arXiv subjects

Jaehoon Kim

Publications and source records attributed to Jaehoon Kim.

At least 19 recordsLinked to original sources

arXiv2026

Self-Evolving Search Index

Information retrieval is increasingly important as LLM agents tackle complex tasks involving diverse information needs. Because retrieval relies on an index that represents each document through index keys, retrieval quality depends heavily on how effectively these keys expose the knowledge contained in each document. However, effective index representations vary across retrieval environments, making it difficult for any fixed optimization strategy to perform consistently. Yet evolving an index to its retrieval environment remains largely human-driven, requiring humans to diagnose retrieval failures, refine the optimization strategy, and reprocess the index accordingly. We propose SELF-INDEX, a framework that enables an index to self-evolve without human intervention. Its Optimizer autonomously diagnoses retrieval shortfalls, selectively revises the responsible index keys, and validates each revision before updating the index. Beyond reacting to observed retrieval demands, SELF-INDEX proactively explores additional demands through a Query Simulator, allowing the index to evolve beyond the queries already available for optimization. Across diverse corpora and retrievers, SELF-INDEX consistently improves retrieval performance while outperforming existing index optimization methods. We further show that these benefits extend to downstream applications, improving the effectiveness and efficiency of search agents and helping agent memory systems retrieve useful past interactions.

cs.IR↗

arXiv2026

Learning Robust Dexterous In-Hand Manipulation from Joint Sensors with Proprioceptive Transformer

In-hand object manipulation is a fundamental yet challenging capability for dexterous robots. Despite significant progress in dexterous manipulation, existing approaches rely heavily on vision or tactile sensing to track object states, while joint sensing -- the most readily available modality on any robotic hand -- remains largely overlooked, particularly for tendon-driven hands. In this paper, we study how far joint sensing alone can go by asking: (i) whether motor encoders or direct joint sensing provides better proprioceptive feedback, (ii) how to extract environment information from joint measurements, and (iii) whether joint-only control can achieve competitive real-world performance without external perception. We present the Proprioceptive Transformer (PT), an exteroceptive-free approach for continuous cube rotation on a tendon-driven dexterous hand that uses only joint sensing feedback. A teacher policy is first trained via reinforcement learning with privileged object information, then distilled into PT, which operates solely on joint position and velocity histories. The Transformer architecture effectively extracts implicit object state information from temporal patterns in joint sensor readings. Experiments on the real ORCA hand show that our approach achieves 3.1x higher rotation speed than baselines. We also demonstrate that our PT achieves a 23.4% lower RMSE for cube position estimation than the MLP baseline, indicating superior extraction of exteroceptive information from proprioceptive sources.

cs.RO↗

arXiv2026

OPSD Compresses What RLVR Teaches: A Post-RL Compaction Stage for Reasoning Models

On-Policy Self-Distillation (OPSD) has recently emerged as an alternative to Reinforcement Learning with Verifiable Rewards (RLVR), promising higher accuracy and shorter responses through token-level credit assignment from a self-teacher conditioned on privileged context. However, this promise does not carry over to thinking-enabled mathematical reasoning, where reported accuracy gains shrink and sometimes turn negative. We hypothesize that hindsight supervision can specify better token-level alternatives in short thinking-disabled outputs, but in long thinking-enabled traces it more readily identifies redundancy than supplies better replacements. To test this, we applied OPSD separately to correct and incorrect rollout groups, so that compression and correction can be observed in isolation. Our results show that in thinking-enabled mathematical reasoning, OPSD behaves most reliably as a compression mechanism rather than a correction mechanism: training only on correct rollouts preserves accuracy while substantially shortening responses, whereas training only on incorrect rollouts damages accuracy. In light of these findings, we propose a revised post-training pipeline for thinking-enabled mathematical reasoning: SFT then RLVR then OPSD.

cs.AI↗

arXiv2026

High-Speed, Scalable Sensor Readout for Dexterous Robotic Hands via Shift-Register Multiplexing

Dexterous robotic hands require high-speed multimodal sensing across many degrees of freedom, yet existing readout architectures often impose trade-offs between sensor count, wiring complexity, and sampling bandwidth. This paper presents a scalable analog sensor readout architecture based on a serial-in parallel-out (SIPO) shift-register principle. The proposed architecture supports versatile integration of heterogeneous analog-output sensors, scalable expansion using only three signal lines between sensor modules, and fast, configurable sampling. We validate the approach on a tendon-driven robotic hand integrating 16 joint sensor modules and one four-channel tactile sensor module, enabling acquisition of 20 sensor channels at a full-scan rate of 1 kHz, with stable operation up to 1.5 kHz. Joint sensor characterization showed a maximum slope absolute percentage error (APE) of 0.446% and sub-degree estimation error, indicating that the proposed readout system does not significantly degrade sensing performance. For tactile sensing, LSTM-based models achieved an RMSE of 0.125 N for force estimation and 93.4% accuracy for five-class contact-location classification, and were deployed for real-time inference at 1 kHz. System-level experiments showed that the joint sensors provide more accurate feedback than motor-based estimation during interaction, while the tactile sensor enables responsive force estimation in contact. The proposed architecture offers a practical path toward fully sensorized robotic hands for dexterous manipulation.

cs.RO↗

arXiv2026

Atomic-Scale Mechanisms of SiO$_2$ Plasma-Enhanced Chemical Vapor Deposition Revealed by Molecular Dynamics with a Machine-Learning Interatomic Potential

Plasma-enhanced chemical vapor deposition (PECVD) of silicon dioxide (SiO$_2$) is widely used for low-temperature fabrication of dielectric thin films, yet its atomic-scale growth mechanisms remain incompletely understood. In this work, we investigate SiO$_2$ PECVD using silane and N$_2$O as source gases via molecular dynamics simulations driven by a machine-learning interatomic potential. By systematically varying the oxidant-to-silane-derived species ratio $r$, we elucidate the evolution of film stoichiometry, density, and hydrogen content. Formation of the Si-O-Si network primarily proceeds via oxidation of surface Si-H groups to form Si-OH species, followed by condensation of neighboring Si-OH groups that produces H$_2$O as the dominant byproduct. At low $r$, H$_2$ formation via reactions between Si-H and Si-OH groups also contributes to the network formation. Increasing oxidant supply promotes the network formation through oxidation of residual Si-H species, suppressing hydrogen incorporation and leading to saturation of the Si/O ratio. Rapid chemisorption of silane-derived species, together with steric hindrance from pre-deposited species, results in localized growth and surface roughness. We further show that high-kinetic-energy plasma species can etch SiO$_2$ films, which potentially limits growth rates and enhances surface roughness under high RF-power conditions. These results provide atomic-scale insight into PECVD growth and guidance for optimizing film composition and quality.

cond-mat.mtrl-sci↗

arXiv2026

Stability with minuscule structure for chromatic thresholds

The chromatic threshold $δ_χ(H)$ of a graph $H$ is the infimum of $d>0$ such that the chromatic number of every $n$-vertex $H$-free graph with minimum degree at least $d n$ is bounded by a constant depending only on $H$ and $d$. Allen, B{ö}ttcher, Griffiths, Kohayakawa, and Morris determined the chromatic threshold for every $H$; in particular, they showed that if $χ(H)=r\ge 3$, then $δ_χ(H) \in\{\frac{r-3}{r-2},~\frac{2 r-5}{2 r-3},~\frac{r-2}{r-1}\}$. While the chromatic thresholds have been completely determined, rather surprisingly the structural behaviors of extremal graphs near the threshold remain unexplored. In this paper, we establish the stability theorems for chromatic threshold problems. We prove that every $n$-vertex $H$-free graph $G$ with $δ(G)\ge (δ_χ(H)-o(1))n$ and $χ(G)=ω(1)$ must be structurally close to one of the extremal configurations. Furthermore, we give a stronger stability result when $H$ is a clique, showing that $G$ admits a partition into independent sets and a small subgraph on sublinear number of vertices. We show that this small subgraph has fractional chromatic number $2+o(1)$ and is homomorphic to a Kneser graph defined by subsets of a logarithmic size set; both these two bounds are best possible. This is the first stability result that captures the lower-order structural features of extremal graphs. We also study two variations of chromatic thresholds. Replacing chromatic number by its fractional counterpart, we determine the fractional chromatic thresholds for all graphs. Another variation is the bounded-VC chromatic thresholds, which was introduced by Liu, Shangguan, Skokan, and Xu very recently. Extending work of Łuczak and Thomass{é} on the triangle case, we determine the bounded-VC chromatic thresholds for all cliques.

math.CO↗

arXiv2026

On Universal Graphs for Trees and Tree-Like Graphs

Chung and Graham [J. London Math. Soc. 1983] claimed to prove that there exists an $n$-vertex graph $G$ with $ \frac{5}{2}n \log_2 n + O(n)$ edges that contains every $n$-vertex tree as a subgraph. Frati, Hoffmann and Tóth [Combin. Probab. Comput. 2023] discovered an error in the proof. By adding more edges to $G$ the error can be corrected, bringing the number of edges in $G$ to $\frac{7}{2}n \log_2 n + O(n). $ We make the first improvement to Chung and Graham's bound in over four decades by showing that there exists an $n$-vertex graph with $ \frac{14}{5}n \log_2 n + O(n) $ edges that contains every $n$-vertex tree as a subgraph. Furthermore, we generalise this bound for treewidth-$k$ graphs by showing that there exists a graph with $O(kn\log(n/k+1))$ edges that contains every $n$-vertex treewidth-$k$ graph as a subgraph. This is best possible in the sense that $Ω(kn\log(n/k+1))$ edges are required.

math.CO↗

arXiv2026

SafePlanner: Testing Safety of the Automated Driving System Plan Model

In this work, we present SafePlanner, a systematic testing framework for identifying safety-critical flaws in the Plan model of Automated Driving Systems (ADS). SafePlanner targets two core challenges: generating structurally meaningful test scenarios and detecting hazardous planning behaviors. To maximize coverage, SafePlanner performs a structural analysis of the Plan model implementation - specifically, its scene-transition logic and hierarchical control flow - and uses this insight to extract feasible scene transitions from code. It then composes test scenarios by combining these transitions with non-player vehicle (NPC) behaviors. Guided fuzzing is applied to explore the behavioral space of the Plan model under these scenarios. We evaluate SafePlanner on Baidu Apollo, a production-grade level 4 ADS. It generates 20635 test cases and detects 520 hazardous behaviors, grouped into 15 root causes through manual analysis. For four of these, we applied patches based on our analysis; the issues disappeared, and no apparent side effects were observed. SafePlanner achieves 83.63 percent function and 63.22 percent decision coverage on the Plan model, outperforming baselines in both bug discovery and efficiency.

cs.SE↗

arXiv2025

On the size of universal graphs for spanning trees

Chung and Graham [J. London Math. Soc., 1983] claimed that there exists an $n$-vertex graph $G$ containing all $n$-vertex trees as subgraphs that has at most $\frac{5}{2}n \log_2 n + O(n)$ edges. We identify an error in their proof. This error can be corrected by adding more edges, which increases the number of edges to $e(G) \leq \frac{7}{2}n \log_2 n + O(n)$. Moreover, we further improve this by showing that there exists such an $n$-vertex graph with at most $\left(5- \frac{1}{3}\right)n \log_3 n + O(n) \leq 2.945 n \log_2 n$ edges. This is the first improvement of the bound since Chung and Graham's pioneering work four decades ago.

math.CO↗

arXiv2025

Fragile minor-monotone parameters under random edge perturbation

We conduct a quantitative analysis of how many random edges need to be added to a base graph $H$ in order to significantly increase natural minor-monotone graph parameters of the resulting graph $R$. Specifically, we show that if $R$ is obtained from a connected graph $H$ by adding only a few random edges, the tree-width, genus, and Hadwiger number of $R$ become very large, irrespective of the structure of $H$.

math.CO↗

arXiv2025

Atomistic Insights into Cu/amorphous-Ta$_x$N Interfacial Adhesion via Machine Learning Interatomic Potentials: Effects of Stoichiometry and Interface Construction

Accurate understanding and control of interfacial adhesion between Cu and Ta$_x$N diffusion barriers are essential for ensuring the mechanical reliability and integrity of Cu interconnect systems in semiconductor devices. Amorphous tantalum nitride (a-Ta$_x$N) barriers are particularly attractive due to their superior barrier performance, attributed to the absence of grain boundaries. However, a systematic atomistic investigation of how varying Ta stoichiometries influences adhesion strength at Cu/a-Ta$_x$N interfaces remains lacking, hindering a comprehensive understanding of interface optimization strategies. In this study, we employ machine learning interatomic potentials (MLIPs) to perform steered molecular dynamics (SMD) simulations of Cu/a-Ta$_x$N interfaces. We simultaneously evaluate three distinct interface construction approaches--static relaxation, high-temperature annealing, and simulated Cu deposition--to comprehensively investigate their influence on adhesion strength across varying Ta compositions ($x=1, 2, 4$). Peak force and work of adhesion values from SMD simulations quantitatively characterize interface strength, while atomic stress and strain analyses elucidate detailed deformation behavior, highlighting the critical role of interfacial morphologies. Additionally, we explore the atomistic mechanisms underlying cohesive failure, revealing how targeted incorporation of Ta atoms into Cu layers enhances the cohesive strength of the interface. This study demonstrates how MLIP-driven simulations can elucidate atomic-scale relationships between interface morphology and adhesion behavior, providing insights that can guide future atomistic engineering strategies toward enhancing intrinsic barrier adhesion, potentially enabling liner-free interconnect technologies.

cond-mat.mtrl-sci↗

arXiv2025

In Their Own Words: Reasoning Traces Tailored for Small Models Make Them Better Reasoners

Transferring reasoning capabilities from larger language models to smaller ones through supervised fine-tuning often fails counterintuitively, with performance degrading despite access to high-quality teacher demonstrations. We identify that this failure stems from distributional misalignment: reasoning traces from larger models contain tokens that are low probability under the student's distribution, exceeding the internal representation capacity of smaller architectures and creating learning barriers rather than helpful guidance. We propose Reverse Speculative Decoding (RSD), a mechanism for generating student-friendly reasoning traces in which the teacher model proposes candidate tokens but the student model determines acceptance based on its own probability distributions, filtering low probability tokens. When applied to Qwen3-0.6B, direct distillation of s1K-1.1 reasoning trace data degrades average performance across major reasoning benchmarks by 20.5\%, while the same model trained on RSD-generated reasoning traces achieves meaningful improvements of 4.9\%. Our analysis reveals that low probability tokens constitute the critical bottleneck in reasoning ability transfer. However, cross-model experiments demonstrate that RSD traces are model-specific rather than universally applicable, indicating that distributional alignment must be tailored for each student architecture's unique internal representation.

cs.CL↗

arXiv2025

Atomistic insights into hydrogen migration in IGZO from machine-learning interatomic potential: linking atomic diffusion to device performance

Understanding hydrogen diffusion is critical for improving the reliability and performance of oxide thin-film transistors (TFTs), where hydrogen plays a key role in carrier modulation and bias instability. In this work, we investigate hydrogen diffusion in amorphous IGZO ($a$-IGZO) and $c$-axis aligned crystalline IGZO (CAAC-IGZO) using machine learning interatomic potential molecular dynamics (MLIP-MD) simulations. We construct accurate phase-specific MLIPs by fine-tuning SevenNet-0, a universal pretrained MLIP, and validate the models against a comprehensive dataset covering hydrogen-related configurations and diffusion environments. Hydrogen diffusivity is evaluated over 650--1700 K, revealing enhanced mobility above 750 K in $a$-IGZO due to the glassy matrix, while diffusion at lower temperatures is constrained by the rigid network. Arrhenius extrapolation of the diffusivity indicates that hydrogen in $a$-IGZO can reach the channel/insulator interface within $10^{4}$ seconds at 300--400 K, likely contributing to negative bias stress-induced device degradation. Trajectory analysis reveals that long-range diffusion in $a$-IGZO is enabled by a combination of hydrogen hopping and flipping mechanisms. In CAAC-IGZO, hydrogen exhibits high in-plane diffusivity but severely restricted out-of-plane transport due to a high energy barrier along the $c$-axis. This limited vertical diffusion in CAAC-IGZO suggests minimal impact on bias instability. This work bridges the atomic-level hydrogen transport mechanism and device-level performance in oxide TFTs by leveraging large-scale MLIP-MD simulations.

cond-mat.mtrl-sci↗

arXiv2025

Hamilton cycles in pseudorandom graphs: resilience and approximate decompositions

Dirac's classical theorem asserts that, for $n \ge 3$, any $n$-vertex graph with minimum degree at least $n/2$ is Hamiltonian. Furthermore, if we additionally assume that such graphs are regular, then, by the breakthrough work of Csaba, Kühn, Lo, Osthus and Treglown, they admit a decomposition into Hamilton cycles and at most one perfect matching, solving the well-known Nash-Williams conjecture. In the pseudorandom setting, it has long been conjectured that similar results hold in much sparser graphs. We prove two overarching theorems for graphs that exclude excessively dense subgraphs, which yield asymptotically optimal resilience and Hamilton-decomposition results in sparse pseudorandom graphs. In particular, our results imply that for every fixed $γ> 0$, there exists a constant $C > 0$ such that if $G$ is a spanning subgraph of an $(n,d,λ)$-graph satisfying $δ(G) \ge (\tfrac12 + γ)d$ and $d/λ\ge C$, then $G$ must contain a Hamilton cycle. Secondly, we show that for every $\varepsilon > 0$, there is $C > 0$ so that every $(n,d,λ)$-graph with $d/λ\ge C$ contains at least $(\tfrac12 - \varepsilon)d$ edge-disjoint Hamilton cycles, and, finally, we prove that the entire edge set of $G$ can be covered by no more than $(\tfrac12 + \varepsilon)d$ such cycles. All bounds are asymptotically optimal and significantly improve earlier results on Hamiltonian resilience, packing, and covering in sparse pseudorandom graphs.

math.CO↗

arXiv2025

On a Ramsey--Turán variant of Roth's theorem

A classical theorem of Roth states that the maximum size of a solution-free set of a homogeneous linear equation $\mathcal{L}$ in $\mathbb{F}_p$ is $o(p)$ if and only if the sum of the coefficients of $\mathcal{L}$ is $0$. In this paper, we prove a Ramsey--Turán variant of Roth's theorem, with respect to a natural notion of ``structured'' sets introduced by Erdős and Sárközy in the 1970's. Namely, we show that the following statements are equivalent: $(a)$ Every solution-free set $A$ of $\mathcal{L}$ in $\mathbb{F}_p$ with $α(\mathrm{Cay}_{\mathbb{F}_p}(A)) = o(p)$ has size $o(p)$. $(b)$ There exists a non-empty \emph{subset} of coefficients of $\mathcal{L}$ with zero sum.

math.CO↗

arXiv2025

A characterization of testable hypergraph properties

We provide a combinatorial characterization of all testable properties of $k$-uniform hypergraphs ($k$-graphs for short). Here, a $k$-graph property $P$ is testable if there is a randomized algorithm which makes a bounded number of edge queries and distinguishes with probability $2/3$ between $k$-graphs that satisfy $P$ and those that are far from satisfying $P$. For the $2$-graph case, such a combinatorial characterization was obtained by Alon, Fischer, Newman and Shapira. Our results for the $k$-graph setting are in contrast to those of Austin and Tao, who showed that for the somewhat stronger concept of local repairability, the testability results for graphs do not extend to the $3$-graph setting. Our proof relies on a random subhypergraph sampling result proved in a companion paper.

math.CO↗

arXiv2025

On the order of intersecting hypergraphs

Determining the maximum number of edges in an intersecting hypergraph on a fixed ground set under additional constraints is one of the central topics in extremal combinatorics. In contrast, there are few results on analogous problems concerning the maximum order of such hypergraphs. In this paper, we systematically study these vertex analogues.

math.CO↗

arXiv2025

On the $(k+2,k)$-problem of Brown, Erdős and Sós for $k=5,6,7$

Let $f^{(r)}(n;s,k)$ denote the maximum number of edges in an $n$-vertex $r$-uniform hypergraph containing no subgraph with $k$ edges and at most $s$ vertices. Brown, Erdős and Sós [New directions in the theory of graphs (Proc. Third Ann Arbor Conf., Univ. Michigan 1971), pp. 53--63, Academic Press 1973] conjectured that the limit $\lim_{n\rightarrow \infty}n^{-2}f^{(3)}(n;k+2,k)$ exists for all $k$. The value of the limit was previously determined for $k=2$ in the original paper of Brown, Erdős and Sós, for $k=3$ by Glock [Bull. Lond. Math. Soc. 51 (2019) 230--236] and for $k=4$ by Glock, Joos, Kim, Kühn, Lichev and Pikhurko [Proc. Amer. Math. Soc., Series B, 11 (2024) 173-186] while Delcourt and Postle [Proc. Amer. Math. Soc., 152 (2024), 1881-1891] proved the conjecture (without determining the limiting value). In this paper, we determine the value of the limit in the Brown-Erdős-Sós Problem for $k\in \{5,6,7\}$. More generally, we obtain the value of $\lim_{n\rightarrow \infty}n^{-2}f^{(r)}(n;rk-2k+2,k)$ for all $r\geq 3$ and $k\in \{5,6,7\}$. In addition, by combining these new values with recent results of Bennett, Cushman and Dudek [arXiv:2309.00182] we obtain new asymptotic values for several generalised Ramsey numbers.

math.CO↗