Allen Tu
Graduate Machine Learning Researcher at UMIACS- Claim this Profile
Click to upgrade to our gold package
for the full feature experience.
-
English Native or bilingual proficiency
-
Chinese Native or bilingual proficiency
Topline Score
Bio
0
/5.0 / Based on 0 ratingsFilter reviews by:
Experience
-
UMIACS
-
United States
-
Research Services
-
1 - 100 Employee
-
Graduate Machine Learning Researcher
-
Aug 2023 - Present
Researching uncertainty estimation in neural radiance fields Advised by Professor Thomas Goldstein Researching uncertainty estimation in neural radiance fields Advised by Professor Thomas Goldstein
-
-
-
STR
-
United Kingdom
-
Services for Renewable Energy
-
1 - 100 Employee
-
Computer Vision Research Co-op
-
Jan 2023 - Aug 2023
Researched face, body, and multimodal recognition for IARPA’s Biometric Recognition and Identification at Accuracy and Range program; implemented feature upgrades to the evaluation pipeline and parallelized data processing for 300x speedup Increased open search performance by over 24.24% for face, 39.18% for body, and 49.10% for multimodal recognition through hard example training, optimized template compilation algorithms, and multi-model voting classifier ensembles Produced a self-supervised, operating condition invariant face recognition model and raised accuracy by 52.10%; engineered a suite of tools for training data analysis and balancing, feature space visualization, and hyperparameter tuning Show less
-
-
-
UMD Department of Computer Science
-
United States
-
Software Development
-
1 - 100 Employee
-
Peer Research Mentor (FIRE: Capital One Machine Learning)
-
Jan 2021 - Dec 2022
Researched realistic talking head generation with Dr. Raymond Tu (no relation); designed the model architecture; created a vector-quantized variational autoencoder and discriminator for realistic face synthesis; presented poster at 2 research eventsProduced project-based coursework, taught classes on research publications, and tutored students during lab hours; lead instruction on face recognition, autoencoders, and generative adversarial networks; advised 4 junior Peer Research MentorsCoordinated teams of 3-6 students building projects in text-to-audio generation, image super-resolution, and facial age and gender classification; facilitated research and development as the project architect, scrum master, and Git administratorCommitted to diversity, equity, and inclusion by providing an introduction to research experience that is accessible to freshmen from all backgrounds; mentored 80 underclassmen, many of whom continue to pursue computer science research Show less
-
-
Undergraduate Machine Learning Researcher
-
Jan 2021 - Dec 2022
Designed a human study with 5 prompt types to rank metrics for evaluating generative models by their alignment to human judgement of image quality, as well as investigate the limits of human perception of Fréchet Inception Distance (FID); Advised by Professor Thomas GoldsteinBuilt a framework for synthesizing and evaluating image distributions of varying qualities using the same generative adversarial network; controls FID by a range of over 100 points without destroying the composition of the images Identified behavioral trends of data analysts through a statistical analysis of 2.3 million dataset-visualization pairs in the Plotly Community Feed; proposed strategies for mitigating training bias in the VizML machine learning visualization recommenderFirst author of Honors Seminar paper "Addressing Bias in Visualization Recommenders by Identifying Trends in Training Data" (arXiv:2203.94937); advised by Drs. Leilani Battle and Dave Levin in the Undergraduate Honors Seminar Show less
-
-
-
-
Computer Vision Research Intern
-
Jun 2022 - Aug 2022
Created a generative model that improves the invariance of whole-body recognition models to clothing by transferring garments between multiple subjects in their training data; achieves FID 12.93 for upper and 9.29 for lower garment transfer Combined novel view synthesis, semantic segmentation, and pose estimation to reclothe subjects with high-fidelity textures while completely preserving the context and identity biometrics such as their face, hair, and gait. Designed the approach to operate from all camera poses, synthesize images congruent with their original operating conditions, and be robust against occlusions and distortions typical of footage from long-range security cameras and UAVs Show less
-
-
-
nCino, Inc.
-
United States
-
Software Development
-
700 & Above Employee
-
Software Engineering Intern
-
Jun 2021 - Aug 2021
Fulfilled bug and feature tickets across 3 nCino cloud Bank Operating System packages using Java, JavaScript, and SQL Integrated the Salesforce platform with 5 partner credit, loan, and insurance services using Informatica, REST, and SOAP Implemented complete regression testing suites for 9 packages; demoed to quality assurance testers and technical writers Fulfilled bug and feature tickets across 3 nCino cloud Bank Operating System packages using Java, JavaScript, and SQL Integrated the Salesforce platform with 5 partner credit, loan, and insurance services using Informatica, REST, and SOAP Implemented complete regression testing suites for 9 packages; demoed to quality assurance testers and technical writers
-
-
Education
-
University of Maryland
Master of Science - MS, Computer Science -
University of Maryland
Bachelor of Science - BS, Computer Science -
Canyon Crest Academy