Author name cluster

Junyan Wang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

5 papers

1 author row

NeurIPS Conference 2025 Conference Paper

FlowerTune: A Cross-Domain Benchmark for Federated Fine-Tuning of Large Language Models

Yan Gao
Massimo R. Scamarcia
Javier Fernandez-Marques
Mohammad Naseri
Chong Ng
Dimitris Stripelis
Zexi Li
Tao Shen

Large Language Models (LLMs) have achieved state-of-the-art results across diverse domains, yet their development remains reliant on vast amounts of publicly available data, raising concerns about data scarcity and the lack of access to domain-specific, sensitive information. Federated Learning (FL) presents a compelling framework to address these challenges by enabling decentralized fine-tuning on pre-trained LLMs without sharing raw data. However, the compatibility and performance of pre-trained LLMs in FL settings remain largely under explored. We introduce the FlowerTune LLM Leaderboard, a first-of-its-kind benchmarking suite designed to evaluate federated fine-tuning of LLMs across four diverse domains: general NLP, finance, medical, and coding. Each domain includes federated instruction-tuning datasets and domain-specific evaluation metrics. Our results, obtained through a collaborative, open-source and community-driven approach, provide the first comprehensive comparison across 26 pre-trained LLMs with different aggregation and fine-tuning strategies under federated settings, offering actionable insights into model performance, resource constraints, and domain adaptation. This work lays the foundation for developing privacy-preserving, domain-specialized LLMs for real-world applications.

PDF Details

TMLR Journal 2025 Journal Article

Frame-wise Conditioning Adaptation for Fine-Tuning Diffusion Models in Text-to-Video Prediction

Zheyuan Liu
Junyan Wang
Zicheng Duan
Cristian Rodriguez-Opazo
Anton van den Hengel

Text-video prediction (TVP) is a downstream video generation task that requires a model to produce subsequent video frames given a series of initial video frames and text describing the required motion. In practice TVP methods focus on a particular category of videos depicting manipulations of objects carried out by human beings or robot arms. Previous methods adapt models pre-trained on text-to-image tasks, and thus tend to generate video that lacks the required continuity. A natural progression would be to leverage more recent pre-trained text-to-video (T2V) models. This approach is rendered more challenging by the fact that the most common fine-tuning technique, low-rank adaptation (LoRA), yields undesirable results. In this work, we propose an adaptation-based strategy we label Frame-wise Conditioning Adaptation (FCA). Within the module, we devise a sub-module that produces frame-wise text embeddings from the input text, which acts as an additional text condition to aid generation. We use FCA to fine-tune the T2V model, which incorporates the initial frame(s) as an extra condition. We compare and discuss the more effective strategy for injecting such embeddings into the T2V model. We conduct extensive ablation studies on our design choices with quantitative and qualitative performance analysis. Our approach establishes a new baseline for the task of TVP. Our code is open-source at https://github.com/Cuberick-Orion/FCA.

PDF Details

YNIMG Journal 2022 Journal Article

Deciphering the developmental order and microstructural patterns of early white matter pathways in a diffusion MRI based fetal brain atlas

Ruike Chen
Cong Sun
Tingting Liu
Yuhao Liao
Junyan Wang
Yi Sun
Yi Zhang
Guangbin Wang

White matter (WM) of the fetal brain undergoes rapid development to form early structural connections. Diffusion magnetic resonance imaging (dMRI) has shown to be a useful tool to depict fetal brain WM in utero, and many studies have observed increasing fractional anisotropy and decreasing diffusivity in the fetal brain during the second-to-third trimester, whereas others reported non-monotonic changes. Unbiased dMRI atlases of the fetal brain are important for characterizing the developmental trajectories of WM and providing normative references for in utero diagnosis of prenatal abnormalities. To date, the sole fetal brain dMRI atlas was collected from a Caucasian/mixed population and was constructed based on the diffusion tensor model with limited spatial resolution. In this work, we proposed a fiber orientation distribution (FOD) based pipeline for generating fetal brain dMRI atlases, which showed better registration accuracy than a diffusion tensor based pipeline. Based on the FOD-based pipeline, we constructed the first Chinese fetal brain dMRI atlas using 89 dMRI scans of normal fetuses at gestational age between 24 and 38 weeks. Complex non-monotonic trends of tensor- and FOD-derived microstructural parameters in eight WM tracts were observed, which jointly pointed to different phases of microstructural development. Specifically, we speculated that the turning point of the diffusivity trajectory may correspond to the starting point of pre-myelination, based on which, the developmental order of WM tracts can be mapped and the order was in agreement with the order of myelination from histological studies. The normative atlas also provided a reference for the detection of abnormal WM development, such as that in congenital heart disease. Therefore, the established high-order fetal brain dMRI atlas depicted the spatiotemporal pattern of early WM development, and findings may help decipher the distinct microstructural events in utero.

Details DOI

NeurIPS Conference 2022 Conference Paper

Entropy-Driven Mixed-Precision Quantization for Deep Network Design

Zhenhong Sun
Ce Ge
Junyan Wang
Ming Lin
Hesen Chen
Hao Li
Xiuyu Sun

Deploying deep convolutional neural networks on Internet-of-Things (IoT) devices is challenging due to the limited computational resources, such as limited SRAM memory and Flash storage. Previous works re-design a small network for IoT devices, and then compress the network size by mixed-precision quantization. This two-stage procedure cannot optimize the architecture and the corresponding quantization jointly, leading to sub-optimal tiny deep models. In this work, we propose a one-stage solution that optimizes both jointly and automatically. The key idea of our approach is to cast the joint architecture design and quantization as an Entropy Maximization process. Particularly, our algorithm automatically designs a tiny deep model such that: 1) Its representation capacity measured by entropy is maximized under the given computational budget; 2) Each layer is assigned with a proper quantization precision; 3) The overall design loop can be done on CPU, and no GPU is required. More impressively, our method can directly search high-expressiveness architecture for IoT devices within less than half a CPU hour. Extensive experiments on three widely adopted benchmarks, ImageNet, VWW and WIDER FACE, demonstrate that our method can achieve the state-of-the-art performance in the tiny deep model regime. Code and pre-trained models are available at https: //github. com/alibaba/lightweight-neural-architecture-search.

PDF Details

YNIMG Journal 2018 Journal Article

Modeling topographic regularity in structural brain connectivity with application to tractogram filtering

Junyan Wang
Dogu Baran Aydogan
Rohit Varma
Arthur W. Toga
Yonggang Shi

Topographic regularity is an important biological principle in brain connections that has been observed in various anatomical studies. However, there has been limited research on mathematically characterizing this property and applying it in the analysis of in vivo connectome imaging data. In this work, we propose a general mathematical model of topographic regularity for white matter fiber bundles based on previous neuroanatomical understanding. Our model is based on a novel group spectral graph analysis (GSGA) framework motivated by spectral graph theory and tensor decomposition. The GSGA provides a common set of eigenvectors for the graphs formed by topographic proximity of nearby tracts, which gives rises to the group graph spectral distance, or G2SD, for measuring the topographic regularity of each fiber tract in a tractogram. Based on this novel model of topographic regularity in fiber tracts, we then develop a tract filtering algorithm that can generally be applied to remove outliers in tractograms generated by any tractography algorithm. In the experimental results, we show that our novel algorithm outperforms existing methods in both simulation data from ISMRM 2015 Tractography Challenge and real data from the Human Connectome Project (HCP). On a large-scale dataset from 215 HCP subjects, we quantitatively show our method can significantly improve the retinotopy in the reconstruction of the optic radiation bundle. The software for the tract filtering algorithm developed in this work has also been publicly released on NITRC (https: //www. nitrc. org/projects/connectopytool).

Details DOI