arXiv subjects

Jin Dong

Publications and source records attributed to Jin Dong.

At least 19 recordsLinked to original sources

arXiv2026

Scalable Composition of Byzantine Agreements under Reorder Attacks

Byzantine agreement (BA) is a foundational building block in distributed systems, and the security analysis of BA protocols under multi-instance executions has attracted increasing attention. However, most existing adversary models focus solely on party corruption and neglect important threats posed by adversarial manipulations of communication channels in the network. Through channel attacks, messages can be reordered across multiple executions and lead to violations of the protocol's security guarantees, In this work, we present the first adversary model that combines party corruption and channel attacks. Based on this model, we establish new security thresholds for Byzantine agreement under parallel and concurrent compositions, supported by complementary impossibility and possibility results that match each other to form a tight bound. For the impossibility result, we show that even authenticated Byzantine agreement protocols cannot be secure under parallel composition when $n \leq 3t$ or $n \leq 2c + 2t + 1$, where $t$ and $c$ denote the number of corrupted parties and communication channels, respectively, and $n$ is the number of parties. For the possibility result, we prove the existence of secure protocols for unauthenticated Byzantine agreement under parallel and concurrent composition, when $n > \max\{3t, 2c+2t+1\}$. We first provide general black-box compilers that transform any single-instance secure BA protocol into one that is secure under parallel and concurrent executions without additional security assumptions. To optimize performance, we further design refined compilers using erasure-correcting codes. These refined versions significantly reduce communication overhead, particularly for long messages, where they achieve a constant multiplicative overhead compared with the original protocol, thus achieving the same asymptotic communication complexity.

cs.CR↗

arXiv2026

Subleading Collinear Limits of Yang--Mills Amplitudes from Recursions

We study the subleading adjacent-collinear limit of tree-level color-ordered Yang--Mills amplitudes. In general dimensions, a Lorentz null rotation transports the polarizations of collinear legs along the collinear path while preserving transversality and gauge equivalence. The finite coefficient then separates into a pole-free hard contribution, computed by a channel-deleted Berends--Giele recursion, and an explicit contribution from the adjacent factorization channel. In four dimensions, we construct the same coefficient directly from a collinear-aware Britto-Cachazo-Feng-Witten recursion whose terminal data are closed MHV and $\overline{\rm MHV}$ formulas. Both constructions apply at arbitrary momentum fraction and extend to collinear gluons with distinct polarizations; at the symmetric split, the auxiliary-vector dependence cancels for equal polarizations and for the polarization-symmetrized mixed limit. Open--closed disk relations then give a direct construction of one-graviton Einstein--Yang--Mills amplitudes and their higher-derivative string corrections from the recursively generated coefficients. A self-contained \textsc{Mathematica} implementation accompanies the paper.

hep-th↗

arXiv2026

Subleading Collinear Limits of Yang-Mills Amplitudes from Gravity

We show that the subleading collinear sector of Yang-Mills (YM) amplitudes is controlled by ordinary Einstein-Yang-Mills (EYM) amplitudes and their higher-derivative corrections. The complete set of strict subleading collinear limits associated with an equal-helicity collinear pair can be extracted from open-closed string disk amplitudes. The latter generate $\frac12 (n-3)!$ BCJ-like relations with non-linear kinematic coefficients, reducing the $(n-3)!$ collinear data to a basis of dimension $\frac12 (n-3)!$. This reduced basis can be represented by gravitational amplitudes in EYM theories and by their higher-derivative open-string corrections. At multiplicity $n$ this gravitational basis consists of the $(n-4)!$ independent EYM subamplitudes together with $\frac12 (n-5)(n-4)!$ higher-order corrections involving one graviton and $n-2$ gluons. The resulting decomposition is governed by unsigned Stirling numbers of the first kind: the ordinary EYM amplitudes correspond to the sector $\left[{n-3\atop 1}\right]$, the higher-order BCJ-like relations to even sectors $\left[{n-3\atop 2j}\right]$, and the higher-derivative EYM corrections to odd sectors $\left[{n-3\atop 2j+1}\right]$ with $j\geq 1$.

hep-th↗

arXiv2026

Keeping Data Centers Online in Weak Grids: PLL-Free VM-DPC With Adaptive Reactive-Power Support for Centralized UPS Systems

Data center power systems are increasingly exposed to weak-grid conditions due to the rapid growth of converter-dominated networks and highly dynamic artificial intelligence (AI) workloads. In centralized uninterruptible power supply (UPS) architectures, the front-end rectifier continuously processes the incoming facility power, making its dynamic performance critical for ensuring stable operation and reliable power delivery to information technology (IT) equipment. Under weak-grid conditions, conventional phase-locked loop (PLL)-based proportional-integral (PI) rectifier controllers may exhibit instability due to strong interactions between converter control dynamics and grid impedance. This paper investigates the stability of centralized UPS data center systems operating under weak-grid conditions using a detailed switching-level model developed in MATLAB/Simulink and validated in real time using an OPAL-RT platform. To enhance weak-grid stability and improve converter-grid interaction, a voltage-modulated direct power control (VM-DPC) strategy with adaptive reactive power support is applied to the front-end rectifier. The proposed approach directly regulates active and reactive power without PLL synchronization while dynamically supporting the point of common coupling (PCC) voltage during rapid IT load variations. Results demonstrate that conventional PI-based rectifier control becomes unstable under SCR<=2 conditions, leading to dc-link oscillations and degradation of downstream power delivery. In contrast, the proposed VM-DPC strategy restores stable operation, improves system damping, and maintains reliable power transfer to highly dynamic IT loads under weak-grid operation.

eess.SY↗

arXiv2026

Stability Enhancement of Centralized UPS Data Center Systems Under Weak-Grid Conditions

Data center power systems are increasingly exposed to weak-grid conditions due to the evolution of modern power systems and the integration of large and dynamic loads. In centralized uninterruptible power supply (UPS) architectures, the front-end rectifier plays a critical role in maintaining stable operation and ensuring reliable power delivery to information technology (IT) equipment. However, conventional phase-locked loop (PLL)-based proportional-integral (PI) control strategies may exhibit degraded performance or instability under low short-circuit ratio (SCR) conditions. This paper investigates the behavior of centralized UPS systems under weak-grid conditions and demonstrates, through electromagnetic transient simulations, that PI-controlled rectifiers can become unstable at SCR=2. To address this issue, a power-based control approach is applied to the three-phase rectifier, enabling direct regulation of active and reactive power without relying on PLL synchronization. Simulation results show that the proposed control strategy improves system damping and restores stable operation under weak-grid conditions. The findings highlight the importance of control design for maintaining reliable operation of data center power systems in emerging low-strength grid environments.

eess.SY↗

arXiv2026

Grid-Orch: An LLM-Powered Orchestrator for Distribution Grid Simulation and Analytics

The power distribution engineering workforce faces a projected shortage of up to 1.5 million engineers by 2030, creating urgent demand for more accessible analysis tools. This paper introduces Grid-Orch, a framework that bridges Large Language Models (LLMs) and power system simulation through the Model Context Protocol (MCP), enabling engineers to perform complex distribution analyses via natural language. Using OpenDSS as the reference implementation, Grid-Orch provides 36 domain-specific tools across eleven categories, covering power flow, voltage analysis, quasi-static time series (QSTS) simulation, and automated optimization. A provider-agnostic LLM layer supports both cloud-hosted (Gemini, Claude) and locally deployed (Ollama, llama-cpp) models, enabling air-gapped operation for security-sensitive utility environments. Three optimization skills, capacitor placement, voltage violation analysis, and overvoltage mitigation, extend the platform beyond single-tool queries to multi-step engineering workflows. Grid-Orch is delivered as an interactive web platform with chat-based interaction, a QSTS dashboard, and feeder topology visualization, and renders simulation results inline. Workflow demonstrations show that distribution analyses formerly requiring hours of scripting, such as distributed energy resource (DER) interconnection screening, complete in under two minutes through natural language, producing numerically identical results to direct OpenDSS scripting.

eess.SY↗

arXiv2026

UFO: Unlocking Ultra-Efficient Quantized Private Inference with Protocol and Algorithm Co-Optimization

Private convolutional neural network (CNN) inference based on secure two-party computation (2PC) suffers from high communication and latency overhead, especially from convolution layers. In this paper, we propose UFO, a quantized 2PC inference framework that jointly optimizes the 2PC protocols and quantization algorithm. UFO features a novel 2PC protocol that systematically combines the efficient Winograd convolution algorithm with quantization to improve inference efficiency. However, we observe that naively combining quantization and Winograd convolution faces the following challenges: 1) From the inference perspective, Winograd transformations introduce extensive additions and require frequent bit width conversions to avoid inference overflow, leading to non-negligible communication overhead; 2) From the training perspective, Winograd transformations introduce weight outliers that make quantization-aware training (QAT) difficult, resulting in inferior model accuracy. To address these challenges, we co-optimize both protocol and algorithm. 1) At the protocol level, we propose a series of graph-level optimizations for 2PC inference to minimize the communication. 2) At the algorithm level, we develop a mixed-precision QAT algorithm based on layer sensitivity to optimize model accuracy given communication constraints. To accommodate the outliers, we further introduce a 2PC-friendly bit re-weighting algorithm to increase the representation range without explicitly increasing bit widths. With extensive experiments, UFO demonstrates 11.7x, 3.6x, and 6.3x communication reduction with 1.29%, 1.16%, and 1.29% higher accuracy compared to state-of-the-art frameworks SiRNN, COINN, and CoPriv, respectively.

cs.CR↗

arXiv2025

On Positive Celestial Geometry: ABHY in the Sky

Celestial amplitudes are multiple Mellin transforms w.r.t. conformal dimensions. For arbitrary multiplicity $n$ of massless states in sufficiently high space--time dimension $D$ we perform all Mellin integrations and find an associahedron description in celestial space. The latter expresses celestial tree--level $\phi^3$ amplitudes as the canonical forms associated with this positive geometry. This yields a geometric interpretation of celestial amplitudes in terms of the underlying boundary geometry. Our universal treatment of Mellin integrals in $D$ dimensions also provides a unified description of celestial amplitudes arising from different bulk theories.

hep-th↗

arXiv2025

On differential operators for scalar-scaffolded gluons

Recently, based on the curve-integral formulation for stringy Tr$\phi^3$ amplitudes, a combinatorial formulation for Yang-Mills amplitudes has been proposed which describes gluons using pairs of scalars and produces the $n$-gluon amplitude from simple kinematical shift of stringy Tr$\phi^3$ amplitudes with $2n$ scalars. It has revealed a variety of new properties and structures even for tree-level gluon amplitudes such as hidden zeros and splits, and in this note we provide another example: we study differential operators acting on Yang-Mills amplitudes with respect to $2n$-scalar kinematic variables, which convert such scalar-scaffolded gluons into scalars. In particular, we find $(n{-}1)$-fold differential operators (using $2n$-scalar variables) that turn the $n$-gluon amplitude into a single planar $\phi^3$ diagram; we then generalize such operators to those that convert $n$ gluons to mixed amplitudes with $r$ scalars and $n{-}r$ gluons (the latter can be viewed as insertions on $\phi^3$ diagrams). We also show that the number of linearly independent mixed amplitudes with $r$ scalars and $n-r$ gluons is given by the number of $\phi^3$ diagrams, the Catalan number $\mathcal{C}_{r-2}$, which can be viewed as a generalization of the ``uniqueness" theorem of gluon amplitudes (with $r=0$). Finally, our construction leads to a planar version of the universal expansion of Yang-Mills amplitudes into a sum of gauge-invariant prefactors built from nested commutators, each accompanied by an mixed amplitude in the natural basis. This formulation significantly reduces the redundancy present in the original expansion.

hep-th↗

arXiv2025

ORCA: Mitigating Over-Reliance for Multi-Task Dwell Time Prediction with Causal Decoupling

Dwell time (DT) is a critical post-click metric for evaluating user preference in recommender systems, complementing the traditional click-through rate (CTR). Although multi-task learning is widely adopted to jointly optimize DT and CTR, we observe that multi-task models systematically collapse their DT predictions to the shortest and longest bins, under-predicting the moderate durations. We attribute this moderate-duration bin under-representation to over-reliance on the CTR-DT spurious correlation, and propose ORCA to address it with causal-decoupling. Specifically, ORCA explicitly models and subtracts CTR's negative transfer while preserving its positive transfer. We further introduce (i) feature-level counterfactual intervention, and (ii) a task-interaction module with instance inverse-weighting, weakening CTR-mediated effect and restoring direct DT semantics. ORCA is model-agnostic and easy to deploy. Experiments show an average 10.6% lift in DT metrics without harming CTR. Code is available at https://github.com/Chrissie-Law/ORCA-Mitigating-Over-Reliance-for-Multi-Task-Dwell-Time-Prediction-with-Causal-Decoupling.

cs.IR↗

arXiv2025

Proto-EVFL: Enhanced Vertical Federated Learning via Dual Prototype with Extremely Unaligned Data

In vertical federated learning (VFL), multiple enterprises address aligned sample scarcity by leveraging massive locally unaligned samples to facilitate collaborative learning. However, unaligned samples across different parties in VFL can be extremely class-imbalanced, leading to insufficient feature representation and limited model prediction space. Specifically, class-imbalanced problems consist of intra-party class imbalance and inter-party class imbalance, which can further cause local model bias and feature contribution inconsistency issues, respectively. To address the above challenges, we propose Proto-EVFL, an enhanced VFL framework via dual prototypes. We first introduce class prototypes for each party to learn relationships between classes in the latent space, allowing the active party to predict unseen classes. We further design a probabilistic dual prototype learning scheme to dynamically select unaligned samples by conditional optimal transport cost with class prior probability. Moreover, a mixed prior guided module guides this selection process by combining local and global class prior probabilities. Finally, we adopt an \textit{adaptive gated feature aggregation strategy} to mitigate feature contribution inconsistency by dynamically weighting and aggregating local features across different parties. We proved that Proto-EVFL, as the first bi-level optimization framework in VFL, has a convergence rate of 1/\sqrt T. Extensive experiments on various datasets validate the superiority of our Proto-EVFL. Even in a zero-shot scenario with one unseen class, it outperforms baselines by at least 6.97%

cs.LG↗

arXiv2025

H2Tune: Federated Foundation Model Fine-Tuning with Hybrid Heterogeneity

Different from existing federated fine-tuning (FFT) methods for foundation models, hybrid heterogeneous federated fine-tuning (HHFFT) is an under-explored scenario where clients exhibit double heterogeneity in model architectures and downstream tasks. This hybrid heterogeneity introduces two significant challenges: 1) heterogeneous matrix aggregation, where clients adopt different large-scale foundation models based on their task requirements and resource limitations, leading to dimensional mismatches during LoRA parameter aggregation; and 2) multi-task knowledge interference, where local shared parameters, trained with both task-shared and task-specific knowledge, cannot ensure only task-shared knowledge is transferred between clients. To address these challenges, we propose H2Tune, a federated foundation model fine-tuning with hybrid heterogeneity. Our framework H2Tune consists of three key components: (i) sparsified triple matrix decomposition to align hidden dimensions across clients through constructing rank-consistent middle matrices, with adaptive sparsification based on client resources; (ii) relation-guided matrix layer alignment to handle heterogeneous layer structures and representation capabilities; and (iii) alternating task-knowledge disentanglement mechanism to decouple shared and specific knowledge of local model parameters through alternating optimization. Theoretical analysis proves a convergence rate of O(1/\sqrt{T}). Extensive experiments show our method achieves up to 15.4% accuracy improvement compared to state-of-the-art baselines. Our code is available at https://anonymous.4open.science/r/H2Tune-1407.

cs.LG↗

arXiv2025

TinyAlign: Boosting Lightweight Vision-Language Models by Mitigating Modal Alignment Bottlenecks

Lightweight Vision-Language Models (VLMs) are indispensable for resource-constrained applications. The prevailing approach to aligning vision and language models involves freezing both the vision encoder and the language model while training small connector modules. However, this strategy heavily depends on the intrinsic capabilities of the language model, which can be suboptimal for lightweight models with limited representational capacity. In this work, we investigate this alignment bottleneck through the lens of mutual information, demonstrating that the constrained capacity of the language model inherently limits the Effective Mutual Information (EMI) between multimodal inputs and outputs, thereby compromising alignment quality. To address this challenge, we propose TinyAlign, a novel framework inspired by Retrieval-Augmented Generation, which strategically retrieves relevant context from a memory bank to enrich multimodal inputs and enhance their alignment. Extensive empirical evaluations reveal that TinyAlign significantly reduces training loss, accelerates convergence, and enhances task performance. Remarkably, it allows models to achieve baseline-level performance with only 40\% of the fine-tuning data, highlighting exceptional data efficiency. Our work thus offers a practical pathway for developing more capable lightweight VLMs while introducing a fresh theoretical lens to better understand and address alignment bottlenecks in constrained multimodal systems.

cs.LG↗

arXiv2025

Superstring amplitudes meet surfaceology

We reformulate tree-level amplitudes in open superstring theory (type-I) in terms of stringy Tr$(\phi^3)$ amplitudes with various kinematical shifts in the "curve-integral" formulation: while the bosonic-string amplitude with $n$ pairs of "scaffolding" scalars comes from a particularly simple shift of the Tr$(\phi^3)$ one (corresponding to $n$ length-$2$ cycles), the analogous superstring amplitude requires "correction" terms given by bosonic-string amplitudes with longer, even-length "cycles", which are also Tr$(\phi^3)$ ones at shifted kinematics dictated by the cycles; in total it is expressed as a sum of $(2n{-}3)!!$ shifted amplitudes originated from the expansion of a reduced Pfaffian. Upon taking $n$ scaffolding residues, this leads to a new formula of the $n$-gluon superstring amplitude, which is manifestly symmetric in $n{-}1$ legs, as a gauge-invariant combination of mixed bosonic string amplitudes with gluons and scalars, which come from length-$2$ cycles and longer ones respectively (the total sum is associated with the expansion a $n\times n$ symmetrical determinant); the corresponding prefactors are nested commutators of $2n$-gon kinematical variables, which nicely become traces of field-strengths for those legs corresponding to scalars in the mixed amplitudes. These interesting linear combinations of bosonic string amplitudes must guarantee the cancellation of tachyon poles and $F^3$ vertices ${\it etc.}$, and they give new relations between the superstring amplitude and its bosonic-string building blocks to all orders in the $\alpha'$ expansion (the first order gives a new formula for gluon amplitudes with a single $F^3$ insertion in terms of Yang-Mills-scalar amplitudes). We provide both the worldsheet and "curve-integral" derivations, and discuss applications to heterotic and type II cases.

hep-th↗

arXiv2025

CodeBC: A More Secure Large Language Model for Smart Contract Code Generation in Blockchain

Large language models (LLMs) excel at generating code from natural language instructions, yet they often lack an understanding of security vulnerabilities. This limitation makes it difficult for LLMs to avoid security risks in generated code, particularly in high-security programming tasks such as smart contract development for blockchain. Researchers have attempted to enhance the vulnerability awareness of these models by training them to differentiate between vulnerable and fixed code snippets. However, this approach relies heavily on manually labeled vulnerability data, which is only available for popular languages like Python and C++. For low-resource languages like Solidity, used in smart contracts, large-scale annotated datasets are scarce and difficult to obtain. To address this challenge, we introduce CodeBC, a code generation model specifically designed for generating secure smart contracts in blockchain. CodeBC employs a three-stage fine-tuning approach based on CodeLlama, distinguishing itself from previous methods by not relying on pairwise vulnerability location annotations. Instead, it leverages vulnerability and security tags to teach the model the differences between vulnerable and secure code. During the inference phase, the model leverages security tags to generate secure and robust code. Experimental results demonstrate that CodeBC outperforms baseline models in terms of BLEU, CodeBLEU, and compilation pass rates, while significantly reducing vulnerability rates. These findings validate the effectiveness and cost-efficiency of our three-stage fine-tuning strategy, making CodeBC a promising solution for generating secure smart contract code.

cs.CR↗

arXiv2025

Privacy-Preserving Federated Embedding Learning for Localized Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) has recently emerged as a promising solution for enhancing the accuracy and credibility of Large Language Models (LLMs), particularly in Question & Answer tasks. This is achieved by incorporating proprietary and private data from integrated databases. However, private RAG systems face significant challenges due to the scarcity of private domain data and critical data privacy issues. These obstacles impede the deployment of private RAG systems, as developing privacy-preserving RAG systems requires a delicate balance between data security and data availability. To address these challenges, we regard federated learning (FL) as a highly promising technology for privacy-preserving RAG services. We propose a novel framework called Federated Retrieval-Augmented Generation (FedE4RAG). This framework facilitates collaborative training of client-side RAG retrieval models. The parameters of these models are aggregated and distributed on a central-server, ensuring data privacy without direct sharing of raw data. In FedE4RAG, knowledge distillation is employed for communication between the server and client models. This technique improves the generalization of local RAG retrievers during the federated learning process. Additionally, we apply homomorphic encryption within federated learning to safeguard model parameters and mitigate concerns related to data leakage. Extensive experiments conducted on the real-world dataset have validated the effectiveness of FedE4RAG. The results demonstrate that our proposed framework can markedly enhance the performance of private RAG systems while maintaining robust data privacy protection.

cs.CL↗

arXiv2025

Unveiling Hidden Vulnerabilities in Digital Human Generation via Adversarial Attacks

Expressive human pose and shape estimation (EHPS) is crucial for digital human generation, especially in applications like live streaming. While existing research primarily focuses on reducing estimation errors, it largely neglects robustness and security aspects, leaving these systems vulnerable to adversarial attacks. To address this significant challenge, we propose the \textbf{Tangible Attack (TBA)}, a novel framework designed to generate adversarial examples capable of effectively compromising any digital human generation model. Our approach introduces a \textbf{Dual Heterogeneous Noise Generator (DHNG)}, which leverages Variational Autoencoders (VAE) and ControlNet to produce diverse, targeted noise tailored to the original image features. Additionally, we design a custom \textbf{adversarial loss function} to optimize the noise, ensuring both high controllability and potent disruption. By iteratively refining the adversarial sample through multi-gradient signals from both the noise and the state-of-the-art EHPS model, TBA substantially improves the effectiveness of adversarial attacks. Extensive experiments demonstrate TBA's superiority, achieving a remarkable 41.0\% increase in estimation error, with an average improvement of approximately 17.0\%. These findings expose significant security vulnerabilities in current EHPS models and highlight the need for stronger defenses in digital human generation systems.

cs.CV↗

arXiv2025

GPMFS: Global Foundation and Personalized Optimization for Multi-Label Feature Selection

As artificial intelligence methods are increasingly applied to complex task scenarios, high dimensional multi-label learning has emerged as a prominent research focus. At present, the curse of dimensionality remains one of the major bottlenecks in high-dimensional multi-label learning, which can be effectively addressed through multi-label feature selection methods. However, existing multi-label feature selection methods mostly focus on identifying global features shared across all labels, which overlooks personalized characteristics and specific requirements of individual labels. This global-only perspective may limit the ability to capture label-specific discriminative information, thereby affecting overall performance. In this paper, we propose a novel method called GPMFS (Global Foundation and Personalized Optimization for Multi-Label Feature Selection). GPMFS firstly identifies global features by exploiting label correlations, then adaptively supplements each label with a personalized subset of discriminative features using a threshold-controlled strategy. Experiments on multiple real-world datasets demonstrate that GPMFS achieves superior performance while maintaining strong interpretability and robustness. Furthermore, GPMFS provides insights into the label-specific strength across different multi-label datasets, thereby demonstrating the necessity and potential applicability of personalized feature selection approaches.

cs.LG↗