About Me

I am a PhD student in Computer Science at the University of Maryland, College Park (UMD), advised by Professor Tom Goldstein. My research spans deep learning, computer vision, and graphics, with emphasis on 3D scene reconstruction, multimodal biometric recognition, and generative priors for robust vision systems. I have published in top machine learning conferences, contributing methods in compression and explainability that enable large-scale recognition and 3D reconstruction in real-world settings. In addition to my academic work, I collaborate with both academic and industry partners on problems spanning vision, language, and generative modeling. I completed my BS/MS in Computer Science at UMD in 2024.

I was a Computer Vision Research Intern at Systems & Technology Research (STR), focusing on multimodal biometric recognition, and a Peer Research Mentor in the Capital One Machine Learning FIRE program at UMD, where I mentored over 80 undergraduates in their first research experiences. These roles deepened my interest in building machine learning systems that are both efficient and trustworthy. A recurring theme in my work is designing methods that not only achieve high accuracy, but also scale effectively and provide interpretable signals for decision making. My goal is to advance AI that can be deployed in real-world applications, ensuring systems that are safe, reliable, and beneficial to society.

News

• October 2025: Released our TransFIRA preprint.
• September 2025: STR has opened internships for Summer 2026. Reach out for a referral!
• June 2025: Thank you for coming to see PUP 3D-GS and Speedy-Splat at CVPR 2025! 🎉
• June 2025: Released our Speedy Deformable 3D Gaussian Splatting paper and code.
• May 2025: Returned to STR as a Computer Vision Research Intern.

Research Highlights

	TransFIRA: Transfer Learning for Face Image Recognizability Assessment Preprint, 2025. PAPER WEBSITE Redefine template-based recognition through encoder-grounded recognizability prediction that learns directly from embedding geometry via class-center similarity and angular separation, enabling principled filtering, calibrated weighting, and cross-modal explainability that surpass prior FIQA methods in accuracy, interpretability, and generality. Allen Tu, Kartik Narayan, Joshua Gleason, Jennifer Xu, Matthew Meyn, Tom Goldstein, Vishal Patel
	Speedy Deformable 3D Gaussian Splatting: Fast Rendering and Compression of Dynamic Scenes Preprint, 2025. PAPER CODE ★ WEBSITE Achieve 10.38× faster rendering, 7.71× smaller model size, and 2.71× shorter training time for dynamic 3D Gaussian Splatting representations through temporal sensitivity pruning and flow-based Gaussian grouping while maintaining visual quality. Allen Tu*, Haiyang Ying*, Alex Hanson, Yonghan Lee, Tom Goldstein, Matthias Zwicker
	Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives CVPR, 2025. PAPER CODE ★ WEBSITE Accelerate 3D Gaussian Splatting rendering speed by over 6× and reduce model size by over 90% through accurately localizing primitives during rasterization and pruning the scene during training, providing a significantly higher speedup than existing techniques while maintaining competitive image quality. Alex Hanson, Allen Tu, Geng Lin, Vasu Singla, Matthias Zwicker, Tom Goldstein
	PUP 3D-GS: Principled Uncertainty Pruning for 3D Gaussian Splatting CVPR, 2025. PAPER CODE ★ WEBSITE Prune 90% of primitives from any pretrained 3D Gaussian Splatting model using a mathematically principled sensitivity score, more than tripling rendering speed while retaining more salient foreground information and higher visual fidelity than previous techniques at a substantially higher compression ratio. Alex Hanson*, Allen Tu*, Vasu Singla, Mayuka Jayawardhana, Matthias Zwicker, Tom Goldstein
	Addressing Bias in Visualization Recommenders by Identifying Trends in Training Data: Improving VizML Through a Statistical Analysis of the Plotly Community Feed Undergraduate Honors Seminar, 2021. PAPER Analyze training bias in the VizML visualization recommender by evaluating 1,000 Plotly Community Feed examples, uncover how user convenience and chart imbalance skew predictions, and propose data-driven strategies for fairer, more representative recommendations. Allen Tu, Priyanka Mehta, Alexander Wu, Nandhini Krishnan, Amar Mujumdar, Leilani Battle, Dave Levin
	IARPA Walk-through Rendering from Images of Varying Altitude (WRIVA) UMIACS, 2023-Present. Unconstrained 3D reconstruction and rendering of real-world environments. Efficient rendering, compression, and training for 3D Gaussian Splatting (3DGS). [1, 2, 3] Image, video, and multiview diffusion model priors for sparse-view 3D reconstruction. Uncertainty quantification algorithms for 3DGS and Neural Radiance Fields (NeRF). PIs: Prof. Tom Goldstein, Prof. Matthias Zwicker, Prof. Abhinav Shrivastava, Dr. Abhay Yadav, Dr. Cheng Peng, Prof. Rama Chellappa
	IARPA Biometric Recognition and Identification at Altitude and Range (BRIAR) STR, 2022-Present. Multimodal fusion of face, body, and gait recognition in severe real-world scenarios. Transfer Learning for Face Image Recognizability Assessment. (2025) Learned Frame Feature Aggregation for Face Recognition with Low-Quality Video. (2024) Operating Condition-Invariant Barlow Twins and Multimodal Ensembling. (2023) Style-Based Appearance Flow for Clothing-Robust Body Representation Learning. (2022) PIs: Dr. Joshua Gleason, Dr. Jennifer Xu, Dr. Nathan Shnidman, Dr. Mark Keck, Prof. Vishal Patel, Prof. Rama Chellappa

Experience