absorb.md

Alfred Lin

Chronological feed of everything captured from Alfred Lin.

paper / alfred-lin / Jun 12

GaiaMoons: Enhancing Binary Asteroid Detection via Stellar Occultations and Gaia Astrometry

The GaiaMoons program leverages Gaia's high-precision astrometry to identify candidate binary asteroids, which are then studied using stellar occultations. This combined approach successfully characterized 101 targets, with four showing potential binary or contact binary features, thereby addressing biases in current detection methods and providing new constraints on asteroid system properties. The methodology establishes a self-improving cycle for discovering and characterizing binary asteroids.

solar-system-objectsasteroidsbinary-systemsstellar-occultationsastro-navigationgaia-mission
“Stellar occultation campaigns combined with Gaia observations can identify new binary asteroids and refine constraints on their size and shape.”
paper / alfred-lin / Jun 12

Multidimensional Symmetric Cooperative Motion and Porous Medium Equation Convergence

This paper extends the study of symmetric cooperative motion to higher dimensions by employing a novel approach that discretizes the porous medium equation. The authors overcome challenges related to finite difference schemes and unbounded initial data, ultimately establishing a new multidimensional convergence result for a scheme approximating the ZKB/Barenblatt solution. This contributes to the understanding of both symmetric cooperative motion and numerical methods for the porous medium equation.

mathematicsprobabilitystochastic-processesanalysisnumerical-methods
“The paper introduces a multidimensional version of symmetric cooperative motion, extending prior work from one dimension.”
paper / alfred-lin / Jun 12

MicroBooNE Reports First Measurement of Sub-GeV Neutrino-Argon Coherent Pion Production

The MicroBooNE experiment reports the first measurement of sub-GeV $\nu_\mu$ charged-current coherent pion production on argon. Utilizing a dataset corresponding to $1.26 \times 10^{21}$ POT with a mean neutrino energy of 0.8 GeV, the flux-averaged cross section was determined to be $(9.1 \pm 1.2_{\text{stat}} \pm 1.2_\text{syst}) \times 10^{-40}\,\text{cm}^2/\text{Ar}$. This measurement is significant as the clean kinematics of coherent pion production make it a valuable process for constraining neutrino flux uncertainties in oscillation experiments like DUNE.

charged-current-coherent-pion-productionneutrino-physicsmicroboone-experimentliquid-argon-tpcdune-oscillation-experimentcross-section-measurementhigh-energy-physics
“The MicroBooNE experiment has successfully measured the charged-current coherent pion production cross section on argon at sub-GeV neutrino energies.”
paper / alfred-lin / Jun 10

OpenPCC: An Open, TEE-Based Framework for Confidential LLM Serving

OpenPCC addresses the limitations of proprietary secure cloud compute solutions by offering an open framework for confidential LLM serving using commercially available Trusted Execution Environments (TEEs). This approach aims to provide robust privacy protection for sensitive user data processed by Large Language Models, overcoming the hardware restrictions and design flaws of existing systems. Evaluations on a Llama-3 8B workload confirm the feasibility and security of this open, TEE-based solution.

confidential-computingteellm-servingcloud-inferenceprivacy
“Existing secure cloud compute solutions like Apple PCC and Google Private AI Compute are not universally adoptable due to their reliance on proprietary hardware and closed ecosystems.”
paper / alfred-lin / Jun 10

P3D-Bench: Evaluating MLLMs in Parametric 3D Generation with a Focus on Structural and Geometric Precision

P3D-Bench is introduced as a novel benchmark for evaluating multimodal large language models (MLLMs) in parametric 3D generation. It addresses the limitations of existing benchmarks by assessing not only the appearance but also the structural and geometric precision of 3D models generated from code. The benchmark comprises three task families—Text-to-3D, Image-to-3D, and Assembly-3D—evaluating outputs on executability, fidelity, topology, semantic alignment, and structural consistency. Findings indicate that MLLMs excel at global shape and semantic recovery but struggle with precise parametric geometry and, notably, with part-level composition in complex assemblies.

multimodal-llm-benchmark3d-generationparametric-modelingcomputer-visionstructural-reasoningcode-generation
“Current multimodal large language models (MLLMs) struggle with parametric 3D generation, particularly in complex assembly tasks where composing multiple parts into a coherent structure is a significant challenge.”