CV

A PDF version of my CV is available here.

Research Interests

Inverse problems and sparse recovery; computational imaging; generative modeling (diffusion, flow matching, autoregressive); attention mechanisms and token mixing in measurement space; continual learning and model merging under fixed budgets; emerging interest in physics-constrained latent dynamics and world models.

Education

• Ph.D in Electrical & Computer Engineering, Johns Hopkins University, 2023 - Present
  • Advisor: Prof. Trac D. Tran
  • Selected coursework: Optimization in Data Science, Matrix Analysis, Random Matrix Theory, Probability Theory, Equivariant Machine Learning, Compressed Sensing, Wavelets
• M.S. in Electrical & Computer Engineering, University of Southern California, 2021 - 2023
  • Advisor: Prof. Justin P. Haldar
  • Honor Student
• B.S. in Automation, Xi’an Jiaotong University, 2016 - 2020

Honors & Awards

• Top Reviewer, Conference on Neural Information Processing Systems (NeurIPS), 2026
• MRM Highlights Selection, Magnetic Resonance in Medicine, 2025
• Honor Student, M.S. in Electrical & Computer Engineering, University of Southern California, 2023

Research Experience

• 2024 - Present: Graduate Research Assistant — Attention-Guided Sparse Recovery and Token Mixing
  • Johns Hopkins University (Advisor: Prof. Trac D. Tran)
  • Proved that self-attention computed on measurements approximates attention on the underlying signal under RIP-style conditions, and established an asymmetric design principle — global analysis, local reconstruction — through a 2x2 factorial study of encoder/decoder cell types.
  • Derived a pre-training diagnostic that predicts when attention beats uniform mixing from a conditional-variance statistic estimated on image pairs alone.
• 2025 - Present: Graduate Research Assistant — Memory-Bounded Continual Learning
  • Johns Hopkins University (Advisors: Prof. Trac D. Tran & Prof. Vishal M. Patel)
  • Developed sparse consolidation of task adapters into a fixed-size store, paired with a compressed-sensing forecast of how far back stored adaptations remain recoverable.
  • Benchmarked against parameter-efficient and merging baselines on ImageNet-R and OmniBench-1K across horizons up to T=100 tasks.
• Sep 2023 - Present: Graduate Research Assistant — Lensless Epi-Fluorescent Microendoscopy
  • Johns Hopkins University (Advisors: Prof. Trac D. Tran & Prof. Mark A. Foster)
  • Built calibration-free sparse recovery (ElasticNet sensing-matrix estimation, OMP reconstruction) for fluorescence imaging through coded-aperture optics, reaching ~10x speedup over iterative baselines.
  • Extended the framework to region-of-interest reconstruction for real-time neuron imaging in deep-brain applications.
• 2024 - Present: Research Collaborator — Refinement and Correction for Autoregressive Image Generation
  • Tencent, Shenzhen (with Cheng Cheng)
  • Reformulated AR generation as a Markov process over overlapping tensors with a discrete tensor-noising scheme, connecting iterative refinement to discrete diffusion.
  • Designed an acceptance-gated correction policy for frozen AR generators that preserves well-tuned samples token-for-token while repairing degraded ones, with no test-time degradation label.

Teaching Experience

• Teaching Assistant, EN.520.646 Wavelets and Filter Banks, Johns Hopkins University, Fall 2026
• Teaching Assistant, EN.520.651 Random Signal Analysis, Johns Hopkins University, Fall 2024 & Fall 2025
  • Led discussion sections, designed and graded problem sets and exams, and held weekly office hours for graduate students.

Publications

Refine or Preserve? Acceptance-Gated Correction for Autoregressive Image Generation

D. An, C. Cheng, Z. Zhou, D. Poppert, T. D. Tran. "Refine or Preserve? Acceptance-Gated Correction for Autoregressive Image Generation." Under review at ICLR 2027.

Learning While Forgetting: Sparse Adapter Consolidation for Memory-Bounded Continual Learning

D. An, D. Ye, J. Xu, C. Cheng, Y. Wei, K. Zhang, V. M. Patel, T. D. Tran. "Learning While Forgetting: Sparse Adapter Consolidation for Memory-Bounded Continual Learning." Under review at ICLR 2027.

When Attention Beats Uniform Mixing: A Conditional-Variance Hinge for Image Restoration

D. An, D. Poppert, T. D. Tran. "When Attention Beats Uniform Mixing: A Conditional-Variance Hinge for Image Restoration." Under review at ICASSP 2027.

When Do Encoder-Decoders Need Attention?

D. An, T. D. Tran. "When Do Encoder-Decoders Need Attention?" Under review at UniReps Workshop, NeurIPS 2026. Extended abstract track.

From Prediction to Perfection: Introducing Refinement to Autoregressive Image Generation

C. Cheng, L. Song, D. An, Y. Xiao, X. Zhang, H. Sun, Y. Shan. (2026). "From Prediction to Perfection: Introducing Refinement to Autoregressive Image Generation." International Conference on Learning Representations (ICLR).

Attention Networks for Spotlight SAR Motion Compensation

D. Poppert, D. An, T. D. Tran. (2025). "Attention Networks for Spotlight SAR Motion Compensation." 59th Asilomar Conference on Signals, Systems, and Computers.

Region-of-Interest Sparse Reconstruction for Lensless Coded-Aperture Optical Imaging

D. An, D. Poppert, J. Li, J. Pham, B. Sun, M. Foster, T. D. Tran. (2025). "Region-of-Interest Sparse Reconstruction for Lensless Coded-Aperture Optical Imaging." 59th Asilomar Conference on Signals, Systems, and Computers, pp. 1517-1521.

A Minimally-Invasive Lensless Epi-Fluorescent Microendoscope Leveraging Learned Sensing

J. Li, D. An, M. Foster, T. D. Tran, et al. "A Minimally-Invasive Lensless Epi-Fluorescent Microendoscope Leveraging Learned Sensing." In preparation for Optics Express.

Motion Compensation for Synthetic Aperture Radar with the Vision Transformer

D. Poppert, D. An, E. D. Jansing, T. D. Tran. (2025). "Motion Compensation for Synthetic Aperture Radar with the Vision Transformer." IEEE International Geoscience and Remote Sensing Symposium (IGARSS).

TRUST: Transformer-Driven U-Net for Sparse Target Recovery

D. An, D. Poppert, J. Li, M. Foster, T. D. Tran. (2025). "TRUST: Transformer-Driven U-Net for Sparse Target Recovery." Under review at IEEE TPAMI. arXiv:2506.01112.

The "hidden noise" problem in MR image reconstruction

J. Wang, D. An, J. P. Haldar. (2024). "The ‘hidden noise’ problem in MR image reconstruction." Magnetic Resonance in Medicine, 92(3), 982-996.

The problem of hidden noise in MR image reconstruction

J. Wang, D. An, J. P. Haldar. (2023). "The problem of hidden noise in MR image reconstruction." Proc. International Society for Magnetic Resonance in Medicine (ISMRM).

Technical Skills

• Programming: Python (PyTorch, NumPy, SciPy), MATLAB, C/C++
• Methods: Optimization, sparse recovery, signal processing, generative modeling, deep learning, continual learning
• Languages: Mandarin (native), English (fluent)
• Tools: Linux, Git, LaTeX, CUDA, multi-GPU training (DDP)

Professional Service

• Reviewer, Conference on Neural Information Processing Systems (NeurIPS), 2026 — Main Reviewer, recognized as Top Reviewer
• Reviewer, International Conference on Learning Representations (ICLR), 2027
• Reviewer, IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2024 - Present
• Reviewer, UniReps Workshop, NeurIPS, 2026