arXiv subjects

David Xu

Publications and source records attributed to David Xu.

6 recordsLinked to original sources

arXiv2026

PersonaTrace: Synthesizing Realistic Digital Footprints with LLM Agents

Digital footprints (records of individuals' interactions with digital systems) are essential for studying behavior, developing personalized applications, and training machine learning models. However, research in this area is often hindered by the scarcity of diverse and accessible data. To address this limitation, we propose a novel method for synthesizing realistic digital footprints using large language model (LLM) agents. Starting from a structured user profile, our approach generates diverse and plausible sequences of user events, ultimately producing corresponding digital artifacts such as emails, messages, calendar entries, reminders, etc. Intrinsic evaluation results demonstrate that the generated dataset is more diverse and realistic than existing baselines. Moreover, models fine-tuned on our synthetic data outperform those trained on other synthetic datasets when evaluated on real-world out-of-distribution tasks.

cs.CL↗

arXiv2025

An explicit exotic representation of a rank-one simple Lie group via convex bodies

In [DP12], Delzant and Py showed that there exist continuous irreducible isometric actions of $\mathrm{PSL}_2(\mathbb{R})$ on the infinite-dimensional hyperbolic space $\mathbb{H}^\infty$. Such continuous irreducible actions do not exist on the hyperbolic spaces $\mathbb{H}^n$ when $n>2$ and their associated embeddings $\mathbb{H}^2 \to \mathbb{H}^\infty$ given by the orbit maps were later called \emph{exotic} by Monod and Py in [MP14]. In this article, we produce a continuous and irreducible representation of $\mathrm{PSL}_2(\mathbb{R})\to \mathrm{Isom}(\mathbb{H}^\infty)$ using the hyperbolic model for convex bodies introduced in [DF22]. This yields a convex cocompact $\mathrm{PSL}_2(\mathbb{R})$-action on the infinite-dimensional hyperbolic space $\mathbb{H}^\infty$, of which the compact quotient over the minimal $\mathrm{PSL}_2(\mathbb{R})$-invariant convex set is homeomorphic to the 2-dimensional oriented Banach--Mazur compactum. Moreover, we study the geometry of one of its orbit maps and compute the Hausdorff dimension of the limit set of this representation.

math.GR↗

arXiv2025

Role of prompt cusps in driving the core collapse of SIDM halos

Prompt cusps (PCs) form from the direct collapse of overdensities in the early Universe, reside at the center of every dark matter halo, and have density profiles steeper than $r^{-1}$ NFW cusps. Using a suite of high-resolution N-body simulations, we study the evolution of isolated halos in self-interacting dark matter (SIDM) with massive PCs embedded at their centers, a scenario that could be realized in certain SIDM models with light mediators that predict a small-scale suppression of the linear matter power spectrum. We track the evolution of three equally concentrated $10^7\,{\rm{M}}_\odot$ halos, hosting PCs of various total masses, and quantify how the presence of a PC affects the processes of core formation and collapse. Early in the core-formation phase, halos with more prominent PCs exhibit a delayed evolution by a factor of $\sim 2$ due to smaller velocity dispersion gradients in the inner region. During most of the core-collapse phase, the halo evolution becomes closely aligned in physical time, with appropriate rescaling of densities, radii, and velocity dispersions. The scale densities and radii preserve the virial mass of the original halos, but with increased concentration. Deviations occur at the late phase of core-collapse at the level of $\sim 5\%$ relative to the reference collapse track of an NFW halo. These deviations depend non-trivially on both the increased concentration incurred by the PCs, as well as the velocity dispersion (temperature) of the outer halo regions, which can inhibit or enhance the heat transfer process. Our simulations illustrate the complex interplay between the inner and outer halo profiles in determining the onset of core collapse and motivate future studies in the full cosmological context.

astro-ph.GA↗

arXiv2024

AudioSetMix: Enhancing Audio-Language Datasets with LLM-Assisted Augmentations

Multi-modal learning in the audio-language domain has seen significant advancements in recent years. However, audio-language learning faces challenges due to limited and lower-quality data compared to image-language tasks. Existing audio-language datasets are notably smaller, and manual labeling is hindered by the need to listen to entire audio clips for accurate labeling. Our method systematically generates audio-caption pairs by augmenting audio clips with natural language labels and corresponding audio signal processing operations. Leveraging a Large Language Model, we generate descriptions of augmented audio clips with a prompt template. This scalable method produces AudioSetMix, a high-quality training dataset for text-and-audio related models. Integration of our dataset improves models performance on benchmarks by providing diversified and better-aligned examples. Notably, our dataset addresses the absence of modifiers (adjectives and adverbs) in existing datasets. By enabling models to learn these concepts, and generating hard negative examples during training, we achieve state-of-the-art performance on multiple benchmarks.

eess.AS↗

arXiv2024

Convex-cocompact representations into the isometry group of the infinite-dimensional hyperbolic space

We prove that convex-cocompact representations of finitely generated groups in the group of isometries of the infinite-dimensional hyperbolic space form an open set in the space of representations, allowing us to deform these convex-cocompact representations. We then use bending to obtain convex-cocompact representations of surface groups that are not conjugate to any exotic representation of PSL(2,R) classified by Monod and Py.

math.GT↗

arXiv2022

Neural network accelerator for quantum control

Efficient quantum control is necessary for practical quantum computing implementations with current technologies. Conventional algorithms for determining optimal control parameters are computationally expensive, largely excluding them from use outside of the simulation. Existing hardware solutions structured as lookup tables are imprecise and costly. By designing a machine learning model to approximate the results of traditional tools, a more efficient method can be produced. Such a model can then be synthesized into a hardware accelerator for use in quantum systems. In this study, we demonstrate a machine learning algorithm for predicting optimal pulse parameters. This algorithm is lightweight enough to fit on a low-resource FPGA and perform inference with a latency of 175 ns and pipeline interval of 5 ns with $~>~$0.99 gate fidelity. In the long term, such an accelerator could be used near quantum computing hardware where traditional computers cannot operate, enabling quantum control at a reasonable cost at low latencies without incurring large data bandwidths outside of the cryogenic environment.

quant-ph↗