BZPEERPapers, people, evidence and research paths
Menu

Discover

SearchFind papers, people, topics, institutions and opportunities from one place.GraphSee how papers, authors, topics and evidence connect.

Research tools

HomeContinue from the papers, people and topics that matter now.CollectionsTurn saved papers into reading lists, evidence maps and shareable context.ResearchReconstruct the path that led you from search to paper to question.CollaborationMove from a paper, topic or expert request into a focused thread.PublishCreate full-text research articles with metadata, formulas and versions.

Network

ExpertsFind specialists whose work and reviews explain why they can help.CommunitiesJoin research groups organized around papers, topics and shared evidence.

Opportunities

ListingsBrowse roles, calls and collaborations connected to real research context.

Account

Log inReturn to your saved papers, graph views and active threads.Create accountStart saving papers, building a research trail and joining teams.
Log inCreate account

Source author record

Lina Chen

Lina Chen appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record
math.DGComputer Visioncond-mat.mtrl-scicond-mat.str-elArtificial IntelligenceComputation and Languagecond-mat.mes-halleess.IV

Catalog footprint

What is connected

10works
8topics
4close collaborators

Actions

Connect this record

Log in to claim
Open graphBrowse works

Research graph

See the researcher in context

Open full explorer

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

Published work

10 published item(s)

preprint2026arXiv

STEP3-VL-10B Technical Report

We present STEP3-VL-10B, a lightweight open-source foundation model designed to redefine the trade-off between compact efficiency and frontier-level multimodal intelligence. STEP3-VL-10B is realized through two strategic shifts: first, a unified, fully unfrozen pre-training strategy on 1.2T multimodal tokens that integrates a language-aligned Perception Encoder with a Qwen3-8B decoder to establish intrinsic vision-language synergy; and second, a scaled post-training pipeline featuring over 1k iterations of reinforcement learning. Crucially, we implement Parallel Coordinated Reasoning (PaCoRe) to scale test-time compute, allocating resources to scalable perceptual reasoning that explores and synthesizes diverse visual hypotheses. Consequently, despite its compact 10B footprint, STEP3-VL-10B rivals or surpasses models 10$\times$-20$\times$ larger (e.g., GLM-4.6V-106B, Qwen3-VL-235B) and top-tier proprietary flagships like Gemini 2.5 Pro and Seed-1.5-VL. Delivering best-in-class performance, it records 92.2% on MMBench and 80.11% on MMMU, while excelling in complex reasoning with 94.43% on AIME2025 and 75.95% on MathVision. We release the full model suite to provide the community with a powerful, efficient, and reproducible baseline.

0 citations0 reviewsNo code linkedOpen access
preprint2024arXiv

Can Large Language Models Understand Real-World Complex Instructions?

Large language models (LLMs) can understand human instructions, showing their potential for pragmatic applications beyond traditional NLP tasks. However, they still struggle with complex instructions, which can be either complex task descriptions that require multiple tasks and constraints, or complex input that contains long context, noise, heterogeneous information and multi-turn format. Due to these features, LLMs often ignore semantic constraints from task descriptions, generate incorrect formats, violate length or sample count constraints, and be unfaithful to the input text. Existing benchmarks are insufficient to assess LLMs' ability to understand complex instructions, as they are close-ended and simple. To bridge this gap, we propose CELLO, a benchmark for evaluating LLMs' ability to follow complex instructions systematically. We design eight features for complex instructions and construct a comprehensive evaluation dataset from real-world scenarios. We also establish four criteria and develop corresponding metrics, as current ones are inadequate, biased or too strict and coarse-grained. We compare the performance of representative Chinese-oriented and English-oriented models in following complex instructions through extensive experiments. Resources of CELLO are publicly available at https://github.com/Abbey4799/CELLO.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

Almost maximal volume entropy rigidity for integral Ricci curvature in the non-collapsing case

In this note we will show the almost maximal volume entropy rigidity for manifolds with lower integral Ricci curvature bound in the non-collapsing case: Given $n, d, p>\frac{n}{2}$, there exist $δ(n, d, p), ε(n, d, p)>0$, such that for $δ<δ(n, d, p)$, $ε<ε(n, d, p)$, if a compact $n$-manifold $M$ satisfies that the integral Ricci curvature has lower bound $\bar k(-1, p)\leq δ$, the diameter $diam(M)\leq d$ and volume entropy $h(M)\geq n-1-ε$, then the universal cover of $M$ is Gromov-Hausdorff close to a hyperbolic space form $\Bbb H^k$, $k\leq n$; If in addition the volume of $M$, $vol(M)\geq v>0$, then $M$ is diffeomorphic and Gromov-Hausdorff close to a hyperbolic manifold where $δ, ε$ also depends on $v$.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

Almost volume cone implies almost metric cone for annuluses centered at a compact set in $RCD(K, N)$-spaces

In \cite{CC1}, Cheeger-Colding considered manifolds with lower Ricci curvature bound and gave some almost rigidity results about warped products including almost metric cone rigidity and quantitative splitting theorem. As a generalization of manifolds with lower Ricci curvature bound, for metric measure spaces in $RCD(K, N)$, $1<N<\infty$, splitting theorem \cite{Gi13} and "volume cone implies metric cone" rigidity for balls and annuluses of a point \cite{PG} have been proved. In this paper we will generalize Cheeger-Colding's \cite{CC1} result about "almost volume cone implies almost metric cone for annuluses of a compact subset " to $RCD(K, N)$-spaces. More precisely, consider a $RCD(K, N)$-space $(X, d, \mathfrak m)$ and a Borel subset $Ω\subset X$. If the closed subset $S=\partial Ω$ has finite outer curvature, the diameter ${diam}(S)\leq D$ and the mean curvature of $S$ satisfies $$m(x)\leq m, \, \forall x\in S,$$ and \begin{equation*}\mathfrak m(A_{a, b}(S))\geq (1-ε)\int_a^b \left({sn}'_H(r)+ \frac{m}{n-1}{sn}_H(r)\right)^{n-1}dr \mathfrak m_S(S)\end{equation*} then $A_{a', b'}(S)$ is measured Gromov-Hausdorff close to a warped product $(a', b')\times_{{sn}'_H(r)+ \frac{m}{n-1}{sn}_H(r)}Y,$ $A_{a, b}(S)=\{x\in X\setminus Ω, \, a<d(x, S)<b\}$, $a<a'<b'<b$, $Y$ is a metric space with finite components with each component is a $RCD(0, N-1)$-space when $m=0, K=0$ and is a $RCD(N-2, N-1)$-space for other cases and $H=\frac{K}{N-1}$. Note that when $m=0, K=0$, our result is a kind of quantitative splitting theorem and in other cases it is an almost metric cone rigidity. To prove this result, different from \cite{Gi13, PG}, we will use \cite{GiT}'s second order differentiation formula and a method similar as \cite{CC1}.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

HistoKT: Cross Knowledge Transfer in Computational Pathology

The lack of well-annotated datasets in computational pathology (CPath) obstructs the application of deep learning techniques for classifying medical images. %Since pathologist time is expensive, dataset curation is intrinsically difficult. Many CPath workflows involve transferring learned knowledge between various image domains through transfer learning. Currently, most transfer learning research follows a model-centric approach, tuning network parameters to improve transfer results over few datasets. In this paper, we take a data-centric approach to the transfer learning problem and examine the existence of generalizable knowledge between histopathological datasets. First, we create a standardization workflow for aggregating existing histopathological data. We then measure inter-domain knowledge by training ResNet18 models across multiple histopathological datasets, and cross-transferring between them to determine the quantity and quality of innate shared knowledge. Additionally, we use weight distillation to share knowledge between models without additional training. We find that hard to learn, multi-class datasets benefit most from pretraining, and a two stage learning framework incorporating a large source domain such as ImageNet allows for better utilization of smaller datasets. Furthermore, we find that weight distillation enables models trained on purely histopathological features to outperform models using external natural image data.

0 citations0 reviewsNo code linkedOpen access
preprint2020arXiv

Maximizing spin-orbit torque efficiency of Ta(O)/Py via modulating oxygen-induced interface orbital hybridization

Spin-orbit torques due to interfacial Rashba and spin Hall effects have been widely considered as a potentially more efficient approach than the conventional spin-transfer torque to control the magnetization of ferromagnets. We report a comprehensive study of spin-orbit torque efficiency in Ta(O)/Ni81Fe19 bilayers by tuning low-oxidation of \b{eta}-phase tantalum, and find that the spin Hall angle θDL increases from ~ -0.18 of the pure Ta/Py to the maximum value ~ -0.30 of Ta(O)/Py with 7.8% oxidation. Furthermore, we distinguish the efficiency of the spin-orbit torque generated by the bulk spin Hall effect and by interfacial Rashba effect, respectively, via a series of Py/Cu(0-2 nm)/Ta(O) control experiments. The latter has more than twofold enhancement, and even more significant than that of the former at the optimum oxidation level. Our results indicate that 65% enhancement of the efficiency should be related to the modulation of the interfacial Rashba-like spin-orbit torque due to oxygen-induced orbital hybridization cross the interface. Our results suggest that the modulation of interfacial coupling via oxygen-induced orbital hybridization can be an alternative method to boost the change-spin conversion rate.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

A Remark on Regular Points of Ricci Limit Spaces

Let $Y$ be a Gromov-Hausdorff limit of complete Riemannian n-manifolds with Ricci curvature bounded from below. A point in $Y$ is called $k$-regular, if its tangent is unique and is isometric to an $k$-dimensional Euclidean space. By \cite{B5}, there is $k>0$ such that the set of all $k$-regular point $\mathcal{R}_k$ has a full renormalized measure. An open problem is if $\mathcal{R}_l=\emptyset$ for all $l<k$? The main result in this paper asserts that if $\mathcal{R}_1\ne \emptyset$, then $Y$ is a one dimensional topological manifold. Our result improves the Handa's result \cite{Honda} that under the assumption that $1\leq \mathrm{dim}_H(Y)<2$.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

Polar compensation at the surface of SrTiO3 (111)

We have systematically investigated the annealing effect on the structure and composition of the polar surface of SrTiO3 (111), starting with an ex-situ chemical etch. The relative surface concentration between Ti and Sr strongly depends on both the annealing temperature and the oxygen processing. There is a critical annealing temperature at which the maximum concentration ratio of Ti to Sr is achieved, while still maintaining a (1 x 1) surface structure. We demonstrate that with proper processing it is possible to avoid surface reconstruction over a broad temperature range. Our results provide an optimal temperature window for epitaxial film growth.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

Quantitative Volume Space From Rigidity with lower Ricci curvature bound II

This is the second paper of two in a series under the same title ([CRX]); both study the quantitative volume space form rigidity conjecture: a closed $n$-manifold of Ricci curvature at least $(n-1)H$, $H=\pm 1$ or $0$ is diffeomorphic to a $H$-space form if for every ball of definite size on $M$, the lifting ball on the Riemannian universal covering space of the ball achieves an almost maximal volume, provided the diameter of $M$ is bounded for $H\ne 1$. In [CRX], we verified the conjecture for the case that $M$ or its Riemannian universal covering space $\tilde M$ is not collapsed for $H=1$ or $H\ne 1$ respectively. In the present paper, we will verify this conjecture for the case that Ricci curvature is also bounded above, while the above non-collapsing condition is not required.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

Unusual Fe-H bonding associated with oxygen vacancies at the (001) surface of Fe3O4

An unusual Fe-H bonding rather than conventional OH bonding is identified at Fe3O4 (001) surface. This abnormal behavior is associated with the oxygen vacancies which exist on the surface region but also penetrate deep into the bulk Fe3O4. In contrast, OH bonding becomes preferential as generally expected on an ozone processed surface, which has appreciably less oxygen vacancies. Such bonding site selective behavior, depending on oxygen vacancy concentrations, is further confirmed with DFT calculations. The results demonstrate an opportunity for tuning the chemical properties of oxide surfaces or oxide clusters.

0 citations0 reviewsNo code linkedOpen access