arXiv subjects

Yiming Song

Publications and source records attributed to Yiming Song.

4 recordsLinked to original sources

arXiv2026

Carousel theorems for compact sets and homothets

We prove that if $A_0$ and $A_1$ are compact sets contained in a convex $n$-gon with vertices $G_1, \dots, G_n$, and $2 \lceil \frac{n}{2} \rceil$ is strictly greater than the number of common supporting lines of $A_0$ and $A_1$, then there exist $i\in \{0,1\}$ and $j\in \{1, \dots, n\}$ such that $A_i$ is contained in the convex hull of $A_{1-i}$ and $\{G_1,\dots,G_n\}\setminus\{G_j\}$. This generalizes and recovers results of Adaricheva-Bolat and Czédli-Kurusa concerning disks and homothetic sets. We construct examples to prove that the bound is sharp. We also construct a family of convex geometries not representable by positive homothets of any fixed planar convex body.

math.CO↗

arXiv2026

RouteMoA: Dynamic Routing without Pre-Inference Boosts Efficient Mixture-of-Agents

Mixture-of-Agents (MoA) improves LLM performance through layered collaboration, but its dense topology raises costs and latency. Existing methods employ LLM judges to filter responses, yet still require all models to perform inference before judging, failing to cut costs effectively. They also lack model selection criteria and struggle with large model pools, where full inference is costly and can exceed context limits. To address this, we propose RouteMoA, an efficient mixture-of-agents framework with dynamic routing. It employs a lightweight scorer to perform initial screening by predicting coarse-grained performance from the query, narrowing candidates to a high-potential subset without inference. A mixture of judges then refines these scores through lightweight self- and cross-assessment based on existing model outputs, providing posterior correction without additional inference. Finally, a model ranking mechanism selects models by balancing performance, cost, and latency. RouteMoA outperforms MoA across varying tasks and model pool sizes, reducing cost by 89.8% and latency by 63.6% in the large-scale model pool.

cs.AI↗

arXiv2025

Superlubricity of Borophene: Tribological Properties in Comparison to hBN

The tribological performance of 2D materials makes them good candidates toward a reduction of friction at the macroscale. Superlubricity has been observed for graphene, MoS\textsubscript{2} and MXenes and hexagonal boron nitride (hBN) is used to reduce or tune friction, but other materials are investigated as potential candidates for low-lubricity applications. Specifically, borophene is predicted to have ultra-low friction. Here, we experimentally investigate frictional properties of borophene and use a borophene-hBN lateral heterostructure to directly compare the tribological properties of the two complementary 2D materials. In particular, we investigate the friction between a sliding tip and (i) the weakly corrugated $\mathcal{X}_6$-borophene layer on Ir(111) or (ii) the hBN/Ir(111) superlattice structures with a strongly corrugated moiré reconstruction. Our experimental study performed in ultra-high vacuum at room temperature combined with a Prandtl-Tomlinson (PT) model calculation confirms the superlubricity predicted for borophene, while hBN, which exhibits a higher friction, is nevertheless confirmed as a low friction material. Ab initio calculations show that the lower friction of $\mathcal{X}_6$-borophene with respect to hBN can be rationalized by weaker tip/surface interactions. In addition, we assess structural and electrical properties of borophene and hBN by using scanning probe techniques and compare their dissipation under the oscillating tip to investigate the possible path of energy dissipation occurring during friction. Our study demonstrates the low frictional properties of borophene and the potential of lateral heterostructure investigations to directly compare the properties of these 2D materials.

cond-mat.mtrl-sci↗

arXiv2025

An In-Situ Spatial-Temporal Sequence Detector for Neuromorphic Vision Sensor Empowered by High Density Vertical NAND Storage

Neuromorphic vision sensors require efficient real-time pattern recognition, yet conventional architectures struggle with energy and latency constraints. Here, we present a novel in-situ spatiotemporal sequence detector that leverages vertical NAND storage to achieve massively parallel pattern detection. By encoding each cell with two single-transistor-based multi-level cell (MLC) memory elements, such as ferroelectric field-effect transistors (FeFETs), and mapping a pixel's temporal sequence onto consecutive word lines (WLs), we enable direct temporal pattern detection within NAND strings. Each NAND string serves as a dedicated reference for a single pixel, while different blocks store patterns for distinct pixels, allowing large-scale spatial-temporal pattern recognition via simple direct bit-line (BL) sensing, a well-established operation in vertical NAND storage. We experimentally validate our approach at both the cell and array levels, demonstrating that vertical NAND-based detector achieves more than six orders of magnitude improvement in energy efficiency and more than three orders of magnitude reduction in latency compared to conventional CPU-based methods. These findings establish vertical NAND storage as a scalable and energy-efficient solution for next-generation neuromorphic vision processing.

cs.ET↗