Searcharxiv⌕ Search

arXiv subjects

Dong Dong

Publications and source records attributed to Dong Dong.

15 recordsLinked to original sources
arXiv2026

Cordon: Semantic Transactions for Tool-Using LLM Agents

Tool-using LLM agents are shifting the unit of computation from explicit human-issued commands to model-driven tasks with stateful consequences. Yet today's agent runtimes still expose tools as isolated RPCs. This interface gives runtimes a convenient integration point, but it lacks a task-scoped execution boundary for commit, rollback, recovery, and audit across multi-step agent workflows. We argue that this mismatch calls for a runtime containment boundary rather than another per-call guardrail. This paper introduces Cordon, a transactional runtime system for staging and validating irreversible agent effects before commit. A semantic transaction is a task-level execution boundary that binds tool intents and runtime-tracked result lineage to reversible local state, staged external effects, delegated authority, and audit metadata. Cordon implements this abstraction with a transaction manager that tracks derived result objects, executes reversible mutations in shadow state, stages outward-facing actions in an effect outbox, and records recovery metadata. The runtime then validates the composed execution flow before it commits state or releases external effects. Our evaluation across adversarial and benign workflows shows that Cordon exposes cross-step violations missed by existing defenses. It also reduces irreversible-effect failures while preserving benign task completion with modest approval and latency overhead.

cs.OS↗
arXiv2026

SkillSmith: Compiling Agent Skills into Boundary-Guided Runtime Interfaces

Recently, skills have been widely adopted in large language model (LLM)-based agent systems across various domains. In existing frameworks, skills are typically injected into the agent reasoning loop as contextual guidance once matched to a runtime task, enabling specialized task-solving capabilities. We find that this execution paradigm introduces two major sources of redundancy: irrelevant context injection and repeated skill-specific reasoning and planning. To this end, we propose SkillSmith, a boundary-first compiler-runtime framework that compiles skill packages offline into minimal executable interfaces. By extracting fine-grained operational boundaries from skills, SkillSmith enables agents to dynamically access and execute only the relevant components at runtime, thereby minimizing unnecessary context injection and redundant reasoning overhead. In the evaluation on SkillsBench benchmark, SkillSmith reduces solve-stage token usage by 57.44%, thinking iterations by 42.99%, solve time by 50.57% (2.02x faster), and token-proportional monetary cost by 57.44% compared with using raw-skills. Moreover, compiled artifacts produced by a stronger model can be reused by a smaller or more efficient runtime model, improving task accuracy in cases where raw skill interpretation fails. The source code and data are available at https://github.com/AetherHeart-AI/Aeloon.

cs.AI↗
arXiv2025

Length-MAX Tokenizer for Language Models

We introduce a new tokenizer for language models that minimizes the average tokens per character, thereby reducing the number of tokens needed to represent text during training and to generate text during inference. Our method, which we refer to as the Length-MAX tokenizer, obtains its vocabulary by casting a length-weighted objective maximization as a graph partitioning problem and developing a greedy approximation algorithm. On FineWeb and diverse domains, it yields 14--18\% fewer tokens than Byte Pair Encoding (BPE) across vocabulary sizes from 10K to 50K, and the reduction is 13.0\% when the size is 64K. Training GPT-2 models at 124M, 355M, and 1.3B parameters from scratch with five runs each shows 18.5\%, 17.2\%, and 18.5\% fewer steps, respectively, to reach a fixed validation loss, and 13.7\%, 12.7\%, and 13.7\% lower inference latency, together with a 16\% throughput gain at 124M, while consistently improving on downstream tasks including reducing LAMBADA perplexity by 11.7\% and enhancing HellaSwag accuracy by 4.3\%. Moreover, the Length-MAX tokenizer achieves 99.62\% vocabulary coverage and the out-of-vocabulary rate remains low at 0.12\% on test sets. These results demonstrate that optimizing for average token length, rather than frequency alone, offers an effective approach to more efficient language modeling without sacrificing -- and often improving -- downstream performance. The tokenizer is compatible with production systems and reduces embedding and KV-cache memory by 18\% at inference.

cs.CL↗
arXiv2025

Holographic Transformers for Complex-Valued Signal Processing: Integrating Phase Interference into Self-Attention

Complex-valued signals encode both amplitude and phase, yet most deep models treat attention as real-valued correlation, overlooking interference effects. We introduce the Holographic Transformer, a physics-inspired architecture that incorporates wave interference principles into self-attention. Holographic attention modulates interactions by relative phase and coherently superimposes values, ensuring consistency between amplitude and phase. A dual-headed decoder simultaneously reconstructs the input and predicts task outputs, preventing phase collapse when losses prioritize magnitude over phase. We demonstrate that holographic attention implements a discrete interference operator and maintains phase consistency under linear mixing. Experiments on PolSAR image classification and wireless channel prediction show strong performance, achieving high classification accuracy and F1 scores, low regression error, and increased robustness to phase perturbations. These results highlight that enforcing physical consistency in attention leads to generalizable improvements in complex-valued learning and provides a unified, physics-based framework for coherent signal modeling. The code is available at https://github.com/EonHao/Holographic-Transformers.

eess.SP↗
arXiv2024

Human-LLM Collaborative Construction of a Cantonese Emotion Lexicon

Large Language Models (LLMs) have demonstrated remarkable capabilities in language understanding and generation. Advanced utilization of the knowledge embedded in LLMs for automated annotation has consistently been explored. This study proposed to develop an emotion lexicon for Cantonese, a low-resource language, through collaborative efforts between LLM and human annotators. By integrating emotion labels provided by LLM and human annotators, the study leveraged existing linguistic resources including lexicons in other languages and local forums to construct a Cantonese emotion lexicon enriched with colloquial expressions. The consistency of the proposed emotion lexicon in emotion extraction was assessed through modification and utilization of three distinct emotion text datasets. This study not only validates the efficacy of the constructed lexicon but also emphasizes that collaborative annotation between human and artificial intelligence can significantly enhance the quality of emotion labels, highlighting the potential of such partnerships in facilitating natural language processing tasks for low-resource languages.

cs.HC↗
arXiv2024

SAM-Lightening: A Lightweight Segment Anything Model with Dilated Flash Attention to Achieve 30 times Acceleration

Segment Anything Model (SAM) has garnered significant attention in segmentation tasks due to their zero-shot generalization ability. However, a broader application of SAMs to real-world practice has been restricted by their low inference speed and high computational memory demands, which mainly stem from the attention mechanism. Existing work concentrated on optimizing the encoder, yet has not adequately addressed the inefficiency of the attention mechanism itself, even when distilled to a smaller model, which thus leaves space for further improvement. In response, we introduce SAM-Lightening, a variant of SAM, that features a re-engineered attention mechanism, termed Dilated Flash Attention. It not only facilitates higher parallelism, enhancing processing efficiency but also retains compatibility with the existing FlashAttention. Correspondingly, we propose a progressive distillation to enable an efficient knowledge transfer from the vanilla SAM without costly training from scratch. Experiments on COCO and LVIS reveal that SAM-Lightening significantly outperforms the state-of-the-art methods in both run-time efficiency and segmentation accuracy. Specifically, it can achieve an inference speed of 7 milliseconds (ms) per image, for images of size 1024*1024 pixels, which is 30.1 times faster than the vanilla SAM and 2.1 times than the state-of-the-art. Moreover, it takes only 244MB memory, which is 3.5\% of the vanilla SAM. The code and weights are available at https://anonymous.4open.science/r/SAM-LIGHTENING-BC25/.

cs.CV↗
arXiv2023

Xenos: Dataflow-Centric Optimization to Accelerate Model Inference on Edge Devices

Edge computing has been emerging as a popular scenario for model inference. However, the inference performance on edge devices (e.g., Multi-Core DSP, FGPA, etc.) suffers from inefficiency due to the lack of highly optimized inference frameworks. Previous model inference frameworks are mainly developed in an operator-centric way, which provides insufficient acceleration to edge-based inference. Besides, the operator-centric framework incurs significant costs for continuous development and maintenance. In this paper, we propose Xenos, which can automatically conduct dataflow-centric optimization of the computation graph and accelerate inference in two dimensions. Vertically, Xenos develops operator linking technique to improve data locality by restructuring the inter-operator dataflow. Horizontally, Xenos develops DSP-aware operator split technique to enable higher parallelism across multiple DSP units. Our evaluation proves the effectiveness of vertical and horizontal dataflow optimization, which reduce the inference time by 21.2\%--84.9\% and 17.9\%--96.2\% , respectively. Besides, Xenos also outperforms the widely-used TVM by 3.22$\times$--17.92$\times$. Moreover, we extend Xenos to a distributed solution, which we call d-Xenos. d-Xenos employs multiple edge devices to jointly conduct the inference task and achieves a speedup of 3.68x--3.78x compared with the single device.

cs.DC↗
arXiv2019

Transport Model for Feature Extraction

We present a new feature extraction method for complex and large datasets, based on the concept of transport operators on graphs. The proposed approach generalizes and extends the many existing data representation methodologies built upon diffusion processes, to a new domain where dynamical systems play a key role. The main advantage of this approach comes from the ability to exploit different relationships than those arising in the context of e.g., Graph Laplacians. Fundamental properties of the transport operators are proved. We demonstrate the flexibility of the method by introducing several diverse examples of transformations. We close the paper with a series of computational experiments and applications to the problem of classification of hyperspectral satellite imagery, to illustrate the practical implications of our algorithm and its ability to quantify new aspects of relationships within complicated datasets.

cs.LG↗
arXiv2019

Special cases of power decay in multilinear oscillatory integrals

We use tools from the multilinear oscillatory integral program developed by Christ, Li, Thiele, and Tao to treat special cases which are not covered by existing theory. Consideration of special cases leads to an extended class of examples for which $λ$-power decay holds.

math.CA↗
arXiv2018

On bilinear Hilbert transform along two polynomials

We prove that the bilinear Hilbert transform along two polynomials $B_{P,Q}(f,g)(x)=\int_{\mathbb{R}}f(x-P(t))g(x-Q(t))\frac{dt}{t}$ is bounded from $L^p \times L^q$ to $L^r$ for a large range of $(p,q,r)$, as long as the polynomials $P$ and $Q$ have distinct leading and trailing degrees. The same boundedness property holds for the corresponding bilinear maximal function $\mathcal{M}_{P,Q}(f,g)(x)=\sup_{ε>0}\frac{1}{2ε}\int_{-ε}^ε |f(x-P(t))g(x-Q(t))|dt$.

math.CA↗
arXiv2017

Discrete bilinear Radon transforms along arithmetic functions with many common values

We prove that for a large class of functions $P$ and $Q$, there exists $d\in (0,1)$ such that the discrete bilinear Radon transform $$B^{\rm dis}_{P,Q}(f,g)(n)=\sum_{m\in\mathbb{Z}\setminus\{0\}} f(n-P(m))g(n-Q(m))\frac{1}{m}$$ is bounded from $l^2\times l^2$ into $l^{1+ε}$ for any $ε\in (d,1)$. In particular, the boundedness holds for any $ε\in (0,1)$ when $P$ (or $Q$) is the Euler totient function $ϕ(|m|)$ or the prime counting function $π(|m|)$.

math.NT↗
arXiv2017

Improved estimates for polynomial Roth type theorems in finite fields

We prove that, under certain conditions on the function pair $φ_1$ and $φ_2$, bilinear average $p^{-1}\sum_{y\in \mathbb{F}_p}f_1(x+φ_1(y)) f_2(x+φ_2(y))$ along curve $(φ_1, φ_2)$ satisfies certain decay estimate. As a consequence, Roth type theorems hold in the setting of finite fields. In particular, if $φ_1,φ_2\in \mathbb{F}_p[X]$ with $φ_1(0)=φ_2(0)=0$ are linearly independent polynomials, then for any $A\subset \mathbb{F}_p, |A|=δp$ with $δ>c p^{-\frac{1}{12}}$, there are $\gtrsim δ^3p^2$ triplets $x,x+φ_1(y), x+φ_2(y)\in A$. This extends a recent result of Bourgain and Chang who initiated this type of problems, and strengthens the bound in a result of Peluse, who generalized Bourgain and Chang's work. The proof uses discrete Fourier analysis and algebraic geometry.

math.NT↗
arXiv2015

Effect of carbon vacancies on structural and mechanical properties of stable zirconium carbides: A first principles study

By using evolutionary algorithm USPEX, we have predicted a number of stable zirconium carbides. In addition to the well-known rocksalt-type stoichiometric ZrC (Fm-3m), present prediction also identifies five stable substoichiometric zirconium carbides adopting rocksalt-type structures with ordered carbon vacancies, Zr8C7 (P-1), Zr6C5 (C2/m), Zr5C4 (P-1), Zr3C2 (C2/m), and Zr2C (Fd-3m). The effects of carbon vacancies on structural and mechanical properties are investigated. We highlight that the distribution of carbon vacancies has significant influence on volume, Pugh's ratio, and hardness. We further propose that hardness can be enhanced by replacing carbon vacancies with suitable elements, in particular nitrogen and oxygen.

cond-mat.mtrl-sci↗
arXiv2013

Prediction of stable hafnium carbides: their stoichiometries, mechanical properties, and electronic structure

Hafnium carbides are studied by a systematic search for possible stable stoichiometric compounds in the Hf-C system at ambient pressure using variable-composition ab initio evolutionary algorithm implemented in the USPEX code. In addition to well-known HfC, we predicted two additional compounds Hf3C2 and Hf6C5. The structure of Hf6C5 with space group C2/m contains 11 atoms in the primitive cell and this prediction revives the earlier proposal by A. I. Gusev. The stable structure of Hf3C2 also has space group C2/m, and is more energetically favorable than the Immm, P-3m1, P2 and C2221 structures put forward by A. I. Gusev. Dynamical and mechanical stability of the newly predicted structures have been verified by calculations of their phonons and elastic constants. The bulk and shear moduli of Hf3C2 are 195.8 GPa and 143.1 GPa, respectively, while for Hf6C5 they are 227.9 GPa and 187.2 GPa, respectively. Their mechanical properties are inferior to those of HfC due to the presence of structural vacancies. Chemical bonding, band structure, and Bader charge are presented and discussed.

cond-mat.mtrl-sci↗