arXiv subjects

Jaehoon Kim

Publications and source records attributed to Jaehoon Kim.

At least 19 recordsLinked to original sources

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

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 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 T\'oth [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 $\Omega(kn\log(n/k+1))$ edges are required.

math.CO↗

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 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

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

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

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, K\"uhn, 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 $\gamma > 0$, there exists a constant $C > 0$ such that if $G$ is a spanning subgraph of an $(n,d,\lambda)$-graph satisfying $\delta(G) \ge (\tfrac12 + \gamma)d$ and $d/\lambda \ge C$, then $G$ must contain a Hamilton cycle. Secondly, we show that for every $\varepsilon > 0$, there is $C > 0$ so that every $(n,d,\lambda)$-graph with $d/\lambda \ge C$ 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--Tur\'{a}n 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--Tur\'{a}n variant of Roth's theorem, with respect to a natural notion of ``structured'' sets introduced by Erd\H{o}s and S\'ark\"ozy 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 $\alpha(\mathrm{Cay}_{\mathbb{F}_p}(A)) = o(p)$ has size $o(p)$. $(b)$ There exists a non-empty \emph{subset} of coefficients of $\mathcal{L}$ with zero sum.

math.CO↗

arXiv2025

Stability with minuscule structure for chromatic thresholds

The chromatic threshold $\delta_\chi(H)$ 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{\"o}ttcher, Griffiths, Kohayakawa, and Morris determined the chromatic threshold for every $H$; in particular, they showed that if $\chi(H)=r\ge 3$, then $\delta_\chi(H) \in\{\frac{r-3}{r-2},~\frac{2 r-5}{2 r-3},~\frac{r-2}{r-1}\}$. 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 $\delta(G)\ge (\delta_\chi(H)-o(1))n$ and $\chi(G)=\omega(1)$ 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 {\L}uczak and Thomass{\'e} on the triangle case, we determine the bounded-VC chromatic thresholds for all cliques.

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↗

arXiv2024

Ramsey--Dirac theory for bounded degree hypertrees

Ramsey--Tur\'an theory considers Tur\'an type questions in Ramsey-context, asking for the existence of a small subgraph in a graph $G$ where the complement $\overline{G}$ lacks an appropriate subgraph $F$, such as a clique of linear size. Similarly, one can consider Dirac-type questions in Ramsey context, asking for the existence of a spanning subgraph $H$ in a graph $G$ where the complement $\overline{G}$ lacks an appropriate subgraph $F$, which we call a Ramsey--Dirac theory question. When $H$ is a connected spanning subgraph, the disjoint union $K_{n/2}\cup K_{n/2}$ of two large cliques shows that it is natural to consider complete bipartite graphs $F$. Indeed, Han, Hu, Ping, Wang, Wang and Yang in 2024 proved that if $G$ is an $n$-vertex graph with $\delta(G)=\Omega(n)$ where the complement $\overline{G}$ does not contain any complete bipartite graph $K_{m,m}$ with $m=\Omega(n)$, then $G$ contains every $n$-vertex bounded degree tree $T$ as a subgraph. Extending this result to the Ramsey--Dirac theory for hypertrees, we prove that if $G$ is an $n$-vertex $r$-uniform hypergraph with $\delta(G)=\Omega(n^{r-1})$ where the complement $\overline{G}$ does not contain any complete $r$-partite hypergraph $K_{m,m,\dots, m}$ with $m=\Omega(n)$, then $G$ contains every $n$-vertex bounded degree hypertree $T$ as a subgraph. We also prove the existence of matchings and loose Hamilton cycles in the same setting, which extends the result of Mcdiarmid and Yolov into hypergraphs. This result generalizes the universality result on randomly perturbed graphs by B\"ottcher, Han, Kohayakawa, Montgomery, Parczyk and Person in 2019 into hypergraphs and also strengthen the results on quasirandom hypergraphs by Lenz, Mubayi and Mycroft in 2016 and Lenz and Mubayi in 2016 into hypergraphs satisfying a much weaker pseudorandomness condition.

math.CO↗

arXiv2024

Bayesian optimal change point detection in high-dimensions

We propose the first Bayesian methods for detecting change points in high-dimensional mean and covariance structures. These methods are constructed using pairwise Bayes factors, leveraging modularization to identify significant changes in individual components efficiently. We establish that the proposed methods consistently detect and estimate change points under much milder conditions than existing approaches in the literature. Additionally, we demonstrate that their localization rates are nearly optimal in terms of rates. The practical performance of the proposed methods is evaluated through extensive simulation studies, where they are compared to state-of-the-art techniques. The results show comparable or superior performance across most scenarios. Notably, the methods effectively detect change points whenever signals of sufficient magnitude are present, irrespective of the number of signals. Finally, we apply the proposed methods to genetic and financial datasets, illustrating their practical utility in real-world applications.

stat.ME↗

arXiv2024

Data-efficient multi-fidelity training for high-fidelity machine learning interatomic potentials

Machine learning interatomic potentials (MLIPs) are used to estimate potential energy surfaces (PES) from ab initio calculations, providing near quantum-level accuracy with reduced computational costs. However, the high cost of assembling high-fidelity databases hampers the application of MLIPs to systems that require high chemical accuracy. Utilizing an equivariant graph neural network, we present an MLIP framework that trains on multi-fidelity databases simultaneously. This approach enables the accurate learning of high-fidelity PES with minimal high-fidelity data. We test this framework on the Li$_6$PS$_5$Cl and In$_x$Ga$_{1-x}$N systems. The computational results indicate that geometric and compositional spaces not covered by the high-fidelity meta-gradient generalized approximation (meta-GGA) database can be effectively inferred from low-fidelity GGA data, thus enhancing accuracy and molecular dynamics stability. We also develop a general-purpose MLIP that utilizes both GGA and meta-GGA data from the Materials Project, significantly enhancing MLIP performance for high-accuracy tasks such as predicting energies above hull for crystals in general. Furthermore, we demonstrate that the present multi-fidelity learning is more effective than transfer learning or $\Delta$-learning an d that it can also be applied to learn higher-fidelity up to the coupled-cluster level. We believe this methodology holds promise for creating highly accurate bespoke or universal MLIPs by effectively expanding the high-fidelity dataset.

cond-mat.mtrl-sci↗

arXiv2024

Transversal cycles and paths in tournaments

Thomason [$\textit{Trans. Amer. Math. Soc.}$ 296.1 (1986)] proved that every sufficiently large tournament contains Hamilton paths and cycles with all possible orientations, except possibly the consistently oriented Hamilton cycle. This paper establishes $\textit{transversal}$ generalizations of these classical results. For a collection $\mathbf{T}=\{T_1,\dots,T_m\}$ of not-necessarily distinct tournaments on the common vertex set $V$, an $m$-edge directed subgraph $\mathcal{D}$ with the vertices in $V$ is called a transversal if there exists an bijection $\varphi\colon E(\mathcal{D})\to [m]$ such that $e\in E(T_{\varphi(e)})$ for all $e\in E(\mathcal{D})$. We prove that for sufficiently large $n$, there exist transversal Hamilton cycles of all possible orientations possibly except the consistently oriented one. We also obtain a similar result for the transversal Hamilton paths of all possible orientations. These results generalize the classical theorem of Thomason, and our approach provides another proof of this theorem.

math.CO↗