arXiv subjects

Jiali Zhu

Publications and source records attributed to Jiali Zhu.

5 recordsLinked to original sources

arXiv2026

Exploring the Capability Boundaries of LLMs in Mastering of Chinese Chouxiang Language

While large language models (LLMs) have achieved remarkable success in general language tasks, their performance on Chouxiang Language, a representative subcultural language in the Chinese internet context, remains largely unexplored. In this paper, we introduce Mouse, a specialized benchmark designed to evaluate the capabilities of LLMs on NLP tasks involving Chouxiang Language across six tasks. Experimental results show that, current state-of-the-art (SOTA) LLMs exhibit clear limitations on multiple tasks, while performing well on tasks that involve contextual semantic understanding. In addition, we further discuss the reasons behind the generally low performance of SOTA LLMs on Chouxiang Language, examine whether the LLM-as-a-judge approach adopted for translation tasks aligns with human judgments and values, and analyze the key factors that influence Chouxiang translation. Our study aims to promote further research in the NLP community on multicultural integration and the dynamics of evolving internet languages. Our code and data are publicly available.

cs.CL↗

arXiv2024

MobileFlow: A Multimodal LLM For Mobile GUI Agent

Currently, the integration of mobile Graphical User Interfaces (GUIs) is ubiquitous in most people's daily lives. And the ongoing evolution of multimodal large-scale models, such as GPT-4v, Qwen-VL-Max, has significantly bolstered the capabilities of GUI comprehension and user action analysis, showcasing the potentiality of intelligent GUI assistants. However, current GUI Agents often need to access page layout information through calling system APIs, which may pose privacy risks. Fixing GUI (such as mobile interfaces) to a certain low resolution might result in the loss of fine-grained image details. At the same time, the multimodal large models built for GUI Agents currently have poor understanding and decision-making abilities for Chinese GUI interfaces, making them difficult to apply to a large number of Chinese apps. This paper introduces MobileFlow, a multimodal large language model meticulously crafted for mobile GUI agents. Transforming from the open-source model Qwen-VL-Chat into GUI domain, MobileFlow contains approximately 21 billion parameters and is equipped with novel hybrid visual encoders, making it possible for variable resolutions of image inputs and good support for multilingual GUI. By incorporating Mixture of Experts (MoE) expansions and pioneering alignment training strategies, MobileFlow has the capacity to fully interpret image data and comprehend user instructions for GUI interaction tasks. Finally, MobileFlow outperforms Qwen-VL-Max and GPT-4v in terms of task execution by GUI agents on both public and our proposed evaluation metrics, and has been successfully deployed in real-world business contexts, proving its effectiveness for practical applications.

cs.CV↗

arXiv2019

Arithmetic on self-similar sets

Let $K_1$ and $K_2$ be two one-dimensional homogeneous self-similar sets. Let $f$ be a continuous function defined on an open set $U\subset \mathbb{R}^{2}$. Denote the continuous image of $f$ by $$ f_{U}(K_1,K_2)=\{f(x,y):(x,y)\in (K_1\times K_2)\cap U\}. $$ In this paper we give an sufficient condition which guarantees that $f_{U}(K_1,K_2)$ contains some interiors. Our result is different from Simon and Taylor's \cite[Proposition 2.9]{ST} as we do not need the condition that the multiplication of the thickness of $K_1$ and $K_2$ is strictly greater than $1$. As a consequence, we give an application to the univoque sets in the setting of $q$-expansions.

math.DS↗

arXiv2019

Arithmetic on Moran sets

Let $(\mathcal{M}, c_k,n_k)$ be a class of Moran sets. We assume that the convex hull of any $E\in (\mathcal{M}, c_k,n_k)$ is $[0,1]$. Let $A,B$ be two non-empty sets in $\mathbb{R}$. Suppose that $f$ is a continuous function defined on an open set $U\subset \mathbb{R}^{2}$. Denote the continuous image of $f$ by \begin{equation*} f_{U}(A,B)=\{f(x,y):(x,y)\in (A\times B)\cap U\}. \end{equation*} In this paper, we prove the following result. Let $E_1,E_2\in(\mathcal{M}, c_k, n_k)$. If there exists some $(x_0,y_0)\in (E_1\times E_2)\cap U$ such that $$\sup_{k\geq 1}\left\{1-c_kn_k\right\}<\left\vert \frac{\partial _{y}f|_{(x_{0},y_{0})}}{\partial _{x}f|_{(x_{0},y_{0})}}\right\vert <\inf_{k\geq 1}\left\{\dfrac{c_k}{1-n_kc_k}\right\},$$ then $f_U(E_1, E_2)$ contains an interior.

math.DS↗

arXiv2019

Multiple representations of real numbers on self-similar sets with overlaps

Let $K$ be the attractor of the following IFS $$\{f_1(x)=λx, f_2(x)=λx +c-λ,f_3(x)=λx +1-λ\}, $$ where $f_1(I)\cap f_2(I)\neq \emptyset, (f_1(I)\cup f_2(I))\cap f_3(I)=\emptyset,$ and $I=[0,1]$ is the convex hull of $K$. The main results of this paper are as follows: $$\sqrt{K}+\sqrt{K}=[0,2]$$ if and only if $$\sqrt{c}+1\geq 2\sqrt{1-λ},$$ where $\sqrt{K}+\sqrt{K}=\{\sqrt{x}+\sqrt{y}:x,y\in K\}$. If $c\geq (1-λ)^2$, then $$\dfrac{K}{K}=\left\{\dfrac{x}{y}:x,y\in K, y\neq 0\right\}=\left[0,\infty\right).$$ As a consequence, we prove that the following conditions are equivalent: (1) For any $u\in [0,1]$, there are some $x,y\in K$ such that $u=x\cdot y;$ (2) For any $u\in [0,1]$, there are some $x_1,x_2,x_3,x_4,x_5,x_6,x_7,x_8, x_9,x_{10}\in K$ such that $$u=x_1+x_2=x_3-x_4=x_5\cdot x_6=x_7÷x_8=\sqrt{x_9}+\sqrt{x_{10}};$$ (3) $c\geq (1-λ)^2$.

math.DS↗