Jingewi Zhang (张敬玮)

Jingewi Zhang (张敬玮)

CS Ph.D candidate in Stony Brook University

Publications

You can also find my articles on my Google Scholar profile.

Journal Articles

LBMamba: Locally Bi-directional Mamba

Published in Transactions on Machine Learning Research (TMLR), 2025

LBMamba: Locally Bi-directional Mamba

We present LBMamba, a locally bi-directional State Space Model block that integrates a lightweight, register-level backward scan within the forward scan to restore a full receptive field without the computational overhead of a secondary global backward pass.

Recommended citation: Jingwei Zhang*, Xi Han*, Hong Qin, Mahdi S. Hosseini, and Dimitris Samaras, "LBMamba: Locally Bi-directional Mamba", Transactions on Machine Learning Research (TMLR), 2025.
Download Paper

Conference Papers

ChannelSFormer: A Channel Agnostic Vision Transformer for Multi-Channel Cell Painting Images

Published in NeurIPS 2025 Workshop for Imageomics: Discovering Biological Knowledge from Images Using AI, 2025

ChannelSFormer: A Channel Agnostic Vision Transformer for Multi-Channel Cell Painting Images

ChannelSFormer is a channel-agnostic Vision Transformer that decomposes self-attention into spatial-wise and channel-wise attention for flexible and efficient processing of multi-channel Cell Painting images.

Recommended citation: Jingwei Zhang, Srinivasan Sivanandan, "ChannelSFormer: A Channel Agnostic Vision Transformer for Multi-Channel Cell Painting Images", NeurIPS 2025 Workshop for Imageomics, 2025.
Download Paper

2DMamba: Efficient State Space Model for Image Representation with Applications on Giga-Pixel Whole Slide Image Classification

Published in Proceedings of 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

2DMamba: Efficient State Space Model for Image Representation with Applications on Giga-Pixel Whole Slide Image Classification

2DMamba is an efficient state space model for image representation, which extends 1D Mamba into 2D while maintaining its modeling capabilities, high parallelism, and memory access efficiency.

Recommended citation: Jingwei Zhang*, Anh Tien Nguyen*, Xi Han*, Vincent Quoc-Huy Trinh, Hong Qin, Dimitris Samaras, and Mahdi S. Hosseini, "2DMamba: Efficient State Space Model for Image Representation with Applications on Giga-Pixel Whole Slide Image Classification", In Proceedings of the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025.
Download Paper

SAM-Path: A Segment Anything Model for Semantic Segmentation in Digital Pathology

Published in MedAGI at International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) (Oral), 2023

SAM-Path: A Segment Anything Model for Semantic Segmentation in Digital Pathology

SAM-Path adapts the Segment Anything Model for semantic segmentation in digital pathology by introducing trainable class prompts and incorporating a pathology foundation model encoder.

Recommended citation: Jingwei Zhang, Ke Ma, Saarthak Kapse, Joel Saltz, Maria Vakalopoulou, Prateek Prasanna, Dimitris Samaras, "SAM-Path: A Segment Anything Model for Semantic Segmentation in Digital Pathology", MedAGI at International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2023.
Download Paper

Prompt-MIL: Boosting Multi-Instance Learning Schemes via Task-specific Prompt Tuning

Published in International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) (Oral), 2023

Prompt-MIL: Boosting Multi-Instance Learning Schemes via Task-specific Prompt Tuning

Prompt-MIL integrates prompt tuning into multi-instance learning for end-to-end whole slide image classification, enabling task-specific feature calibration with minimal additional parameters within GPU memory limitation.

Recommended citation: Jingwei Zhang*, Saarthak Kapse*, Ke Ma, Prateek Prasanna, Joel Saltz, Maria Vakalopoulou, Dimitris Samaras, "Prompt-MIL: Boosting Multi-Instance Learning Schemes via Task-specific Prompt Tuning", International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2023.
Download Paper

Precise Location Matching Improves Dense Contrastive Learning in Digital Pathology

Published in International Conference on Information Processing in Medical Imaging (IPMI), 2022

Precise Location Matching Improves Dense Contrastive Learning in Digital Pathology

We propose a precise location-based matching mechanism for dense contrastive learning that utilizes overlapping information between geometric transformations to accurately match dense features in histopathology images.

Recommended citation: Jingwei Zhang*, Saarthak Kapse*, Ke Ma, Prateek Prasanna, Maria Vakalopoulou, Joel Saltz, Dimitris Samaras, "Precise Location Matching Improves Dense Contrastive Learning in Digital Pathology", International Conference on Information Processing in Medical Imaging (IPMI), 2023.
Download Paper

Gigapixel Whole-Slide Images Classification using Locally Supervised Learning

Published in International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) (Oral), 2022

Gigapixel Whole-Slide Images Classification using Locally Supervised Learning

We propose a locally supervised learning framework that divides a pre-trained network into modules optimized locally with auxiliary models, enabling efficient whole slide image classification without exhaustive patch-level processing.

Recommended citation: Jingwei Zhang*, Xin Zhang*, Ke Ma, Rajarsi Gupta, Joel Saltz, Maria Vakalopoulou, Dimitris Samaras, "Gigapixel Whole-Slide Images Classification using Locally Supervised Learning", International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2022.
Download Paper

A Joint Spatial and Magnification Based Attention Framework for Large Scale Histopathology Classification

Published in CVMI at IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (Best paper), 2021

A Joint Spatial and Magnification Based Attention Framework for Large Scale Histopathology Classification

We propose a spatial and magnification-based attention sampling strategy that estimates attention maps from down-sampled images to efficiently select informative patches at optimal magnifications for large-scale histopathology classification.

Recommended citation: Jingwei Zhang, Ke Ma, John Van Arnam, Rajarsi Gupta, Joel Saltz, Maria Vakalopoulou, Dimitris Samaras, "A Joint Spatial and Magnification Based Attention Framework for Large Scale Histopathology Classification", CVMI at IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023.
Download Paper