arXiv subjects

Chenxi Zhou

Publications and source records attributed to Chenxi Zhou.

14 recordsLinked to original sources

arXiv2026

Beyond Factual Knowledge: Benchmarking and Learning Step-Level Procedural Rule Reasoning in Large Language Models

Large language models (LLMs) excel at text understanding and generation, yet still struggle to reliably understand and apply externally provided procedural rules at scale. To evaluate this capability, we introduce RuleWorld, a large-scale benchmark that reformulates rules as globally reusable abstract units rather than instance-specific facts. In RuleWorld, several scenarios, including single-rule, parallel multi-rule, and multi-hop reasoning, are settled for comprehensive evaluation. We further propose DynaRule, an end-to-end framework that injects the given rules into the KV cache and turns retrieval into an internal, learnable, step-wise process. Specifically, DynaRule employs Stacked Step-Level Attention Training with a special token to enable dynamic rule re-attention and updating during inference. In this way, the model can re-attend to the most relevant rules at each step, dynamically replacing outdated ones to support more stable multi-step reasoning. Experiments on RuleWorld show that existing LLMs face challenges under large rule pools, while DynaRule improves average QA accuracy by up to 19 points and achieves over 85% Recall@1 at 10K rules, outperforming strong baselines by large margins. We make our code and dataset available here: https://github.com/SharkSpicy-NLP/Beyond-Factual-Knowledge.

cs.CL↗

arXiv2026

Quantization Degradation in Large Language Models: A Signal-Noise Perspective

Post-training quantization reduces the deployment cost of large language models, yet how severely a quantized model degrades is not determined by bit-width alone. We systematically study weight-only post-training quantization across bit-widths, quantization methods, model scales and downstream tasks on multiple model families. We observe that such degradation varies substantially across these factors: 4-bit quantization usually preserves performance, 2-bit often causes broad degradation, and at 3-bit, degradation becomes apparent but varies markedly with task type, quantization method and model scale. To explain this variability, we use the signal-to-noise ratio (SNR) to measure how strongly quantization perturbs full-precision representations. We trace degradation back to two linked processes: how quantization errors arise within individual modules, and how they accumulate across layers. First, a source SNR decomposition shows that newly introduced errors depend on three factors: the magnitude of the weight error, the strength of the task-specific signal, and how strongly the quantization error aligns with task-specific activations. Different factors affect these components in distinct ways. Second, a cross-layer propagation analysis shows that these errors can be attenuated, preserved, or amplified as they pass across layers, and that larger models benefit from weaker error amplification. Together, these results establish that quantization degradation is governed by how errors are introduced at the source and how they accumulate across the network.

cs.AI↗

arXiv2026

Rethinking Generative AI Literacy: An Integrative, Developmental, and Dialectical Framework for K-12 Teacher Education

Generative artificial intelligence (GenAI) has entered classrooms faster than teachers have been prepared to use it well, producing a GenAI literacy lag in which technological diffusion outpaces educators' conceptual, pedagogical, and ethical readiness. Established AI literacy frameworks predate the widespread adoption of large language models and, while acknowledging ethics, position it as a discrete competency rather than a constitutive commitment, with equity and agency as supplementary design principles. Recent GenAI-specific efforts address isolated features but remain fragmented. We introduce the Responsible AI Literacy in Education (RAIL-Ed) framework, developed through a systematic review and qualitative framework analysis of 67 studies (2023-2025), grounded in critical, pragmatist, sociocultural, and human-centered traditions (Freire, Dewey, Vygotsky, Shneiderman). RAIL-Ed specifies six interdependent pillars: Technical Fluency, Critical Evaluation, Human-AI Collaboration, Contextual Awareness, Ethical Reasoning, and Empowered Agency, marked by three commitments. It is integrative: the absence of any pillar produces a characteristic pedagogical failure. It is developmental: a three-level rubric (Emerging, Competent, Advanced) specifies how each pillar matures across the K-12 teacher-preparation continuum. It is dialectical: the same generative affordance can deepen or displace learning depending on the literacy a teacher brings to it, making the cultivation of that literacy, not the adoption of the tool, the object of design. By treating ethics, equity, and agency as constitutive, RAIL-Ed offers a theoretically grounded basis for curriculum design, teacher education, and policy, aligned with the UNESCO AI Competency Framework for Teachers and the OECD/European Commission AILit Framework. The framework is conceptual, advancing falsifiable propositions for empirical validation.

cs.CY↗

arXiv2026

DecoEvo: Score-Decoupled Co-Evolution of Solver and Rubric-Generator Skills in Text Space

Text-space optimization adapts large language models (LLMs) by editing external natural-language artifacts rather than model weights, so the optimized artifacts remain inspectable and the model can be treated as a black box. However, most existing text-space methods keep evaluation fixed. On open-ended tasks, this can become a bottleneck: once the solver improves on the criteria a rubric measures, omitted dimensions remain invisible to the optimization signal. Simply evolving the rubric is also unreliable when updates are selected by the current solver's score, because apparent progress can come from making the rubric easier to satisfy. We introduce DecoEvo (Decoupled Co-Evolution), which co-evolves a solver skill and a rubric-generator skill under decoupled objectives without using gold rubrics during optimization. The solver skill is updated using criterion-level feedback, while the rubric-generator skill is revised through complementary audits of requirement coverage and response discrimination that are independent of aggregate solver score. This separation focuses generator updates on newly exposed solver weaknesses, reducing repeated emphasis on criteria the solver already satisfies. Under each benchmark's official evaluation, DecoEvo outperforms all compared methods across five benchmarks and three LLM backbones, yielding 2.8--5.0\% relative gains over SkillOpt in the five-benchmark average.

cs.AI↗

arXiv2026

From Proprietary to Open-Source: Bridging the Distribution Gap via Multi-Agent Protocol Distillation in Agentic Search

Agentic search enables large language models to solve knowledge-intensive tasks by interleaving multi-step reasoning with retrieval, yet optimizing this with outcome-based reinforcement learning (RL) provides only sparse supervision. Knowledge distillation can supply denser guidance, and advanced proprietary models with their strong reasoning capabilities are promising teachers. While distilling from proprietary models can densify this supervisory signal, conventional logit-matching is precluded by hidden logits and mismatched tokenizers, whereas raw natural language trajectory imitation transfers superficial stylistic artifacts rather than core reasoning competence. To address the heterogeneous distillation problem and bridge the distribution gap, we propose Multi-Agent Protocol Distillation (MAPD), a joint distillation and RL framework uses a structured, style-normalized protocol as an intermediate representation. An offline multi-agent system (MAS) decomposes each query, retrieves supporting evidence, repairs failed searches, and converts the resulting exploration trace into a JSON protocol containing the task type, reasoning plan, and extractive grounding facts. During training, the protocol is provided only to a privileged branch of the student policy, whose token distributions furnish a dense distillation signal alongside the sparse RL objective. Extensive evaluations across seven QA benchmarks demonstrate that MAPD consistently outperforms competitive distillation and RL, achieving average success rates of 39.4\% on Qwen3-1.7B and 44.4\% on Qwen3-4B. Crucially, the framework generalizes robustly across diverse proprietary teachers while effectively mitigating the student policy from style drift and verbosity degeneration.

cs.AI↗

arXiv2026

PReM: Learning What to Preserve and When to Refresh for Context Compression

Efficient long-context inference is not only about reducing memory cost, but also about keeping useful contextual evidence accessible as generation proceeds. However, existing compression-oriented approaches, such as key-value (KV) cache compression and context compression, often either make an early decision about which contextual information to keep or rely on an external compressor. Such designs make it difficult to adapt the compressed context to the evidence needed by later reasoning steps. This paper introduces PReM (Preserve and Refresh Memory), a context-compression framework that maintains the long context as the model's internal layer-wise KV memory and learns what to preserve and when to refresh it. Specifically, PReM uses a dedicated memory layer to make memory-selection decisions, and a special memory token to trigger refreshes during generation. To train this behavior, PReM introduces Phase-Separated Refresh Training, aligning memory selection with memory-conditioned generation while preserving continuity across refreshes. Experiments with 32K-token contexts show that PReM outperforms strong baselines under both 16x and 32x compression, while maintaining a favorable balance between answer quality and inference efficiency.

cs.CL↗

arXiv2026

LongCrafter: Towards Diverse Long-Context Understanding via Evidence-Graph-Guided Instruction Synthesis

Synthesizing long-context supervised fine-tuning (SFT) data is a scalable way to enhance the long-context understanding of large language models (LLMs), yet existing approaches share three limitations: narrow task coverage, insufficient instruction difficulty, and a lack of faithfulness supervision. We propose \textbf{LongCrafter}, a structured synthesis framework that couples a hierarchical task taxonomy with an evidence-grounded pipeline. The taxonomy organizes long-context understanding into local/shallow and global/deep levels and yields 32 fine-grained task types that serve as a global generative prior. Guided by this taxonomy, LongCrafter constructs task-aligned long contexts, decomposes them into explicit evidence graphs that model cross-paragraph dependencies, and generates instruction--response pairs strictly grounded in the located evidence spans, ensuring both controllable difficulty and faithful, traceable reasoning. Models fine-tuned on LongCrafter data outperform all SFT baselines and even the official post-trained models on LongBench, LongBench~v2, and LooGLE across both Qwen2.5-7B and LLaMA-3.1-8B, with the largest gains on high-difficulty tasks. Further analysis shows that LongCrafter data is more diverse and better spread across difficulty levels, and that the trained models locate evidence robustly regardless of position, effectively mitigating the ``lost in the middle'' problem.

cs.CL↗

arXiv2026

Task-Stratified Knowledge Scaling Laws for Post-Training Quantized Large Language Models

Post-Training Quantization (PTQ) is a critical strategy for efficient Large Language Models (LLMs) deployment. However, existing scaling laws primarily focus on general performance, overlooking crucial fine-grained factors and how quantization differentially impacts diverse knowledge capabilities. To address this, we establish Task-Stratified Knowledge Scaling Laws. By stratifying capabilities into memorization, application, and reasoning, we develop a framework that unifies model size, bit-width, and fine-grained factors: group size and calibration set size. Validated on 293 diverse PTQ configurations, our framework demonstrates strong fit and cross-architecture consistency. It reveals distinct sensitivities across knowledge capabilities: reasoning is precision-critical, application is scale-responsive, and memorization is calibration-sensitive. We highlight that in low-bit scenarios, optimizing these fine-grained factors is essential for preventing performance collapse. These findings provide an empirically-backed foundation for designing knowledge-aware quantization strategies.

cs.CL↗

arXiv2026

From Signal Degradation to Computation Collapse: Uncovering the Two Failure Modes of LLM Quantization

Post-Training Quantization (PTQ) is critical for the efficient deployment of Large Language Models (LLMs). While 4-bit quantization is widely regarded as an optimal trade-off, reducing the precision to 2-bit usually triggers a catastrophic ``performance cliff.'' It remains unclear whether the underlying mechanisms differ fundamentally. Consequently, we conduct a systematic mechanistic analysis, revealing two qualitatively distinct failure modes: Signal Degradation, where the computational patterns remain intact but information precision is impaired by cumulative error; and Computation Collapse, where key components fail to function, preventing correct information processing and destroying the signal in the early layers. Guided by this diagnosis, we conduct mechanism-aware interventions, demonstrating that targeted, training-free repair can mitigate Signal Degradation, but remains ineffective for Computation Collapse. Our findings provide a systematic diagnostic framework for PTQ failures and suggest that addressing Computation Collapse requires structural reconstruction rather than mere compensation.

cs.CL↗

arXiv2026

Pangenome-guided sequence assembly via binary optimisation

De novo genome assembly is challenging in highly repetitive regions; however, reference-guided assemblers often suffer from bias. We propose a framework for pangenome-guided sequence assembly, which can resolve short-read data in complex regions without bias towards a single reference genome. Our primary contribution is to frame the assembly as a graph traversal optimisation problem, which can be implemented classically or on a quantum computer. The workflow involves first annotating pangenome graphs with estimated copy numbers for each node, then finding a path on the graph that best explains those copy numbers. On simulated data, our approach significantly reduces the number of contigs compared to de novo assemblers. While they introduce a small increase in inaccuracies, such as false joins, our optimisation-based methods are competitive with current exhaustive search techniques. They are also designed to scale more efficiently as the problem size grows and will run effectively on future quantum computers; a small experiment on a real quantum device showcases this behaviour. Moreover, they are more resilient to noise in copy number estimation inherent in short-read-based assembly. We also develop novel tools for creating realistic synthetic pangenomes, aligning reads to pangenomes and for evaluating assembly quality.

quant-ph↗

arXiv2024

Field-free current-induced magnetization switching of a room temperature van der Waals magnet for neuromorphic computing

Spin orbit torque (SOT) has become a promising approach to efficiently manipulate the magnetization switching in spintronic devices. As a main factor to impact the device performance, the high quality interface is essentially desired, which can be readily acquired by using the two-dimensional (2D) van der Waals (vdW) materials. Recently, a 2D ferromagnetic material Fe3GaTe2 has been discovered to possess the above-room-temperature Curie temperature and strong perpendicular magnetic anisotropy (PMA), providing an excellent candidate to build spintronic devices. On the other hand, an external magnetic field is necessary for the SOT-driven deterministic switching of perpendicular magnetization, which has become a block for the real applications. Here, we realize the field-free SOT switching of Fe3GaTe2 at room temperature based on the Fe3GaTe2/MnPt heterostructure. In addition, inspired by the superiority of 2D materials in 3D heterogeneous integration, we explore the potential of our device in the computing in memory (CIM). With the application of the current pulses, the gradual switching of our device at zero field imitates the function of artificial synapse in the convolutional neural network (CNN), achieving a high accuracy (~92.8%) pattern recognition. Our work proposes a feasible solution for field-free SOT switching in 2D vdW spintronic devices, which paves the way for applications in magnetic memory and neuromorphic computing.

physics.app-ph↗

arXiv2024

Specular Polynomials

Finding valid light paths that involve specular vertices in Monte Carlo rendering requires solving many non-linear, transcendental equations in high-dimensional space. Existing approaches heavily rely on Newton iterations in path space, which are limited to obtaining at most a single solution each time and easily diverge when initialized with improper seeds. We propose specular polynomials, a Newton iteration-free methodology for finding a complete set of admissible specular paths connecting two arbitrary endpoints in a scene. The core is a reformulation of specular constraints into polynomial systems, which makes it possible to reduce the task to a univariate root-finding problem. We first derive bivariate systems utilizing rational coordinate mapping between the coordinates of consecutive vertices. Subsequently, we adopt the hidden variable resultant method for variable elimination, converting the problem into finding zeros of the determinant of univariate matrix polynomials. This can be effectively solved through Laplacian expansion for one bounce and a bisection solver for more bounces. Our solution is generic, completely deterministic, accurate for the case of one bounce, and GPU-friendly. We develop efficient CPU and GPU implementations and apply them to challenging glints and caustic rendering. Experiments on various scenarios demonstrate the superiority of specular polynomial-based solutions compared to Newton iteration-based counterparts.

cs.GR↗

arXiv2024

Line identification of extreme ultraviolet spectra from aluminum ions in EAST Tokamak plasmas

Extreme ultraviolet (EUV) spectra emitted from aluminum in the 5-340 A wavelength range were observed in Experimental Advanced Superconducting Tokamak (EAST) discharges. Several spectral lines from aluminum ions with different degrees of ionization were successfully observed with sufficient spectral intensities and resolutions using three fast-time-response EUV spectrometers. The line identification uses three independent state-of-art computational codes for the atomic structure calculations, which provide the wavelengths and radiative transition probabilities rate coefficients. These programs are HULLAC (Hebrew University - Lawrence Livermore Atomic Code), AUTOSTRUCTURE, and FAC (Flexible Atomic Code). Using three different codes allows us to resolve some ambiguities in identifying certain spectral lines and assess the validity of the theoretical predictions.

physics.plasm-ph↗

arXiv2021

Modelling brain based on canonical ensemble with functional MRI: A thermodynamic exploration on neural system

Objective. Modelling is an important way to study the working mechanism of brain. While the characterization and understanding of brain are still inadequate. This study tried to build a model of brain from the perspective of thermodynamics at system level, which brought a new thinking to brain modelling. Approach. Regarding brain regions as systems, voxels as particles, and intensity of signals as energy of particles, the thermodynamic model of brain was built based on canonical ensemble theory. Two pairs of activated regions and two pairs of inactivated brain regions were selected for comparison in this study, and the analysis on thermodynamic properties based on the model proposed were performed. In addition, the thermodynamic properties were also extracted as input features for the detection of Alzheimer's disease. Main results. The experiment results verified the assumption that the brain also follows the thermodynamic laws. It demonstrated the feasibility and rationality of brain thermodynamic modelling method proposed, indicating that thermodynamic parameters could be applied to describe the state of neural system. Meanwhile, the brain thermodynamic model achieved much better accuracy in detection of Alzheimer's disease, suggesting the potential application of thermodynamic model in auxiliary diagnosis. Significance. (1) Instead of applying some thermodynamic parameters to analyze neural system, a brain model at system level was proposed from perspective of thermodynamics for the first time in this study. (2) The study discovered that the neural system also follows the laws of thermodynamics, which leads to increased internal energy, increased free energy and decreased entropy when system is activated. (3) The detection of neural disease was demonstrated to be benefit from thermodynamic model, implying the immense potential of thermodynamics in auxiliary diagnosis.

q-bio.NC↗