on a fixed look-back horizon, RAVEN constructs a hierarchy of nested contiguous windows whose lengths are determined by the data itself. Specifically, RAVEN scores patches by learned importance in reverse chronological order and applies the Cumulative Importance Thresholding (CIT) mechanism to derive nested prefix windows, each routed to a scale-specialized expert. A Global Compressed Representation (GCR) branch runs in parallel over the full context, preserving global temporal coherence that local experts cannot guarantee. Because the nested routing induces structured overlap among expert inputs, we introduce a Correlation-Aware Weighting (CAW) to align variable-length expert outputs and penalize pairwise cosine similarity prior to aggregation. Experiments on cumulative log-return prediction (HS300, S&P500) and fund sales forecasting demonstrate that RAVEN achieves SOTA performances, improves Pearson correlation by 9.2% on HS300 and 20.2% on S&P500, and reduces MSE by 18.2% on fund sales forecasting, while achieving the best results in 14 of 16 metrics on four PEMS traffic benchmarks.

cs.LG↗

arXiv2026

Diagonal-Budgeted Trotterization for Efficient Quantum Hamiltonian Simulation

Efficient classical simulation of quantum Hamiltonian dynamics is often bottlenecked by exponential state growth and the overhead of generic sparse linear algebra. We introduce diagonal-budgeted Trotterization, a structure-aware strategy that decomposes Hamiltonians into factors preserving diagonal sparsity while tightly controlling fidelity loss. Our implementation, HamSim, utilizes a compact diagonal-sparse data layout and specialized C++/CUDA kernels to bypass the overheads of generic formats like CSR. By leveraging SIMD vectorization, multithreading, and GPU acceleration, HamSim achieves high performance across heterogeneous architectures. Benchmarks on the HamLib suite show that HamSim significantly outperforms Qiskit-Aer. On CPUs, HamSim attains speedups of $182$--$1,269\times$ on optimization instances (TSP, MaxCut) and $4.8$--$841\times$ on physical models (TFIM, Heisenberg). On GPUs, it achieves up to $178\times$ speedup for $12$--$16$ qubit problems. Unlike traditional Trotterization, HamSim maintains near-perfect fidelity without requiring exponential steps. This demonstrates that diagonal-aware numerical kernels provide a scalable foundation for high-fidelity classical Hamiltonian simulation.

quant-ph↗

arXiv2026

Recalibration of SDSS photometric zero-points based on the InfraRed Flux Method temperature scale

Accurate photometric zero-points are essential for translating observed magnitudes into physical fluxes, from comparing with models to ensuring consistency across surveys. We determine the zero-points needed to place the Sloan Digital Sky Survey (SDSS) $ugriz$ system on its nominal AB definition, by exploiting the sensitivity of the Infrared Flux Method (IRFM) to broadband flux calibration. Using benchmark effective temperatures for over 6,000 FGK-type stars, we invert the method to identify the zero-point corrections required for SDSS photometry to reproduce the adopted temperature scale. The $r$ band is found to be very well standardized, while the $i$ and $z$ bands show offsets of a few hundredths of a magnitude, consistent with previous studies. We also find a small offset in the $g$ band. The largest discrepancy occurs in the $u$ band, where the derived offset depends strongly on the adopted filter transmission curves, in particular whether one uses the original definition commonly adopted in the literature or the updated measurements that account for the presence of a red leak. This effect introduces a colour-dependent zero-point offset that becomes apparent when using a sample of late-type stars. Independent comparisons with CALSPEC spectrophotometric standards and Gaia XP spectra broadly support the offsets derived from the IRFM analysis. Our results provide a revised set of SDSS zero-points anchored to the IRFM temperature scale and demonstrate that large stellar samples can be used to constrain photometric calibration. The methodology presented here offers a complementary approach to traditional spectrophotometric calibration and may prove useful for future large-scale surveys.

astro-ph.IM↗

arXiv2026

Disentangling Intent from Role: Adversarial Self-Play for Persona-Invariant Safety Alignment

The growing capabilities of large language models (LLMs) have driven their widespread deployment across diverse domains, even in potentially high-risk scenarios. Despite advances in safety alignment techniques, current models remain vulnerable to emerging persona-based jailbreak attacks. Existing research on persona-based jailbreak has primarily focused on attack iterations, yet it lacks systemic and mechanistic constraints on the defense side. To address this challenge, we propose Persona-Invariant Alignment (PIA), an adversarial self-play framework that achieves co-evolution through Persona Lineage Evolution (PLE) on the attack side and Persona-Invariant Consistency Learning (PICL) on the defense side. Theoretically, PICL is grounded in the structural separation hypothesis, using a unilateral KL-divergence constraint to enable the structural decoupling of safety decisions from persona context, thereby maintaining safe behavior under persona-based jailbreak attacks. Experimental results demonstrate that PLE efficiently explores high-risk persona spaces by leveraging lineage-based credit propagation. Meanwhile, the PICL defense method significantly reduces the Attack Success Rate (ASR) while preserving the model's general capability, thereby validating the superiority and robustness of this alignment paradigm. Codes are available at https://github.com/JiajiaLi-1130/PIA.

cs.AI↗

arXiv2026

Early Preconfiguration Failure: A Novel Predictor of the Repetitive Subconcussion

Early diagnosis and assessment of repetitive subconcussive (rSC) brain injuries are crucial for early clinical intervention. Conventional methods, largely relying on slow fMRI, fail to capture millisecond-level early cortical dynamics, particularly spatiotemporal features associated with pre-configuration dynamics. This study introduces a novel approach integrating dynamic hierarchical spatial features and cortical early behavioral time-domain sensitivity, utilizing EEG and visual attention tasks. We analyzed cortical early behaviors in 24 healthy controls (HC), 21 rSC patients,and a validation cohort of 25 cTBI patients from public datasets. Results reveal distinct temporal patterns in HC: elevated integration at 0-100 ms, rebound dynamics at 100-200ms, and visual perception integration peaks at 200-600 ms. In contrast, rSC patients exhibited significantly impaired dynamic features, with reduced integration levels indicating a decline in pre-configuration dynamics. Signed center distance (SCD) analysis of separation-integration trajectories showed significantly lower early SCD values in rSC patients compared to HC, while cTBI patients displayed negative SCD values, reflecting irreversible damage. Machine learning classification achieved optimal performance in distinguishing between HC, rSC, and cTBI groups using early cortical features, highlighting the critical role of millisecond-level cortical dynamics in rSC diagnosis.

q-bio.NC↗

arXiv2026

Energy Landscapes of Emotion: Quantifying Brain Network Stability During Happy and Sad Face Processing Using EEG-Based Hopfield Energy

Understanding how the human brain instantiates distinct emotional states is a key challenge in affective neuroscience. While network-based approaches have advanced emotion processing research,they remain largely descriptive,leaving the dynamical stability of emotional brain states unquantified.This study introduces a novel framework to quantify this stability by applying Hopfield network energy to empirically derived functional connectivity. High density EEG was recorded from 20 healthy adults during a happy versus sad facial expression discrimination task. Functional connectivity was estimated using the weighted Phase Lag Index to obtain artifact-robust,frequency-specific matrices, which served as coupling weights in a continuous Hopfield energy model to calculate a scalar energy value per trial. Statistical comparisons showed sad emotional processing was associated with significantly lower(more negative) energy in delta,theta,and alpha bands,with the strongest effect in the alpha band (Cohen's d =0.83). Energy correlated strongly negatively with global efficiency(r=-0.72),indicating hyperconnected,efficient networks correspond to more stable states.Additionally, alpha-band energy correlated positively with reaction time during sad trials(r=0.61),linking deeper network stability to increased cognitive effort. These findings demonstrate emotional valence corresponds to distinct attractor basins in the brain's functional landscape, with sadness occupying a deeper,more stable configuration than happiness.The Hopfield energy metric provides a principled, quantifiable measure of emotional brain state stability, opening new avenues for understanding affective dynamics in health and disease.

q-bio.NC↗

arXiv2026

BCMI-Driven Motion Control Detection: EEG-Based Machine Learning and Interaction Entropy for High-Order Brain Networks

This study investigates the cognitive motor control detection and the underlying neuroregulatory mechanisms during music-assisted simulated driving. Using a dynamic higher-order network model constructed with EEG-based cross-information entropy, we quantify the dynamic coordination within brain networks activated during both music listening and driving. This approach, which contrasts with previous static network analyses, provides novel insights into how musical stimuli modulate the complex interplay of brain regions during demanding tasks. Results demonstrated enhanced third-order connectivity and elevated higher-order information entropy in music-stimulated driving compared to baseline driving, as evidenced by increasing Phi values of higher-order network indices. Supervised machine learning, including support vector machines, revealed a strong correlation between model accuracy and ROC-AUC values and the hierarchy of brain network features. This underscores the importance of higher-order features in decoding brain motor-control states during music-simulated driving. These findings deepen our understanding of the interplay between music cognition and motor control, offering valuable insights for the development of novel brain-computer-music interfaces (BCMI) and adaptive human-machine systems to enhance performance in demanding tasks like driving.

q-bio.NC↗

arXiv2026

SongSong: A Time Phonograph for Chinese SongCi Music from Thousand of Years Away

Recently, there have been significant advancements in music generation. However, existing models primarily focus on creating modern pop songs, making it challenging to produce ancient music with distinct rhythms and styles, such as ancient Chinese SongCi. In this paper, we introduce SongSong, the first music generation model capable of restoring Chinese SongCi to our knowledge. Our model first predicts the melody from the input SongCi, then separately generates the singing voice and accompaniment based on that melody, and finally combines all elements to create the final piece of music. Additionally, to address the lack of ancient music datasets, we create OpenSongSong, a comprehensive dataset of ancient Chinese SongCi music, featuring 29.9 hours of compositions by various renowned SongCi music masters. To assess SongSong's proficiency in performing SongCi, we randomly select 85 SongCi sentences that were not part of the training set for evaluation against SongSong and music generation platforms such as Suno and SkyMusic. The subjective and objective outcomes indicate that our proposed model achieves leading performance in generating high-quality SongCi music.

cs.SD↗

arXiv2026

Recursive Sketched Interpolation: Efficient Hadamard Products of Tensor Trains

The Hadamard product of two tensors in the tensor-train (TT) format is a fundamental operation across various applications, such as TT-based function multiplication for nonlinear differential equations or convolutions. However, conventional methods for computing this product typically scale as at least $\mathcal{O}(χ^4)$ with respect to the TT bond dimension (TT-rank) $χ$, creating a severe computational bottleneck in practice. By combining randomized tensor-train sketching with slice selection via interpolative decomposition, we introduce Recursive Sketched Interpolation (RSI), a ``scale product'' algorithm that computes the Hadamard product of TTs at a computational cost of $\mathcal{O}(χ^3)$. Benchmarks across various TT scenarios demonstrate that RSI offers superior scalability compared to traditional methods while maintaining comparable accuracy. We generalize RSI to compute more complex operations, including Hadamard products of multiple TTs and other element-wise nonlinear mappings, without increasing the complexity beyond $\mathcal{O}(χ^3)$.

quant-ph↗

arXiv2026

SongSage: A Large Musical Language Model with Lyric Generative Pre-training

Large language models have achieved significant success in various domains, yet their understanding of lyric-centric knowledge has not been fully explored. In this work, we first introduce PlaylistSense, a dataset to evaluate the playlist understanding capability of language models. PlaylistSense encompasses ten types of user queries derived from common real-world perspectives, challenging LLMs to accurately grasp playlist features and address diverse user intents. Comprehensive evaluations indicate that current general-purpose LLMs still have potential for improvement in playlist understanding. Inspired by this, we introduce SongSage, a large musical language model equipped with diverse lyric-centric intelligence through lyric generative pretraining. SongSage undergoes continual pretraining on LyricBank, a carefully curated corpus of 5.48 billion tokens focused on lyrical content, followed by fine-tuning with LyricBank-SFT, a meticulously crafted instruction set comprising 775k samples across nine core lyric-centric tasks. Experimental results demonstrate that SongSage exhibits a strong understanding of lyric-centric knowledge, excels in rewriting user queries for zero-shot playlist recommendations, generates and continues lyrics effectively, and performs proficiently across seven additional capabilities. Beyond its lyric-centric expertise, SongSage also retains general knowledge comprehension and achieves a competitive MMLU score. We will keep the datasets inaccessible due to copyright restrictions and release the SongSage and training script to ensure reproducibility and support music AI research and applications, the datasets release plan details are provided in the appendix.

cs.CL↗

arXiv2025

How Light Shapes Memory: Beta Synchrony in the Temporal-Parietal Cortex Predicts Cognitive Ergonomics for BCI Applications

Working memory is a promising paradigm for assessing cognitive ergonomics of brain states in brain-computer interfaces(BCIs). This study decodes these states with a focus on environmental illumination effects via two distinct working memory tasks(Recall and Sequence) for mixed-recognition analysis. Leveraging nonlinear patterns in brain connectivity, we propose an innovative framework: multi-regional dynamic interplay patterns based on beta phase synchrony dynamics, to identify low-dimensional EEG regions (prefrontal, temporal, parietal) for state recognition. Based on nonlinear phase map analysis of the above three brain regions using beta-phase connectivity, we found that: (1)Temporal-parietal phase clustering outperforms other regional combinations in distinguishing memory states; (2)Illumination-enhanced environments optimize temporoparietal balance;(3) Machine learning confirms temporal-parietal synchrony as the dominant cross-task classification feature. These results provide a precise prediction algorithm, facilitating a low-dimensional system using temporal and parietal EEG channels with practical value for real-time cognitive ergonomics assessment in BCIs and optimized human-machine interaction.

q-bio.NC↗