SearcharxivSearch

arXiv subjects

Wei-Hsiang Lin

Publications and source records attributed to Wei-Hsiang Lin.

5 recordsLinked to original sources

Do Before You Judge: Self-Reference as a Pathway to Better LLM Evaluation

LLM-as-Judge frameworks are increasingly popular for AI evaluation, yet research findings on the relationship between models' generation and judgment abilities remain inconsistent. We investigate this relationship through systematic dataset- and instance-level analyses across 11 models and 21 diverse tasks. Despite both capabilities relying on the same underlying knowledge, our analyses reveal they are only weakly correlated, primarily due to LLMs' sensitivity to the responses being judged. To address this, we propose a self-reference-guided evaluation strategy that leverages a model's own answers as references. This approach significantly strengthens the correlation between generation and judgment abilities, offering a practical path to align these skills and providing a reliable proxy for model selection in evaluation tasks.

cs.CL

Biomass transfer on autocatalytic reaction network: a delay differential equation formulation

For a biological system to grow and expand, mass must be transferred from the environment to the system and be assimilated into its reaction network. Here, I characterize the biomass transfer process for growing autocatalytic systems. By track biomass along reaction pathways, an n-dimensional ordinary differential equation (ODE) of the reaction network can be reformulated into a one-dimensional delay differential equation (DDE) for its long-term dynamics. The kernel function of the DDE summarizes the overall amplification and transfer delay of the system and serves as a signature for autocatalysis dynamics. The DDE formulation allows reaction networks of various topologies and complexities to be compared and provides rigorous estimation scheme for growth rate upon dimensional reduction of reaction networks.

q-bio.MN

Direct growth of mm-size twisted bilayer graphene by plasma-enhanced chemical vapor deposition

Plasma enhanced chemical vapor deposition (PECVD) techniques have been shown to be an efficient method to achieve single-step synthesis of high-quality monolayer graphene (MLG) without the need of active heating. Here we report PECVD-growth of single-crystalline hexagonal bilayer graphene (BLG) flakes and mm-size BLG films with the interlayer twist angle controlled by the growth parameters. The twist angle has been determined by three experimental approaches, including direct measurement of the relative orientation of crystalline edges between two stacked monolayers by scanning electron microscopy, analysis of the twist angle-dependent Raman spectral characteristics, and measurement of the Moiré period with scanning tunneling microscopy. In mm-sized twisted BLG (tBLG) films, the average twist angle can be controlled from 0 to approximately 20 \degree, and the angular spread for a given growth condition can be limited to < 7 \degree. Different work functions between MLG and BLG have been verified by the Kelvin probe force microscopy and ultraviolet photoelectron spectroscopy. Electrical measurements of back-gated field-effect-transistor devices based on small-angle tBLG samples revealed high-quality electric characteristics at 300 K and insulating temperature dependence down to 100 K. This controlled PECVD-growth of tBLG thus provides an efficient approach to investigate the effect of varying Moiré potentials on tBLG.

cond-mat.mtrl-sci

Atomic-scale Structural and Chemical Characterization of Hexagonal Boron Nitride Layers Synthesized at the Wafer-Scale with Monolayer Thickness Control

Hexagonal boron nitride (h-BN) is a promising two-dimensional insulator with a large band gap and low density of charged impurities that is isostructural and isoelectronic with graphene. Here we report the chemical and atomic-scale structure of CVD-grown wafer-scale (~25 cm2) h-BN sheets ranging in thickness from 1-20 monolayers. Atomic-scale images of h-BN on Au and graphene/Au substrates obtained by scanning tunneling microscopy (STM) reveal high h-BN crystalline quality in monolayer samples. Further characterization of 1-20 monolayer samples indicates uniform thickness for wafer-scale areas; this thickness control is a result of precise control of the precursor flow rate, deposition temperature and pressure. Raman and infrared spectroscopy indicate the presence of B-N bonds and reveal a linear dependence of thickness with growth time. X-ray photoelectron spectroscopy (XPS) shows the film stoichiometry, and the B/N atom ratio in our films is 1 + 0.6% across the range of thicknesses. Electrical current transport in metal/insulator/metal (Au/h-BN/Au) heterostructures indicates that our CVD-grown h-BN films can act as excellent tunnel barriers with a high hard-breakdown field strength. Our results suggest that large-area h-BN films are structurally, chemically and electronically uniform over the wafer scale, opening the door to pervasive application as a dielectric in layered nanoelectronic and nanophotonic heterostructures.

physics.app-ph

Field Effect Optoelectronic Modulation of Quantum-Confined Carriers in Black Phosphorus

We report measurements of the infrared optical response of thin black phosphorus under field-effect modulation. We interpret the observed spectral changes as a combination of an ambipolar Burstein-Moss (BM) shift of the absorption edge due to band-filling under gate control, and a quantum confined Franz-Keldysh (QCFK) effect, phenomena which have been proposed theoretically to occur for black phosphorus under an applied electric field. Distinct optical responses are observed depending on the flake thickness and starting carrier concentration. Transmission extinction modulation amplitudes of more than two percent are observed, suggesting the potential for use of black phosphorus as an active material in mid-infrared optoelectronic modulator applications.

cond-mat.mes-hall