About

Tracy Anne Hammond is a Professor in the Department of Computer Science & Engineering at Texas A&M University. Her research interests include sketch recognition, perception, cognitive behavior, computer-human interaction, artificial intelligence, concept learning, computer graphics, psychology, and anthropology. She has a strong background in computer science and anthropology, holding a Ph.D. in Computer Science from the Massachusetts Institute of Technology, as well as master's degrees from Columbia University. Hammond's work focuses on developing perception-based languages for sketch recognition, recognizing free-hand diagrams, and creating interactive systems for sketch recognition, including applications for Japanese Kanji and primitive sketch beautification. Throughout her career, Hammond has received numerous awards and honors, including best paper awards, distinguished fellowships, and invitations to speak at international conferences. Her contributions have advanced understanding in sketch recognition systems, human-computer interaction, and artificial intelligence, with a particular emphasis on perception and cognitive processes involved in sketching and recognition tasks.

Research topics

• Computer Science
• Data Mining
• Political Science
• Sociology
• Machine Learning
• Geography
• Mathematics
• Public relations
• Psychology
• Engineering
• Cartography
• Management
• Virology
• Pedagogy
• Social psychology
• Business
• Medicine
• Economics
• Ecology
• Telecommunications
• Remote sensing
• Biology
• Statistics

Selected publications

• Masked Contrastive Pre-Training Improves Music Audio Key Detection

  arXiv (Cornell University) · 2026-04-11

  preprintOpen access

  Self-supervised music foundation models underperform on key detection, which requires pitch-sensitive representations. In this work, we present the first systematic study showing that the design of self-supervised pretraining directly impacts pitch sensitivity, and demonstrate that masked contrastive embeddings uniquely enable state-of-the-art (SOTA) performance in key detection in the supervised setting. First, we discover that linear evaluation after masking-based contrastive pretraining on Mel spectrograms leads to competitive performance on music key detection out of the box. This leads us to train shallow but wide multi-layer perceptrons (MLPs) on features extracted from our base model, leading to SOTA performance without the need for sophisticated data augmentation policies. We further analyze robustness and show empirically that the learned representations naturally encode common augmentations. Our study establishes self-supervised pretraining as an effective approach for pitch-sensitive MIR tasks and provides insights for designing and probing music foundation models.

  PublisherDOI

• DependencyAI: Detecting AI Generated Text through Dependency Parsing

  arXiv (Cornell University) · 2026-02-17

  preprintOpen accessSenior author

  As large language models (LLMs) become increasingly prevalent, reliable methods for detecting AI-generated text are critical for mitigating potential risks. We introduce DependencyAI, a simple and interpretable approach for detecting AI-generated text using only the labels of linguistic dependency relations. Our method achieves competitive performance across monolingual, multi-generator, and multilingual settings. To increase interpretability, we analyze feature importance to reveal syntactic structures that distinguish AI-generated from human-written text. We also observe a systematic overprediction of certain models on unseen domains, suggesting that generator-specific writing styles may affect cross-domain generalization. Overall, our results demonstrate that dependency relations alone provide a robust signal for AI-generated text detection, establishing DependencyAI as a strong linguistically grounded, interpretable, and non-neural network baseline.

  PublisherDOI

• Masked Contrastive Pre-Training Improves Music Audio Key Detection

  arXiv (Cornell University) · 2026-04-11

  articleOpen access

  Self-supervised music foundation models underperform on key detection, which requires pitch-sensitive representations. In this work, we present the first systematic study showing that the design of self-supervised pretraining directly impacts pitch sensitivity, and demonstrate that masked contrastive embeddings uniquely enable state-of-the-art (SOTA) performance in key detection in the supervised setting. First, we discover that linear evaluation after masking-based contrastive pretraining on Mel spectrograms leads to competitive performance on music key detection out of the box. This leads us to train shallow but wide multi-layer perceptrons (MLPs) on features extracted from our base model, leading to SOTA performance without the need for sophisticated data augmentation policies. We further analyze robustness and show empirically that the learned representations naturally encode common augmentations. Our study establishes self-supervised pretraining as an effective approach for pitch-sensitive MIR tasks and provides insights for designing and probing music foundation models.

  PublisherOA PDF

• MambaLiteUNet: Cross-Gated Adaptive Feature Fusion for Robust Skin Lesion Segmentation

  arXiv (Cornell University) · 2026-04-22

  articleOpen accessSenior author

  Recent segmentation models have demonstrated promising efficiency by aggressively reducing parameter counts and computational complexity. However, these models often struggle to accurately delineate fine lesion boundaries and texture patterns essential for early skin cancer diagnosis and treatment planning. In this paper, we propose MambaLiteUNet, a compact yet robust segmentation framework that integrates Mamba state space modeling into a U-Net architecture, along with three key modules: Adaptive Multi-Branch Mamba Feature Fusion (AMF), Local-Global Feature Mixing (LGFM), and Cross-Gated Attention (CGA). These modules are designed to enhance local-global feature interaction, preserve spatial details, and improve the quality of skip connections. MambaLiteUNet achieves an average IoU of 87.12% and average Dice score of 93.09% across ISIC2017, ISIC2018, HAM10000, and PH2 benchmarks, outperforming state-of-the-art models. Compared to U-Net, our model improves average IoU and Dice by 7.72 and 4.61 points, respectively, while reducing parameters by 93.6% and GFLOPs by 97.6%. Additionally, in domain generalization with six unseen lesion categories, MambaLiteUNet achieves 77.61% IoU and 87.23% Dice, performing best among all evaluated models. Our extensive experiments demonstrate that MambaLiteUNet achieves a strong balance between accuracy and efficiency, making it a competitive and practical solution for dermatological image segmentation. Our code is publicly available at: https://github.com/maklachur/MambaLiteUNet.

  PublisherOA PDF

• MambaLiteUNet: Cross-Gated Adaptive Feature Fusion for Robust Skin Lesion Segmentation

  arXiv (Cornell University) · 2026-04-22

  preprintOpen accessSenior author

  Recent segmentation models have demonstrated promising efficiency by aggressively reducing parameter counts and computational complexity. However, these models often struggle to accurately delineate fine lesion boundaries and texture patterns essential for early skin cancer diagnosis and treatment planning. In this paper, we propose MambaLiteUNet, a compact yet robust segmentation framework that integrates Mamba state space modeling into a U-Net architecture, along with three key modules: Adaptive Multi-Branch Mamba Feature Fusion (AMF), Local-Global Feature Mixing (LGFM), and Cross-Gated Attention (CGA). These modules are designed to enhance local-global feature interaction, preserve spatial details, and improve the quality of skip connections. MambaLiteUNet achieves an average IoU of 87.12% and average Dice score of 93.09% across ISIC2017, ISIC2018, HAM10000, and PH2 benchmarks, outperforming state-of-the-art models. Compared to U-Net, our model improves average IoU and Dice by 7.72 and 4.61 points, respectively, while reducing parameters by 93.6% and GFLOPs by 97.6%. Additionally, in domain generalization with six unseen lesion categories, MambaLiteUNet achieves 77.61% IoU and 87.23% Dice, performing best among all evaluated models. Our extensive experiments demonstrate that MambaLiteUNet achieves a strong balance between accuracy and efficiency, making it a competitive and practical solution for dermatological image segmentation. Our code is publicly available at: https://github.com/maklachur/MambaLiteUNet.

  PublisherDOI

• SKETCH ANALYSIS AND ALGORITHMIC GRADING OF REY-OSTERRIETH EXAMINATIONS

  2026-04-03

  articleOpen accessSenior author

  The Rey-Osterrieth Complex Figure Test (ROCF) is among the most widely used neuropsychological examinations to analyze visual spatial constructional ability and memory skills, but grading the patient's sketched complex figure is subjective in nature and can be time consuming.With increasing demand for tools to help detect cognitive decline, there is a need to leverage sketch recognition research to assist in detecting fine details within an ROCF's inherently abstract figure.We present a series of recognition algorithms to detect all 18 official ROCF details using a top-down sub-shape recognition approach.This automated grader transforms a sketch into an undirected graph, identifies and isolates detail sub-shapes, and validates sub-shape neatness via a point-density matrix template matcher.Experimental results from hand-drawn ROCFs confirm that our approach can automatically grade ROCF Tests on the same 18-item sketch detail checklist used by neuropsychologists with marginal error margin.

  PublisherOA PDFDOI

• DependencyAI: Detecting AI Generated Text through Dependency Parsing

  arXiv (Cornell University) · 2026-02-17

  articleOpen accessSenior author

  As large language models (LLMs) become increasingly prevalent, reliable methods for detecting AI-generated text are critical for mitigating potential risks. We introduce DependencyAI, a simple and interpretable approach for detecting AI-generated text using only the labels of linguistic dependency relations. Our method achieves competitive performance across monolingual, multi-generator, and multilingual settings. To increase interpretability, we analyze feature importance to reveal syntactic structures that distinguish AI-generated from human-written text. We also observe a systematic overprediction of certain models on unseen domains, suggesting that generator-specific writing styles may affect cross-domain generalization. Overall, our results demonstrate that dependency relations alone provide a robust signal for AI-generated text detection, establishing DependencyAI as a strong linguistically grounded, interpretable, and non-neural network baseline.

  PublisherOA PDF

• WIP: Design and Usability of an Interactive Educational Application for Teaching Undergraduate Sedimentology

  2025-11-02

  article

  This work-in-progress, innovative research paper presents the design of SedimentSketch, an application designed to facilitate the teaching of sedimentology concepts. Effectively teaching these concepts often requires specialized equipment, hands-on experience with sedimentological samples, and personalized instruction. SedimentSketch was designed to complement in-class instruction on sedimentology. SedimentSketch embodies active learning principles, allowing students to interact meaningfully with the material. Aligned with existing research, SedimentSketch seeks to engage students with science concepts through educational videos, games, and exercises, designed to mimic geologists' real-world tasks. In this paper, we evaluated the usability of SedimentSketch. Specifically, we addressed the research question: What are the perceptions of geoscience experts regarding SedimentSketch? Four geoscience experts and instructors were interviewed about their experience using SedimentSketch. A qualitative thematic analysis of participant feedback revealed perceptions regarding the application's ease of use, novelty, and relevance to geoscience instruction. Participants also identified areas of improvement and potential enhancements. These preliminary findings will inform future development and help improve the design. Future work will involve testing the application in real classroom settings and evaluating its impact on student outcomes.

  PublisherDOI

• Assessment of student retention and satisfaction in computer science service courses when using competency-based grading and assignment choice

  Computers in education journal · 2025-12-01

  articleSenior author

  Introductory engineering courses that include computer science components—particularly for non-computer science majors—often provoke anxiety, especially among students with limited or no prior programming experience. This apprehension is exacerbated by the presence of peers who possess advanced coding backgrounds. To address these challenges, this study explores the implementation of assignment choice and competency-based grading in a Computer Science I (CS-1) service course. Grounded in Self-Determination Theory, the intervention allows students to select from multiple assignment pathways, each aligned with the same core learning objectives. This autonomy is intended to promote student agency, intrinsic motivation, and engagement while still ensuring mastery of fundamental programming concepts. A mixed-methods design compares two parallel course sections: one using traditional instructional methods and another offering assignment choice and competency-based assessment. Quantitative data includes standard performance metrics, such as D, F, and withdrawal (DFQ) rates, while qualitative data is drawn from student feedback and course evaluations. Results indicate that students in the assignment choice section report increased satisfaction, deeper engagement, and a stronger sense of ownership over their learning. The section also showed a reduced DFQ rate, suggesting improved student retention. The approach was especially effective for non-CS majors, many of whom felt more confident and capable navigating the course content at their own pace and with personalized relevance. This study contributes a flexible instructional framework that preserves rigor while promoting student-centered learning. The findings support the use of assignment choice and competency-based grading as strategies to improve student outcomes in engineering service courses involving programming. Implications for scaling and adaptation to other technical domains are also discussed.

  PublisherDOI

• Advancing Micropaleontology Education: Study Design and Evaluation of FossilSketch application, Student and Instructor Interfaces

  Abstracts with programs - Geological Society of America · 2025-01-01

  articleSenior author
  PublisherDOI

Recent grants

• Enhancing Visualization Skills and Conceptual Understanding Using a Drawing-Recognition Tutoring System for Engineering Students

  NSF · $989k · 2017–2023

• Collaborative Research: Enabling Instructors to Teach Statics Actively

  NSF · $541k · 2011–2017

• SCC-PG: Fostering Aging-in-Place and Autonomy in Elderly Persons through Intelligent Tracking

  NSF · $150k · 2020–2022

• EAGER: Exploring Children's Use of Online Social Networks Using the KidGab Network

  NSF · $192k · 2016–2019

• EXP: Collaborative Research: PerSketchTivity- Empowering and Inspiring Creative, Competent, Communicative, and Effective Engineers through Perspective Sketching

  NSF · $433k · 2014–2019

Frequent coauthors

• Julie Linsey

  Massachusetts Institute of Technology

  114 shared

• Lance White

  Texas A&M University

  93 shared

• Samantha Ray

  Texas A&M University

  87 shared

• Seth Polsley

  University of Nebraska–Lincoln

  78 shared

• Donna Jaison

  Texas A&M University

  65 shared

• Paul Taele

  Texas A&M University

  60 shared

• Hillary Merzdorf

  Texas A&M University

  58 shared

• Larry Powell

  Institute for Learning Innovation

  47 shared

Labs

• Computer Science & EngineeringPI

Education

• Ph.D. in Electrical Engineering and Computer Science

  Massachusetts Institute of Technology

  2007

• F.T.O. in Finance Technology

  Massachusetts Institute of Technology

  2004

• M.A. in Anthropology

  Columbia University

  2001

• M.S. in Computer Science

  Columbia University

  2000

• B.S. in Applied Physics and Applied Mathematics

  Columbia University

  1997

• B.A. in Mathematics

  Columbia University Columbia College

  1997

Awards & honors

• Best Paper Case Study Honorable Mention, CHI 2013
• IAAIA Innovative Applications of AI Deployed Application Awa…
• College of Engineering Faculty Fellow, 2012
• Best Paper Nomination, SBIM 2010
• Scholarship Recipient: TAMU Assessment Conference, 2008