arXiv subjects

Hongli Sun

Publications and source records attributed to Hongli Sun.

7 recordsLinked to original sources

arXiv2026

Application of the Shape Design Method to Hidden Regularity of Degenerate Hyperbolic Equations with Degenerate Boundary

This paper investigates the well-posedness and hidden regularity of boundary-degenerate hyperbolic equations in a two-dimensional setting. The shape design method is employed, which approximates the original degenerate problem by a family of uniformly elliptic problems on truncated subdomains and passes to the limit through uniform estimates. Within this framework, the existence and uniqueness of weak solutions are established in suitable weighted Sobolev spaces. A hidden regularity estimate for the conormal derivative on the nondegenerate portion of the boundary is obtained, providing a uniform bound in terms of the natural weighted energy norms of the initial data and source term. This estimate yields the boundary trace information essential for observability and controllability of degenerate hyperbolic systems.

math.AP↗

arXiv2024

MedOdyssey: A Medical Domain Benchmark for Long Context Evaluation Up to 200K Tokens

Numerous advanced Large Language Models (LLMs) now support context lengths up to 128K, and some extend to 200K. Some benchmarks in the generic domain have also followed up on evaluating long-context capabilities. In the medical domain, tasks are distinctive due to the unique contexts and need for domain expertise, necessitating further evaluation. However, despite the frequent presence of long texts in medical scenarios, evaluation benchmarks of long-context capabilities for LLMs in this field are still rare. In this paper, we propose MedOdyssey, the first medical long-context benchmark with seven length levels ranging from 4K to 200K tokens. MedOdyssey consists of two primary components: the medical-context "needles in a haystack" task and a series of tasks specific to medical applications, together comprising 10 datasets. The first component includes challenges such as counter-intuitive reasoning and novel (unknown) facts injection to mitigate knowledge leakage and data contamination of LLMs. The second component confronts the challenge of requiring professional medical expertise. Especially, we design the ``Maximum Identical Context'' principle to improve fairness by guaranteeing that different LLMs observe as many identical contexts as possible. Our experiment evaluates advanced proprietary and open-source LLMs tailored for processing long contexts and presents detailed performance analyses. This highlights that LLMs still face challenges and need for further research in this area. Our code and data are released in the repository: \url{https://github.com/JOHNNY-fans/MedOdyssey.}

cs.CL↗

arXiv2023

Null controllability of n-dimensional parabolic equations degenerated on partial boundary

This paper extends the Carleman estimates to high dimensional parabolic equations with highly degenerate symmetric coefficients on a bounded domain of Lipschitz boundary and use these estimates to study the controlla?bility the corresponding equations. Due to the nonsmoothness and degeneracy of boundary, the partial integration by parts in Carleman estimates have no meaning on the degenerate and nonsmooth parts of the boundary. To get around of this difficulty, we construct special weight function, and transform some integral terms in degenerate regions into a non-degenerate ones carefully so that the obtained Carleman estimates can still be used to the controllability problem. Our results includes some well-known works as some special cases as well as some interesting new examples.

math.AP↗

arXiv2023

Extremal properties of the first eigenvalue and the fundamental gap of a sub-elliptic operator

We consider the problems of extreming the first eigenvalue and the fundamental gap of a sub-elliptic operator with Dirichlet boundary condition, when the potential $V$ is subjected to a $p$-norm constraint. The existence results for weak solutions, compact embedding theorem and spectral theory for sub-elliptic equation are given. Moreover, we provide the specific characteristics of the corresponding optimal potential function.

math.AP↗

arXiv2022

The fundamental gap of a kind of two dimensional sub-elliptic operator

This paper is concerned at the minimization fundamental gap problem for a class of two-dimensional degenerate sub-elliptic operators. We establish existence results for weak solutions, Sobolev embedding theorem and spectral theory of sub-elliptic operators. We provide the existence and characterization theorems for extremizing potentials $V(x)$ when $V(x)$ is subject to $L^\infty$ norm constraint.

math.AP↗

arXiv2022

FL-Tuning: Layer Tuning for Feed-Forward Network in Transformer

Prompt tuning is an emerging way of adapting pre-trained language models to downstream tasks. However, the existing studies are mainly to add prompts to the input sequence. This way would not work as expected due to the intermediate multi-head self-attention and feed-forward network computation, making model optimization not very smooth. Hence, we propose a novel tuning way called layer tuning, aiming to add learnable parameters in Transformer layers. Specifically, we focus on layer tuning for feed-forward network in the Transformer, namely FL-tuning. It introduces additional units into the hidden layer of each feed-forward network. We conduct extensive experiments on the public CLUE benchmark. The results show that: 1) Our FL-tuning outperforms prompt tuning methods under both full-data and few-shot settings in almost all cases. In particular, it improves accuracy by 17.93% (full-data setting) on WSC 1.0 and F1 by 16.142% (few-shot setting) on CLUENER over P-tuning v2. 2) Our FL-tuning is more stable and converges about 1.17 times faster than P-tuning v2. 3) With only about 3% of Transformer's parameters to be trained, FL-tuning is comparable with fine-tuning on most datasets, and significantly outperforms fine-tuning (e.g., accuracy improved by 12.9% on WSC 1.1) on several datasets. The source codes are available at https://github.com/genggui001/FL-Tuning.

cs.CL↗

arXiv2022

Designer magnetic topological graphene nanoribbons

The interplay of magnetism and topology lies at the heart of condensed matter physics, which offers great opportunities to design intrinsic magnetic topological materials hosting a variety of exotic topological quantum states including the quantum anomalous Hall effect (QAHE), axion insulator state, and Majorana bound states. Extending this concept to one-dimension (1D) systems offers additional rich quantum spin physics with great promise for molecular-scale spintronics. Despite recent progress in the discovery of symmetry-protected topological quantum phases in 1D graphene nanoribbons (GNRs), the rational design and realization of magnetic topological GNRs (MT-GNRs) represents a grand challenge, as one must tackle multiple dimensions of complexity including time-reversal symmetry (TRS), spatial symmetry (width, edge, end geometry) and many-electron correlations. Here, we devised a new route involving the real- and reciprocal-space descriptions by unifying the chemists and physicists perspectives, for the design of such MT-GNRs with non-trivial electronic topology and robust magnetic terminal. Classic Clar's rule offers a conceptually qualitative real-space picture to predict the transition from closed-shell to open-shell with terminal magnetism, and band gap reopening with possible non-trivial electronic topology in a series of wave-like GNRs, which are further verified by first principle calculations of band-structure topology in a momentum-space. With the advance of on-surface synthesis and careful design of molecular precursors, we have fabricated these MT-GNRs with observation of topological edge bands, whose terminal pi-magnetism can be directly captured using a single-nickelocene spin sensor. Moreover, the transition from strong anti-ferromagnetic to weak coupling (paramagnetism-like) between terminal spins can be controlled by tuning the length of MT-GNRs.

cond-mat.mes-hall↗