arXiv subjects

Jun Zhao

Publications and source records attributed to Jun Zhao.

At least 145 records · Page 8Linked to original sources

arXiv2025

Probabilistic Uncertain Reward Model

Reinforcement learning from human feedback (RLHF) is a critical technique for training large language models. However, conventional reward models based on the Bradley-Terry model (BTRM) often suffer from overconfidence when faced with inconsistent labels or out-of-distribution samples, leading to reward hacking, where the policy model blindly optimizes for proxy rewards while degrading true performance. This paper proposes the Probabilistic Uncertain Reward Model (PURM), which generalizes the Bradley-Terry model to learn the reward distributions that emerged from the preference data. We theoretically derive the loss function of PURM and introduce a novel method that uses the overlap between distributions to quantify uncertainty. Empirical results show that PURM outperforms existing methods with more accurate reward and sound uncertainty estimations, and sustains effective learning for more optimization steps and obtain higher maximum win rate in RLHF. The data and code of this paper are released at https://anonymous.4open.science/r/Probabilistic-Uncertain-Reward-Model/

cs.LG↗

arXiv2025

Improve Rule Retrieval and Reasoning with Self-Induction and Relevance ReEstimate

This paper systematically addresses the challenges of rule retrieval, a crucial yet underexplored area. Vanilla retrieval methods using sparse or dense retrievers to directly search for relevant rules to support downstream reasoning, often suffer from low accuracy. This is primarily due to a significant semantic gap between the instantiated facts in the queries and the abstract representations of the rules. Such misalignment results in suboptimal retrieval quality, which in turn negatively impacts reasoning performance. To overcome these challenges, we propose Self-Induction Augmented Retrieval (SIAR), a novel approach that utilizes Large Language Models (LLMs) to induce potential inferential rules that might offer benefits for reasoning by abstracting the underlying knowledge and logical structure in queries. These induced rules are then used for query augmentation to improve retrieval effectiveness. Additionally, we introduce Rule Relevance ReEstimate (R$^3$), a method that re-estimates the relevance of retrieved rules by assessing whether the abstract knowledge they contain can be instantiated to align with the facts in the queries and the helpfulness for reasoning. Extensive experiments across various settings demonstrate the effectiveness and versatility of our proposed methods.

cs.CL↗

arXiv2025

Private Transformer Inference in MLaaS: A Survey

Transformer models have revolutionized AI, powering applications like content generation and sentiment analysis. However, their deployment in Machine Learning as a Service (MLaaS) raises significant privacy concerns, primarily due to the centralized processing of sensitive user data. Private Transformer Inference (PTI) offers a solution by utilizing cryptographic techniques such as secure multi-party computation and homomorphic encryption, enabling inference while preserving both user data and model privacy. This paper reviews recent PTI advancements, highlighting state-of-the-art solutions and challenges. We also introduce a structured taxonomy and evaluation framework for PTI, focusing on balancing resource efficiency with privacy and bridging the gap between high-performance inference and data privacy.

cs.CR↗

arXiv2025

Reinforced Internal-External Knowledge Synergistic Reasoning for Efficient Adaptive Search Agent

Retrieval-augmented generation (RAG) is a common strategy to reduce hallucinations in Large Language Models (LLMs). While reinforcement learning (RL) can enable LLMs to act as search agents by activating retrieval capabilities, existing ones often underutilize their internal knowledge. This can lead to redundant retrievals, potential harmful knowledge conflicts, and increased inference latency. To address these limitations, an efficient and adaptive search agent capable of discerning optimal retrieval timing and synergistically integrating parametric (internal) and retrieved (external) knowledge is in urgent need. This paper introduces the Reinforced Internal-External Knowledge Synergistic Reasoning Agent (IKEA), which could indentify its own knowledge boundary and prioritize the utilization of internal knowledge, resorting to external search only when internal knowledge is deemed insufficient. This is achieved using a novel knowledge-boundary aware reward function and a knowledge-boundary aware training dataset. These are designed for internal-external knowledge synergy oriented RL, incentivizing the model to deliver accurate answers, minimize unnecessary retrievals, and encourage appropriate external searches when its own knowledge is lacking. Evaluations across multiple knowledge reasoning tasks demonstrate that IKEA significantly outperforms baseline methods, reduces retrieval frequency significantly, and exhibits robust generalization capabilities.

cs.CL↗

arXiv2025

Pangu Ultra MoE: How to Train Your Big MoE on Ascend NPUs

Sparse large language models (LLMs) with Mixture of Experts (MoE) and close to a trillion parameters are dominating the realm of most capable language models. However, the massive model scale poses significant challenges for the underlying software and hardware systems. In this paper, we aim to uncover a recipe to harness such scale on Ascend NPUs. The key goals are better usage of the computing resources under the dynamic sparse model structures and materializing the expected performance gain on the actual hardware. To select model configurations suitable for Ascend NPUs without repeatedly running the expensive experiments, we leverage simulation to compare the trade-off of various model hyperparameters. This study led to Pangu Ultra MoE, a sparse LLM with 718 billion parameters, and we conducted experiments on the model to verify the simulation results. On the system side, we dig into Expert Parallelism to optimize the communication between NPU devices to reduce the synchronization overhead. We also optimize the memory efficiency within the devices to further reduce the parameter and activation management overhead. In the end, we achieve an MFU of 30.0% when training Pangu Ultra MoE, with performance comparable to that of DeepSeek R1, on 6K Ascend NPUs, and demonstrate that the Ascend system is capable of harnessing all the training stages of the state-of-the-art language models. Extensive experiments indicate that our recipe can lead to efficient training of large-scale sparse language models with MoE. We also study the behaviors of such models for future reference.

cs.CL↗

arXiv2025

Dynamic Parametric Retrieval Augmented Generation for Test-time Knowledge Enhancement

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by retrieving relevant documents from external sources and incorporating them into the context. While it improves reliability by providing factual texts, it significantly increases inference costs as context length grows and introduces challenging issue of RAG hallucination, primarily caused by the lack of corresponding parametric knowledge in LLMs. An efficient solution is to enhance the knowledge of LLMs at test-time. Parametric RAG (PRAG) addresses this by embedding document into LLMs parameters to perform test-time knowledge enhancement, effectively reducing inference costs through offline training. However, its high training and storage costs, along with limited generalization ability, significantly restrict its practical adoption. To address these challenges, we propose Dynamic Parametric RAG (DyPRAG), a novel framework that leverages a lightweight parameter translator model to efficiently convert documents into parametric knowledge. DyPRAG not only reduces inference, training, and storage costs but also dynamically generates parametric knowledge, seamlessly enhancing the knowledge of LLMs and resolving knowledge conflicts in a plug-and-play manner at test-time. Extensive experiments on multiple datasets demonstrate the effectiveness and generalization capabilities of DyPRAG, offering a powerful and practical RAG paradigm which enables superior knowledge fusion and mitigates RAG hallucination in real-world applications. Our code is available at https://github.com/Trae1ounG/DyPRAG.

cs.CL↗

arXiv2025

The LHC sensitivity to weak gauginos in light of the latest muon $g-2$ and dark matter results

Among the electroweakinos, the weak gauginos have the largest production rate at the LHC and should therefore be the primary focus of searches. In this work, we examine the LHC sensitivity to weak gauginos in light of the latest constraints from the muon $g-2$ anomaly and dark matter (DM) observations. To simultaneously account for the observed $5σ$ deviation in muon $g-2$ and the correct DM relic abundance, the DM candidate in the MSSM must be bino-like: wino- or higgsino-like neutralinos require TeV-scale masses to avoid underabundance, rendering the electroweakino spectrum too heavy to yield a sizable $g-2$ contribution. Moreover, tight direct detection limits disfavor light higgsinos, which can induce sizable DM-nucleon scattering via bino-higgsino mixing. We thus focus on scenarios with a bino-like LSP and relatively heavy higgsinos. Two classes of wino spectra are considered: a light-wino scenario (LWo) with $m_{\tilde{W}} \lesssim 1$~TeV, and a heavy-wino scenario (HWo) with $m_{\tilde{W}} \gtrsim 1$~TeV. For each, we delineate the viable parameter space under current constraints and assess the discovery potential at future LHC stages. We find that while the high-luminosity LHC (HL-LHC) can probe a portion of the parameter space, the high-energy 27~TeV LHC (HE-LHC) is capable of covering most of it.

hep-ph↗

arXiv2025

A Cognitive-Mechanistic Human Reliability Analysis Framework: A Nuclear Power Plant Case Study

Traditional human reliability analysis (HRA) methods, such as IDHEAS-ECA, rely on expert judgment and empirical rules that often overlook the cognitive underpinnings of human error. Moreover, conducting human-in-the-loop experiments for advanced nuclear power plants is increasingly impractical due to novel interfaces and limited operational data. This study proposes a cognitive-mechanistic framework (COGMIF) that enhances the IDHEAS-ECA methodology by integrating an ACT-R-based human digital twin (HDT) with TimeGAN-augmented simulation. The ACT-R model simulates operator cognition, including memory retrieval, goal-directed procedural reasoning, and perceptual-motor execution, under high-fidelity scenarios derived from a high-temperature gas-cooled reactor (HTGR) simulator. To overcome the resource constraints of large-scale cognitive modeling, TimeGAN is trained on ACT-R-generated time-series data to produce high-fidelity synthetic operator behavior datasets. These simulations are then used to drive IDHEAS-ECA assessments, enabling scalable, mechanism-informed estimation of human error probabilities (HEPs). Comparative analyses with SPAR-H and sensitivity assessments demonstrate the robustness and practical advantages of the proposed COGMIF. Finally, procedural features are mapped onto a Bayesian network to quantify the influence of contributing factors, revealing key drivers of operational risk. This work offers a credible and computationally efficient pathway to integrate cognitive theory into industrial HRA practices.

cs.AI↗

arXiv2025

Effective Length Extrapolation via Dimension-Wise Positional Embeddings Manipulation

Large Language Models (LLMs) often struggle to process and generate coherent context when the number of input tokens exceeds the pre-trained length. Recent advancements in long-context extension have significantly expanded the context window of LLMs but require expensive overhead to train the large-scale models with longer context. In this work, we propose Dimension-Wise Positional Embeddings Manipulation (DPE), a training-free framework to extrapolate the context window of LLMs by diving into RoPE's different hidden dimensions. Instead of manipulating all dimensions equally, DPE detects the effective length for every dimension and finds the key dimensions for context extension. We reuse the original position indices with their embeddings from the pre-trained model and manipulate the key dimensions' position indices to their most effective lengths. In this way, DPE adjusts the pre-trained models with minimal modifications while ensuring that each dimension reaches its optimal state for extrapolation. DPE significantly surpasses well-known baselines such as YaRN and Self-Extend. DPE enables Llama3-8k 8B to support context windows of 128k tokens without continual training and integrates seamlessly with Flash Attention 2. In addition to its impressive extrapolation capability, DPE also dramatically improves the models' performance within training length, such as Llama3.1 70B, by over 18 points on popular long-context benchmarks RULER. When compared with commercial models, Llama 3.1 70B with DPE even achieves better performance than GPT-4-128K.

cs.CL↗

arXiv2025

Algorithmic Mirror: Designing an Interactive Tool to Promote Self-Reflection for YouTube Recommendations

Big Data analytics and Artificial Intelligence systems derive non-intuitive and often unverifiable inferences about individuals' behaviors, preferences, and private lives. Drawing on diverse, feature-rich datasets of unpredictable value, these systems erode the intuitive connection between our actions and how we are perceived, diminishing control over our digital identities. While Explainable Artificial Intelligence scholars have attempted to explain the inner workings of algorithms, their visualizations frequently overwhelm end-users with complexity. This research introduces 'hypothetical inference', a novel approach that uses language models to simulate how algorithms might interpret users' digital footprints and infer personal characteristics without requiring access to proprietary platform algorithms. Through empirical studies with fourteen adult participants, we identified three key design opportunities to foster critical algorithmic literacy: (1) reassembling scattered digital footprints into a unified map, (2) simulating algorithmic inference through LLM-generated interpretations, and (3) incorporating temporal dimensions to visualize evolving patterns. This research lays the groundwork for tools that can help users recognize the influence of data on platforms and develop greater autonomy in increasingly algorithm-mediated digital environments.

cs.HC↗

arXiv2025

Exploiting Contextual Knowledge in LLMs through V-usable Information based Layer Enhancement

Large Language Models (LLMs) have demonstrated remarkable capabilities in various tasks, yet they often struggle with context-faithfulness generations that properly reflect contextual knowledge. While existing approaches focus on enhancing the decoding strategies, they ignore the fundamental mechanism of how contextual information is processed within LLMs' internal states. As a result, LLMs remain limited in their ability to fully leverage contextual knowledge. In this paper, we propose Context-aware Layer Enhancement (CaLE), a novel intervention method that enhances the utilization of contextual knowledge within LLMs' internal representations. By employing V-usable information analysis, CaLE strategically amplifies the growth of contextual information at an optimal layer, thereby enriching representations in the final layer. Our experiments demonstrate that CaLE effectively improves context-faithful generation in Question-Answering tasks, particularly in scenarios involving unknown or conflicting contextual knowledge.

cs.CL↗

arXiv2025

Transparentize the Internal and External Knowledge Utilization in LLMs with Trustworthy Citation

While hallucinations of large language models could been alleviated through retrieval-augmented generation and citation generation, how the model utilizes internal knowledge is still opaque, and the trustworthiness of its generated answers remains questionable. In this work, we introduce Context-Prior Augmented Citation Generation task, requiring models to generate citations considering both external and internal knowledge while providing trustworthy references, with 5 evaluation metrics focusing on 3 aspects: answer helpfulness, citation faithfulness, and trustworthiness. We introduce RAEL, the paradigm for our task, and also design INTRALIGN, an integrated method containing customary data generation and an alignment algorithm. Our experimental results show that our method achieves a better cross-scenario performance with regard to other baselines. Our extended experiments further reveal that retrieval quality, question types, and model knowledge have considerable influence on the trustworthiness in citation generation.

cs.CL↗

arXiv2025

Verifying Properties of Binary Neural Networks Using Sparse Polynomial Optimization

This paper explores methods for verifying the properties of Binary Neural Networks (BNNs), focusing on robustness against adversarial attacks. Despite their lower computational and memory needs, BNNs, like their full-precision counterparts, are also sensitive to input perturbations. Established methods for solving this problem are predominantly based on Satisfiability Modulo Theories and Mixed-Integer Linear Programming techniques, which are characterized by NP complexity and often face scalability issues. We introduce an alternative approach using Semidefinite Programming relaxations derived from sparse Polynomial Optimization. Our approach, compatible with continuous input space, not only mitigates numerical issues associated with floating-point calculations but also enhances verification scalability through the strategic use of tighter first-order semidefinite relaxations. We demonstrate the effectiveness of our method in verifying robustness against both $\|.\|_\infty$ and $\|.\|_2$-based adversarial attacks.

cs.LG↗

arXiv2025

Bulk superconductivity in pressurized trilayer nickelate Pr4Ni3O10 single crystals

The discovery of superconductivity in pressurized bilayer and trilayer nickelates has generated significant interest. However, their superconducting properties are often dependent on sample quality and pressure conditions, complicating the interpretation of the underlying physics. Finding new systems with optimized bulk superconducting properties is therefore important for advancing our understanding of these materials. Unlike cupates, where trilayer compounds typically exhibit the highest transition temperature (Tc), the bilayer nickelate La3Ni2O7 has thus far outperformed the trilayer La4Ni3O10 in reported Tc. Whether the trilayer nickelates have achieved the optimal Tc remains unclear, with various scenarios suggesting different possibilities. Here, we report the discovery of bulk superconductivity in pressurized Pr4Ni3O10 single crystals, achieving a maximum onset Tc of 40.5 K at 80.1 GPa, significantly exceeding the 30 K observed in La4Ni3O10. The bulk nature of superconductivity is confirmed by zero resistance and a strong diamagnetic response below Tc with a superconducting volume fraction exceeding 80%. These findings establish trilayer nickelates as genuine bulk high-temperature superconductors, provide new insights into the mechanisms driving superconductivity, and point to a promising route toward further enhancing superconducting properties in nickelates.

cond-mat.supr-con↗

arXiv2025

Evaluating Personalized Tool-Augmented LLMs from the Perspectives of Personalization and Proactivity

Personalized tool utilization is essential for aligning large language models (LLMs) with user preference in interaction scenarios with various tools. However, most of the current benchmarks primarily focus on either personalization of text generation or direct tool-utilizing, without considering both. In this work, we introduce a novel benchmark ETAPP for evaluating personalized tool invocation, establishing a sandbox environment, and a comprehensive dataset of 800 testing cases covering diverse user profiles. To improve the accuracy of our evaluation, we propose a key-point-based LLM evaluation method, mitigating biases in the LLM-as-a-judge system by manually annotating key points for each test case and providing them to LLM as the reference. Additionally, we evaluate the excellent LLMs and provide an in-depth analysis. Furthermore, we investigate the impact of different tool-invoking strategies on LLMs' personalization performance and the effects of fine-tuning in our task. The effectiveness of our preference-setting and key-point-based evaluation method is also validated. Our findings offer insights into improving personalized LLM agents. Our Code is available at https://github.com/hypasd-art/ETAPP.

cs.CL↗

arXiv2025

Pangu Ultra: Pushing the Limits of Dense Large Language Models on Ascend NPUs

We present Pangu Ultra, a Large Language Model (LLM) with 135 billion parameters and dense Transformer modules trained on Ascend Neural Processing Units (NPUs). Although the field of LLM has been witnessing unprecedented advances in pushing the scale and capability of LLM in recent years, training such a large-scale model still involves significant optimization and system challenges. To stabilize the training process, we propose depth-scaled sandwich normalization, which effectively eliminates loss spikes during the training process of deep models. We pre-train our model on 13.2 trillion diverse and high-quality tokens and further enhance its reasoning capabilities during post-training. To perform such large-scale training efficiently, we utilize 8,192 Ascend NPUs with a series of system optimizations. Evaluations on multiple diverse benchmarks indicate that Pangu Ultra significantly advances the state-of-the-art capabilities of dense LLMs such as Llama 405B and Mistral Large 2, and even achieves competitive results with DeepSeek-R1, whose sparse model structure contains much more parameters. Our exploration demonstrates that Ascend NPUs are capable of efficiently and effectively training dense models with more than 100 billion parameters. Our model and system will be available for our commercial customers.

cs.CL↗

arXiv2025

The non-abelian extension and Wells map of Leibniz conformal algebra

In this paper, we study the theory of non-abelian extensions of a Leibniz conformal algebra $R$ by a Leibniz conformal algebra $H$ and prove that all the non-abelian extensions are classified by non-abelian $2$nd cohomology $H^2_{nab}(R,H)$ in the sense of equivalence. Then we introduce a differential graded Lie algebra $\mathfrak{L}$ and show that the set of its Maurer-Cartan elements in bijection with the set of non-abelian extensions. Finally, as an application of non-abelian extension, we consider the inducibility of a pair of automorphisms about a non-abelian extension, and give the fundamental sequence of Wells of Leibniz conformal algebra $R$. Especially, we discuss the extensibility problem of derivations about an abelian extension of $R$.

math.RA↗

arXiv2025

Libertas: Privacy-Preserving Collective Computation for Decentralised Personal Data Stores

Data and data processing have become an indispensable aspect for our society. Insights drawn from collective data make invaluable contribution to scientific and societal research and business. But there are increasing worries about privacy issues and data misuse. This has prompted the emergence of decentralised personal data stores (PDS) like Solid that provide individuals more control over their personal data. However, existing PDS frameworks face challenges in ensuring data privacy when performing collective computations with data from multiple users. While Secure Multi-Party Computation (MPC) offers input secrecy protection during the computation without relying on any single party, issues emerge when directly applying MPC in the context of PDS, particularly due to key factors like autonomy and decentralisation. In this work, we discuss the essence of this issue, identify a potential solution, and introduce a modular architecture, Libertas, to integrate MPC with PDS like Solid, without requiring protocol-level changes. We introduce a paradigm shift from an `omniscient' view to individual-based, user-centric view of trust and security, and discuss the threat model of Libertas. Two realistic use cases for collaborative data processing are used for evaluation, both for technical feasibility and empirical benchmark, highlighting its effectiveness in empowering gig workers and generating differentially private synthetic data. The results of our experiments underscore Libertas' linear scalability and provide valuable insights into compute optimisations, thereby advancing the state-of-the-art in privacy-preserving data processing practices. By offering practical solutions for maintaining both individual autonomy and privacy in collaborative data processing environments, Libertas contributes significantly to the ongoing discourse on privacy protection in data-driven decision-making contexts.

cs.NI↗