arXiv subjects

Jiannan Yang

Publications and source records attributed to Jiannan Yang.

15 recordsLinked to original sources

arXiv2026

Oracle-Budgeted Molecular Optimization with Short-Term Graph Memory

Molecular optimization is commonly performed under a limited oracle budget, which makes deciding what to evaluate as important as deciding what to generate. We introduce short-term graph memory, a plug-in module that preserves the generator architecture and native update rule while learning from previously evaluated molecules to prioritize subsequent oracle queries. The module maintains an online graph neural surrogate that pre-screens each round's candidate pool, so the fixed oracle budget is spent on molecules with higher predicted utility. Applied to a fragment-based generator on a standard molecular optimization benchmark, it improves the mean top-10 score at no extra oracle cost and never falls behind the base on any oracle; the gain extends to all four generators we tested at a tight budget of one thousand calls. We then analyze how surrogate-guided selection interacts with the exploration and exploitation behavior of different generators. Its benefit at larger budgets is consistent with two properties of the backbone: how broadly it searches, and how effectively its native search already exploits oracle feedback. We provide a simple way to spend a fixed oracle budget more selectively, and evidence on which generators benefit from it.

cs.LG↗

arXiv2026

Beyond Visibility and Technical Reuse: Public Application Transformation in Open-Source Model Ecosystems

Open-source model platforms have made it easier to publish AI models, but model release alone does not reveal whether models become visible, technically reused, or incorporated into public applications. This study introduces public application transformation as a platform-visible dimension of model impact and examines it through structured Model-Space links on Hugging Face. We construct a platform-scale dataset of 2.56 million model repositories, 1.06 million Spaces, 810,087 dataset repositories, and 1.22 million account profiles, together with Model-Space, Dataset-Space, and model-to-model technical reuse links. The analysis shows that public application transformation is highly selective and concentrated: only a small share of models are linked to Spaces, and most Model-Space links are concentrated among a limited set of models. More importantly, application transformation is associated with platform visibility but is not equivalent to technical reuse, indicating that downloads, likes, downstream model reuse, and application-facing uptake capture different forms of model impact. Additional analyses show that application-transformed models tend to exhibit stronger metadata-based readiness and enter heterogeneous Space configurations involving datasets, SDKs, and task-specific application categories. By tracing how models move from repositories into public applications and demos, this study extends the measurement of open-source model impact from artifact availability and technical reuse to platform-mediated transformation across AI information objects.

cs.SE↗

arXiv2026

Efficient Imputation for Patch-based Missing Single-cell Data via Cluster-regularized Optimal Transport

Missing data in single-cell sequencing datasets poses significant challenges for extracting meaningful biological insights. However, existing imputation approaches, which often assume uniformity and data completeness, struggle to address cases with large patches of missing data. In this paper, we present CROT (Cluster-Regularized Optimal Transport), an optimal transport-based imputation algorithm designed to handle patch-based missing data in tabular formats. Our approach effectively captures the underlying data structure in the presence of significant missingness. Notably, it achieves superior imputation accuracy while significantly reducing runtime, demonstrating its scalability and efficiency for large-scale datasets. This work introduces a robust solution for imputation in heterogeneous, high-dimensional datasets with structured data absence, addressing critical challenges in both biological and clinical data analysis. Our code is available on GitHub, https://github.com/yuyuliu11037/CROT.

cs.LG↗

arXiv2026

Resting Neurons, Active Insights: Robustifying Activation Sparsity in LLMs via Spontaneity

Activation sparsity offers a compelling route to accelerate large language model (LLM) inference by selectively suppressing hidden activations, yet existing approaches exhibit severe accuracy degradation at high sparsity. We show that this failure stems from representational instability: *activation sparsity disrupts input-dependent activation learned during pretraining, inducing distribution shifts in hidden states.* We address this issue by reframing activation sparsity as a representational alignment problem and introducing **Spontaneous Neurons (SPON)**, a lightweight mechanism inspired by spontaneous neural activity in biological systems. SPON injects a small set of learnable, input-independent activation vectors that act as persistent representational anchors for sparse computation. These vectors are trained via distribution matching to the dense model and can be absorbed into bias terms after training, incurring negligible inference overhead. Across multiple LLM backbones, SPON consistently restores performance, stabilizes latent representations, and preserves generalization. Our results establish SPON as an effective and principled solution for reliable activation-sparse inference, and offer new insights into knowledge retention in LLMs.

cs.LG↗

arXiv2026

Racing to Release: Priority, Congestion, and Community Recognition in Open-Source LLM Ecosystems

Open-source large language models have made platforms such as Hugging Face central hubs for decentralized AI innovation. Yet these ecosystems are shaped not only by collaboration, but also by competition for priority and community attention. Drawing on Hill and Stein's Race-to-the-Bottom framework, this study extends the logic of project potential, maturation, competition, and quality from scientific production to open-source LLM ecosystems, where prominent base models attract concentrated derivative entry under rapid and highly visible platform feedback. Using a large-scale sample of derivative models on Hugging Face, we find that later releases and more crowded competitive environments are both associated with weaker community recognition, even after accounting for differences in model and ecosystem prominence. These findings suggest that competition for priority remains an important organizing force in open-source LLM ecosystems, shaping which derivative innovations receive community recognition.

cs.SI↗

arXiv2026

Identifying Disruptive Models in the Open-Source LLM Community

The rapid growth of open-source large language models (LLMs) has created a complex ecosystem of model inheritance and reuse. However, existing research has focused mainly on descriptive analyses of lineage evolution, with limited attention to identifying which models play a disruptive role in shaping subsequent development. Using metadata from 2,556,240 models on Hugging Face, this study reconstructs a large-scale lineage network and introduces the Model Disruption Index (MDI) to distinguish between models that reinforce existing technological trajectories and those that become new bases for later development. The results show that most models in the open-source LLM community are consolidative rather than disruptive, reflecting a highly concentrated and path-dependent evolutionary structure. Further analyses suggest that disruptive positions are more likely to emerge among large-scale models and through finetuning strategies. Overall, this study provides a new perspective for identifying disruptive models and understanding uneven technological development in open-source LLM ecosystems.

cs.SI↗

arXiv2026

Physics-informed Gaussian Process Regression in Solving Eigenvalue Problem of Linear Operators

Applying Physics-Informed Gaussian Process Regression to the eigenvalue problem $(\mathcal{L}-λ)u = 0$ poses a fundamental challenge, where the null source term results in a trivial predictive mean and a degenerate marginal likelihood. Drawing inspiration from system identification, we construct a transfer function-type indicator for the unknown eigenvalue/eigenfunction using the physics-informed Gaussian Process posterior. We demonstrate that the posterior covariance is only non-trivial when $λ$ corresponds to an eigenvalue of the partial differential operator $\mathcal{L}$, reflecting the existence of a non-trivial eigenspace, and any sample from the posterior lies in the eigenspace of the linear operator. We demonstrate the effectiveness of the proposed approach through several numerical examples with both linear and non-linear eigenvalue problems.

stat.ML↗

arXiv2025

Self-Supervised Learning on Molecular Graphs: A Systematic Investigation of Masking Design

Self-supervised learning (SSL) plays a central role in molecular representation learning. Yet, many recent innovations in masking-based pretraining are introduced as heuristics and lack principled evaluation, obscuring which design choices are genuinely effective. This work cast the entire pretrain-finetune workflow into a unified probabilistic framework, enabling a transparent comparison and deeper understanding of masking strategies. Building on this formalism, we conduct a controlled study of three core design dimensions: masking distribution, prediction target, and encoder architecture, under rigorously controlled settings. We further employ information-theoretic measures to assess the informativeness of pretraining signals and connect them to empirically benchmarked downstream performance. Our findings reveal a surprising insight: sophisticated masking distributions offer no consistent benefit over uniform sampling for common node-level prediction tasks. Instead, the choice of prediction target and its synergy with the encoder architecture are far more critical. Specifically, shifting to semantically richer targets yields substantial downstream improvements, particularly when paired with expressive Graph Transformer encoders. These insights offer practical guidance for developing more effective SSL methods for molecular graphs.

cs.LG↗

arXiv2025

Derivative based global sensitivity analysis and its entropic link

Variance-based Sobol' sensitivity is one of the most well-known measures in global sensitivity analysis (GSA). However, uncertainties with certain distributions, such as highly skewed distributions or those with a heavy tail, cannot be adequately characterised using the second central moment only. Entropy-based GSA can consider the entire probability density function, but its application has been limited because it is difficult to estimate. Here we present a novel derivative-based upper bound for conditional entropies, to efficiently rank uncertain variables and to work as a proxy for entropy-based total effect indices. To overcome the non-desirable issue of negativity for differential entropies as sensitivity indices, we discuss an exponentiation of the total effect entropy and its proxy. Numerical verifications demonstrate that the upper bound is tight for monotonic functions and it provides the same input variable ranking as the entropy-based indices for about three-quarters of the 1000 random functions tested. We found that the new entropy proxy performs similarly to the variance-based proxies for a river flood physics model with 8 inputs of different distributions, and these two proxies are equivalent in the special case of linear functions with Gaussian inputs. We expect the new entropy proxy to increase the variable screening power of derivative-based GSA and to complement Sobol'-indices proxy for a more diverse type of distributions.

math.NA↗

arXiv2024

Predicting the Risk of Ischemic Stroke in Patients with Atrial Fibrillation using Heterogeneous Drug-protein-disease Network-based Deep Learning

We develop a deep learning model, ABioSPATH, to predict the one-year risk of ischemic stroke (IS) in atrial fibrillation (AF) patients. The model integrates drug-protein-disease pathways and real-world clinical data of AF patients to generate the IS risk and potential pathways for each patient. The model uses a multilayer network to identify the mechanism of drug action and disease comorbidity propagation pathways. The model is tested on the Electronic Health Record (EHR) data of 7859 AF patients from 43 hospitals in Hong Kong. The model outperforms all baselines across all metrics and provides valuable molecular-level insights for clinical use. The model also highlights key proteins in common pathways and potential IS risks tied to less-studied drugs. The model only requires routinely collected data, without requiring expensive biomarkers to be tested.

q-bio.QM↗

arXiv2023

Reciprocal hydrodynamic response estimation in a random spreading sea

Direct estimation of the hydrodynamic response of an offshore structure in a random spreading sea can lead to large computational costs. In this paper the actual spreading sea is replaced by an idealised diffuse wave field and the diffuse field reciprocity (DFR) relationship is derived analytically and verified against diffraction analysis for offshore application. The DFR approach provides an analytical expression for the estimation of the wave loading spectrum in a spreading sea. It is very efficient because only the added damping coefficients are required. Furthermore, if normalised to the peak amplitude of a spreading sea, an upper bound response can be obtained using the reciprocal approach. And this is demonstrated using a spar type floating wind turbine. Given that the hydrodynamic coefficients are routine outputs for offshore structural design, engineers would obtain the upper bound response without additional computational cost using this new approach.

physics.flu-dyn↗

arXiv2023

A general framework for probabilistic sensitivity analysis with respect to distribution parameters

Probabilistic sensitivity analysis identifies the influential uncertain input to guide decision-making. We propose a general sensitivity framework with respect to the input distribution parameters that unifies a wide range of sensitivity measures, including information theoretical metrics such as the Fisher information. The framework is derived analytically via a constrained maximization and the sensitivity analysis is reformulated into an eigenvalue problem. There are only two main steps to implement the sensitivity framework utilising the likelihood ratio/score function method, a Monte Carlo type sampling followed by solving an eigenvalue equation. The resulting eigenvectors then provide the directions for simultaneous variations of the input parameters and guide the focus to perturb uncertainty the most. Not only is it conceptually simple, but numerical examples demonstrate that the proposed framework also provides new sensitivity insights, such as the combined sensitivity of multiple correlated uncertainty metrics, robust sensitivity analysis with an entropic constraint, and approximation of deterministic sensitivities. Three different examples, ranging from a simple cantilever beam to an offshore marine riser, are used to demonstrate the potential applications of the proposed sensitivity framework to applied mechanics problems.

stat.ME↗

arXiv2022

Decision-oriented two-parameter Fisher information sensitivity using symplectic decomposition

The eigenvalues and eigenvectors of the Fisher information matrix (FIM) can reveal the most and least sensitive directions of a system and it has wide application across science and engineering. We present a symplectic variant of the eigenvalue decomposition for the FIM and extract the sensitivity information with respect to two-parameter conjugate pairs. The symplectic approach decomposes the FIM onto an even-dimensional symplectic basis. This symplectic structure can reveal additional sensitivity information between two-parameter pairs, otherwise concealed in the orthogonal basis from the standard eigenvalue decomposition. The proposed sensitivity approach can be applied to naturally paired two-parameter distribution parameters, or decision-oriented pairing via re-grouping or re-parameterization of the FIM. It can be utilised in tandem with the standard eigenvalue decomposition and offer additional insight into the sensitivity analysis at negligible extra cost.

cs.IT↗

arXiv2022

An information upper bound for probability sensitivity

Uncertain input of a mathematical model induces uncertainties in the output and probabilistic sensitivity analysis identifies the influential inputs to guide decision-making. Of practical concern is the probability that the output would, or would not, exceed a threshold, and the probability sensitivity depends on this threshold which is often uncertain. The Fisher information and the Kullback-Leibler divergence have been recently proposed in the literature as threshold-independent sensitivity metrics. We present mathematical proof that the information-theoretical metrics provide an upper bound for the probability sensitivity. The proof is elementary, relying only on a special version of the Cauchy-Schwarz inequality called Titu's lemma. Despite various inequalities exist for probabilities, little is known of probability sensitivity bounds and the one proposed here is new to the present authors' knowledge. The probability sensitivity bound is extended, analytically and with numerical examples, to the Fisher information of both the input and output. It thus provides a solid mathematical basis for decision-making based on probabilistic sensitivity metrics.

cs.IT↗

arXiv2015

Making problems tractable on big data via preprocessing with polylog-size output

To provide a dichotomy between those queries that can be made feasible on big data after appropriate preprocessing and those for which preprocessing does not help, Fan et al. developed the $\sqcap$-tractability theory. This theory provides a formal foundation for understanding the tractability of query classes in the context of big data. Along this line, we introduce a novel notion of $\sqcap'$-tractability in this paper. Inspired by some technologies used to deal big data, we place a restriction on preprocessing function, which limits the function to produce a relatively small database as output, at most polylog-size of the input database. At the same time, we bound the redundancy information when re-factorizing data and queries for preprocessing. These changes aim to make our theory more closely linked to practice. We set two complexity classes to denote the classes of Boolean queries that are $\sqcap'$-tractable themselves and that can be made $\sqcap'$-tractable, respectively. Based on a new factorization in our complexity classes, we investigate two reductions, which differ from whether allowing re-factorizing data and query parts. We verify the transitive and compatible properties of the reductions and analysis the complete problems and sizes of the complexity classes. We conclude that all PTIME classes of Boolean queries can be made $\sqcap'$-tractable, similar to that of the $\sqcap$-tractability theory. With a little surprise, we prove that the set of all $\sqcap'$-tractable queries is strictly smaller than that of all $\sqcap$-tractable queries, and thus the set of $\sqcap'$-tractable queries is properly contained in that of PTIME queries. In this way, we attain a new complexity class inside the complexity class of PTIME queries.

cs.CC↗