Research Interests

Trustworthy Machine Learning:

• Interpretable & Verifiable AI: Understanding how AI models reason and ensuring predictions are supported by valid evidence.

• - [ECCV’26 Oral] Reading "World Knowledge" from ViTs
• - [ICML’26] Inside the mind of vision transformers
• - [ICML’25] Why is there a tumor? Tell me the reason, show me the evidence
• - [NeurIPS’24] Double-correct predictions

• Robust & Generalizable AI: Developing models that remain reliable under distribution shifts, rare events, and previously unseen conditions.

• - [CVPR’26 Highlight] Measuring ViT generalization
• - [ICLR’23] Topology-aware OOD generalization
• - [CVPR’20] Single Domain generalization

• Vision-Language & Foundation Models: Building multimodal models that connect perception, knowledge, reasoning, and evidence.

• - [NeurIPS’26] VIGOR: Benchmarking Visual Rationale Correctness
• - [NeurIPS’24] Localize Concept Explanations for VLM
• - [AAAI’21] Multimodal Learning with Severely Missing Modality

Trustworthy AI in the Real World:

• AI for Health: Trustworthy multimodal AI for medical imaging, diagnosis, and treatment-response modeling.

• - NIH R21 (PI), NSF III CORE (PI), MSK Cancer Center (PI)

• AI for Science: Robust foundation models for scientific discovery, including ocean and seafloor intelligence.

• - NSF CAREER (PI), DoD DEPSCoR (PI)

• AI for Autonomous Systems: Safe and fault-tolerant learning for autonomous vehicles and other intelligent systems.

• - NSF SLES Medium (PI)

Awards

I am honored with prestigious research awards :

• Copenhaver Fellow, UVA, 2026
• Outstanding Faculty Research Award, U Delaware, 2024
• NSF CAREER Award, 2024
• DOD DEPSCoR Award, 2023
• Google Faculty Research Award, 2022
• General University Research Award, U Delaware, 2022
• Research Foundation Award, U Delaware, 2021 

My work won a series of paper awards in top-tier AI/ML conferences:

• Oral Presentation, ECCV 2026, acceptance rate 1.0%
• Highlight Presentation, CVPR 2026, acceptance rate 3.0%
• Oral Presentation, AAAI 2025, acceptance rate 4.7%
• Strong Double Blind, ECCV 2024, acceptance rate 4.5%
• Best Paper Award, NeurIPSW 2021
• Oral Presentation, CVPR 2021, acceptance rate 4.3%
• Spotlight Presentation, ICLR 2021, acceptance rate 3.8%
• Oral Presentation, ICCV 2019, acceptance rate 4.7%
• Oral Presentation, KDD 2019, acceptance rate 9.2%
• Oral Presentation, BMVC 2018, acceptance rate 4.6%
• Best Student Paper Finalist, ECCV 2016, acceptance rate 0.4%
• Oral Presentation, ICPR 2016, acceptance rate 14.1%
• Oral Presentation Award, ACCV 2010, acceptance rate 3.5%

News and media about my research work:

• Dr. Xi Peng develops AI that thinks and reasons like scientists. [Link]
• Dr. Xi Peng develops trustworthy AI for seafloor data intelligence. [Link]
• DOD announces awards ($600K) under the Defense Established Program to Stimulate Competitive Research. [Link]
• Interdisciplinary UD team received ($1M) NSF grant to map global illicit trade of energy-critical materials. [Link]
• A vision of the fast-growing Data Science Institute at UD. [Link]

Funding

Federal Grants: 

• NIH R21 Award (PI), with O. Akin, B. Zhao, L. Schwart
• NSF CAREER Award (PI)
• NSF III CORE (PI), with R. Eigenmann (co-PI)
• NSF Safe Learning-enable System (PI), with W. Shi (co-PI) and C. Yang (co-PI)
• DOD DEPSCoR (PI), with A. Trembanis (co-PI)
• CDC Contract (PI)
• NSF CMMI (Co-PI), with J. Klinger (PI)
• NSF HDR (KP), with F. Bianco (PI)
• NSF E-CORE RES (KP), with A. Slocum
• NSF AI Institute HAIC (KP), with C. Wu

Industrial Grants:

• MSK Cancer Center (PI)
• Google Research Faculty Award (PI)
• Snap Research Award (PI)

Internal Grants: UDRF (PI); GUR (PI); AICoE Seed (PI); DSI Seed (PI); UDRF-SI (Co-PI)

Teaching

At UVA

CS6501 Trustworthy Machine Learning

• 2026 Fall

At U Delaware (internal rating 4.4/5; RateMyProfessors 4.6/5) 

Undergraduate-Level: CISC484 Intro to Machine Learning

• 2019 Fall; 2021 Spring; 2021 Fall; 2022 Fall; 2022 Spring 

Graduate-Level: CISC684 Intro to Machine Learning

• 2022 Fall; 2023 Fall; 2024 Spring; 2024 Fall

Advanced Graduate-Level: CISC889 Advanced Topics in Machine Learning and Deep Neural Networks

• 2020 Spring; 2020 Fall; 2022 Spring

Service

Conference/Journal Committee:

• Conference Area Chair: NeurIPS, CVPR, BMVC, IISE Annual Conf & Expo, ICIG
• Conference Program Committee: NeurIPS, ICML, ICLR, CVPR, ECCV, ICCV, ACL, AAAI, IJCAI
• Journal Guest Editor: Remote Sensing, Neurocomputing, CVIU
• Journal Reviewer: TPAMI, IJCV, TIP, TNNLS, PR, TMM, THMS, JOB...

Proposal and Grant Review Panels:

• NSF Review Panel: RI (2026, 2 panels), CAIG (2026), CPS (2025), OAC (2024), III (2024), RI (2023), III (2023), III (2022), CPS (2022), RI (2021), CPS (2020)
• NIH Review Panel: 2026 (2 panels), 2025
• DOD Review Panel: 2026, 2025
• External Proposal Review: Linz Institute of Technology (2024), University of Sydney (2021), University of Central Florida (2020)
• Internal Proposal Review: AICoE (2024), UDRF (2023), GUR (2023)

Students

PhD students:

• Mengmeng Ma, PhD candidate, Previous: University of Southern California
• Jeffrey Peng, PhD candidate, Previous: Columbia University
• Yanlin Chen, PhD candidate, Previous: Southern University of Science and Technology 
• Jie Wei, PhD candidate, Previous: University of Electronic Science and Technology of China 
• Yuping Zheng, PhD student, Previous: The Chinese University of Hong Kong, Shenzhen
• Jiazheng (Justin) Wang, PhD student, Previous: University of Chicago

Alumni:

• Tang Li, PhD, Now: Applied Scientist @ Amazon
• Fengchun Qiao, PhD, Now: Tenure-track Assistant Professor in Bellini College of AI @ University of South Florida
• Nathaniel Merrill, PhD, (Co-advise), Now: Scientist @ Quidient

Visiting PhD student:

• Ricardo Santos, PhD candidate, Universidade NOVA de Lisboa (Portugal)

Undergraduate Researchers:

• VIP Program: Jakeb Miburn, Coleman Walsh, Furdeen Hasan, Jonathan Ma, Michael Lutz
• Amani A. Kiruga (Junior): Paul D. Amer Meritorious Award; MIT 2023 Summer Research
• Wenxuan Li (Senior): Dean's list; now JHU
• Ruoxi Jin (Senior): Dean's list
• Jonathan Ma (Junior)

Publication

Selected publication from my group (Click here for a full list)

[NeurIPS'26] VIGOR: Benchmarking Visual Rationale Correctness in Multimodal Large Language Models. [PDF] [Code]

[ECCV'26 Oral] "World Knowledge" in the Weights: Reading Concept Circuits of Vision Transformers. [PDF] [Code]

[ECCV'26] Medical AI Encodes a “Feeling of Error”: Verifying Cancer Segmentation via Internal Concepts. [PDF] [Code]

[ICML'26] Inside the Visual Mind: Neuroscience-Motivated Concept Circuits for Interpreting and Steering Vision Transformers. [PDF] [Video] [Code]

[CVPR'26 Highlight] Inside-Out: Measuring Generalization in Vision Transformers Through Inner Workings. [PDF] [Code] [Video]

[ICML'25] Structure-informed Risk Minimization for Robust Ensemble Learning. [PDF]

[ICML'25] "Why Is There a Tumor?": Tell Me the Reason, Show Me the Evidence. [PDF]

[AAAI'25 Oral] Interpretable Failure Detection with Human-Level Concepts. [PDF]

[AAAI'25] Beyond Accuracy: On the Effects of Fine-tuning Towards Vision-Language Model's Prediction Rationality. [PDF]

[NeurIPS'24] Beyond Accuracy: Ensuring Correct Predictions with Correct Rationales. [PDF] [Code]

[NeurIPS'24] SeafloorAI: A Large-scale Vision-Language Dataset for Seafloor Geological Survey. [PDF] [Dataset]

[ICML'24] Ensemble Pruning for Out-of-distribution Generalization. [PDF] [Code]

[ICML'24] Beyond Federation: Topology-aware Federated Learning for Generalization to Unseen Clients. [PDF] [Code]

[ECCV'24 Strong Double Blind] DEAL: Disentangle and Localize Concept-level Explanations for VLM. [PDF] [Code]

[CIKM'24] Adaptive Cascading Network for Continual Test-Time Adaptation. [PDF] [Code]

[ICCV'23] Learning from Semantic Alignment between Unpaired Multiviews for Egocentric Video Recognition. [PDF] [Code]

[CVPR'23] Are Data-driven Explanations Robust against Out-of-Distribution Data?[PDF] [Code]

[ICLR'23] Topology-aware Robust Optimization for Out-of-Distribution Generalization. [PDF] [Code]

[TNNLS'23, IF=14.3] Semi-identical Twins Variational AutoEncoder for Few-Shot Learning. [PDF]

[TPAMI'22, IF=24.3] Out-of-Domain Generalization from a Single Source: An Uncertainty Quantification Approach. [PDF] [Code]

[TMM'22, IF=8.2] Region-aware Arbitrary-shaped Text Detection with Progressive Fusion. [PDF] [Code]

[CVPR'22] Are multimodal transformers robust to missing modality? [PDF] [Code]

[CVPR'22] Symmetry and uncertainty-aware object slam for 6dof object pose estimation. [PDF] [Code]

[NeurIPS'21W Best Paper Award] Deep learning for spatiotemporal modeling of Urbanization. [PDF] [Video-10m]

[ICLR'21 Spotlight] A good image generator is what you need for high-resolution video synthesis. [PDF] [Video-10m] [Code]

[CVPR'21] Uncertainty-guided Model Generalization to Unseen Domains. [PDF] [Video-5m] [Code]

[CVPR'21 Oral] Learning View-Disentangled Human Pose Representation by Contrastive Cross-View Mutual Information Maximization.  [PDF] [Video-5m] [Code]

[AAAI'21] Multimodal learning with severely missing modality. [PDF] [Video-60s] [Video-15m] [Code]

[NSDI'21] Adapting Wireless Mesh Network Configuration from Simulation to Reality via Deep Learning-based Domain Adaptation. [PDF]

[IJCV'20, IF=11.5] Towards image-to-video translation: A structure-aware approach via multi-stage generative adversarial networks. [PDF]

[NeurIPS'20] Maximum-entropy adversarial data augmentation for improved generalization and robustness. [PDF] [Code]

[CVPR'20] Learning to learn single domain generalization. [PDF] [Video-60s] [Code]

[CVPR'20] Knowledge as priors: Cross-modal knowledge generalization for datasets without superior knowledge. [PDF] [Video-60s]

[TPAMI'19, IF=24.3] Towards Efficient U-Nets: A Coupled and Quantized Approach. [PDF]

[NeurIPS'19] Semantic-guided multi-attention localization for zero-shot learning. [PDF]

[NeurIPS'19] Rethinking kernel methods for node representation learning on graphs. [PDF] [Code]

[ICCV'19 Oral] AdaTransform: Adaptive Data Transformation. [PDF]

[CVPR'19] Semantic graph convolutional networks for 3d human pose regression. [PDF]

[KDD'19 Oral] Scalable Global Alignment Graph Kernel Using Random Features: From Node Embedding to Graph Embedding. [PDF]

[CVPR'18] Jointly optimize data and network training: Adversarial data augmentation in human pose estimation. [PDF]

[CVPR'18] A generative adversarial approach for zero-shot learning from noisy texts. [PDF]

[ECCV'18] Quantized densely connected u-nets for efficient landmark localization. [PDF]

[ECCV'18] Learning to forecast and refine residual motion for image-to-video generation. [PDF]

[IJCAI'18] Cr-gan: Learning complete representations for multi-view generation. [PDF]

[IJCV'18, IF=11.5] Red-net: A recurrent encoder-decoder network for video-based face alignment. [PDF]

[IJCV'18, IF=11.5] Toward personalized modeling: Incremental and ensemble alignment for sequential faces in the wild. [PDF]

[ICCV'17] Reconstruction-based disentanglement for pose-invariant face recognition. [PDF]

[ECCV'16 Best Student Paper Finalist] A recurrent encoder-decoder network for sequential face alignment. [PDF]

[ICCV'15] PIEFA: Personalized incremental and ensemble face alignment. [PDF]