arXiv subjects

Jiaying Zhang

Publications and source records attributed to Jiaying Zhang.

5 recordsLinked to original sources

arXiv2026

Tunable Pathway Selection in Coupled Multistable Snap-Through Systems

Multistable mechanical systems can store and release elastic energy through snap-through instabilities, but controlling transition pathways between stable states remains challenging when multiple routes are accessible. Here, we introduce a two-mass von Mises truss as a general model for studying pathway selection governed by coupled saddle-node bifurcations. The system consists of two coupled snap-through units with geometric imperfections, giving rise to four stable configurations: a fully inverted state, a fully natural state, and two intermediate mixed states. We show that the coupling stiffness reorganizes the quasi-static bifurcation structure and selects among three transition pathways under release: sequential snapping through one mixed state, direct cooperative snapping, or sequential snapping through the other mixed state. Using pseudo-arclength continuation, we track the relevant saddle-node bifurcations and identify the parameter regimes associated with each quasi-static pathway. We then demonstrate that dynamic bifurcation delay provides an additional rate-dependent mechanism for pathway selection. Even when the quasi-static bifurcation structure favours a unique sequential pathway, finite-rate loading delays snap-through beyond the corresponding static saddle-node points and can reorder the snapping sequence of the two units. A local reduction of the coupled dynamics near each saddle-node yields normal forms with coupling-dependent critical points and coefficients. The resulting theory identifies distinct rate-dependent delay laws in the inertia-dominated and overdamped regimes and predicts the critical rate at which the snapping order reverses. These results establish a general mechanics framework for tuning transition pathways in multistable systems through elastic coupling and loading-rate control.

cond-mat.soft↗

arXiv2026

Time-periodic solutions of the Boltzmann equation with soft potentials in $\mathbb{R}^3$

We establish the existence of time-periodic solutions to the Boltzmann equation for the full range of soft potentials and prove their global stability. The proof relies on refined energy estimates combining Besov and Sobolev regularity with velocity-weighted energy methods. The analysis is complicated by the lack of a spectral gap in the soft potential regime together with the absence of time integrability of the periodic forcing over $\mathbb{R}^+$. This provides a framework for the study of time-periodic behavior for the Boltzmann equation.

math.AP↗

arXiv2026

GeoRA: Geometry-Aware Low-Rank Adaptation for RLVR

Reinforcement Learning with Verifiable Rewards (RLVR) is a key paradigm for improving large-scale reasoning models. Unlike supervised fine-tuning (SFT), RLVR exhibits distinct optimization dynamics and is sensitive to the preservation of pre-trained geometric structures. However, existing parameter-efficient methods face key limitations in this regime. Low-rank adaptation methods, such as PiSSA, are primarily designed for Supervised Fine-Tuning (SFT) and do not account for the distinct optimization dynamics and geometric structures of RLVR. Conversely, directly fine-tuning the unstructured sparse parameter subspace favored by RLVR encounters efficiency bottlenecks on modern hardware. To address these challenges, we propose GeoRA (Geometry-Aware Low-Rank Adaptation), a low-rank adaptation method tailored for RLVR. Specifically, GeoRA exploits the anisotropic and compressible structure of RL update subspace, and extracts its principal directions via Singular Value Decomposition (SVD) to initialize low-rank adapters, while freezing residual components as a structural anchor during training. This design preserves the pre-trained structure and enables efficient dense computation. Experiments on Qwen and Llama models from 1.5B to 32B parameters show that GeoRA consistently outperforms strong low-rank baselines across RLVR settings in mathematics, medicine, and coding, while showing stronger generalization and less forgetting on out-of-domain tasks.

cs.LG↗

arXiv2019

Enriching Medcial Terminology Knowledge Bases via Pre-trained Language Model and Graph Convolutional Network

Enriching existing medical terminology knowledge bases (KBs) is an important and never-ending work for clinical research because new terminology alias may be continually added and standard terminologies may be newly renamed. In this paper, we propose a novel automatic terminology enriching approach to supplement a set of terminologies to KBs. Specifically, terminology and entity characters are first fed into pre-trained language model to obtain semantic embedding. The pre-trained model is used again to initialize the terminology and entity representations, then they are further embedded through graph convolutional network to gain structure embedding. Afterwards, both semantic and structure embeddings are combined to measure the relevancy between the terminology and the entity. Finally, the optimal alignment is achieved based on the order of relevancy between the terminology and all the entities in the KB. Experimental results on clinical indicator terminology KB, collected from 38 top-class hospitals of Shanghai Hospital Development Center, show that our proposed approach outperforms baseline methods and can effectively enrich the KB.

cs.CL↗

arXiv2016

White matter deficits underlie the loss of consciousness level and predict recovery outcome in disorders of consciousness

This study aimed to identify white matter (WM) deficits underlying the loss of consciousness in disorder of consciousness (DOC) patients using Diffusion Tensor Imaging (DTI) and to demonstrate the potential value of DTI parameters in predicting recovery outcomes of DOC patients. With 30 DOC patients (8 comatose, 8 unresponsive wakefulness syndrome/vegetative state, and 14 minimal conscious state) and 25 patient controls, we performed group comparison of DTI parameters across 48 core WM regions of interest (ROIs) using Analysis of Covariance. Compared with controls, DOC patients had decreased Fractional anisotropy (FA) and increased diffusivities in widespread WM area.The corresponding DTI parameters of those WM deficits in DOC patients significantly correlated with the consciousness level evaluated by Coma Recovery Scale Revised (CRS-R) and Glasgow Coma Scale (GCS). As for predicting the recovery outcomes (i.e., regaining consciousness or not, grouped by their Glasgow Outcome Scale more than 2 or not) at 3 months post scan, radial diffusivity of left superior cerebellar peduncle and FA of right sagittal stratum reached an accuracy of 87.5% and 75% respectively. Our findings showed multiple WM deficits underlying the loss of consciousness level, and demonstrated the potential value of these WM areas in predicting the recovery outcomes of DOC patients who have lost awareness of the environment and themselves.

q-bio.NC↗