arXiv subjects

Charles R. Clark

Publications and source records attributed to Charles R. Clark.

4 recordsLinked to original sources

arXiv2026

DS@GT ARC at MEDIQA-CORE-Task-1 2026: Trimodal Model Fusion with Task-Specific Gates for Brain Tumor Subtype Classification

Brain tumor diagnosis is a time-sensitive process in which patients may wait weeks for a finalized pathology report. This problem motivates automated systems that classify tumor subtype from multimodal inputs. This paper details the DS@GT ARC team's work for ImageCLEFmed MEDIQA-CORE 2026 Task~1, Brain Tumor Subtype Classification. The task evaluates three glioma classification problems: Level-1 Molecular Type, LGG vs HGG, and WHO Grade. We combine pre-extracted MRI (NeuroVFM) and histopathology (Prov-GigaPath) embeddings with free-text radiology reports. Our team explored two trimodal fusion architectures, two report encoders (RadBERT and Llama-3.1-8B-Instruct), and a biologically motivated post-processing stage. We achieve a mean macro-F1 of 0.801 under the Fully Multimodal condition, exceeding the organizers' baseline of 0.796 and ranking second among the teams whose code passed verification. Additional evaluation across modality-dropping conditions shows that this advantage depends heavily on the availability of the histopathology modality, and that our system falls behind the baseline when modalities are missing. Our code is available on GitHub at https://github.com/dsgt-arc/imageclef-mediqacore-2026.

cs.CV↗

arXiv2025

DS@GT AnimalCLEF: Triplet Learning over ViT Manifolds with Nearest Neighbor Classification for Animal Re-identification

This paper details the DS@GT team's entry for the AnimalCLEF 2025 re-identification challenge. Our key finding is that the effectiveness of post-hoc metric learning is highly contingent on the initial quality and domain-specificity of the backbone embeddings. We compare a general-purpose model (DINOv2) with a domain-specific model (MegaDescriptor) as a backbone. A K-Nearest Neighbor classifier with robust thresholding then identifies known individuals or flags new ones. While a triplet-learning projection head improved the performance of the specialized MegaDescriptor model by 0.13 points, it yielded minimal gains (0.03) for the general-purpose DINOv2 on averaged BAKS and BAUS. We demonstrate that the general-purpose manifold is more difficult to reshape for fine-grained tasks, as evidenced by stagnant validation loss and qualitative visualizations. This work highlights the critical limitations of refining general-purpose features for specialized, limited-data re-ID tasks and underscores the importance of domain-specific pre-training. The implementation for this work is publicly available at github.com/dsgt-arc/animalclef-2025.

cs.CV↗

arXiv2025

FAIR-CS: Framework for Interdisciplinary Research Collaborations in Online Computing Programs

Research experience is crucial for computing master's students pursuing academic and scientific careers, yet online students have traditionally been excluded from these opportunities due to the physical constraints of traditional research environments. This paper presents the Framework for Accelerating Interdisciplinary Research in Computer Science (FAIR-CS), a method for achieving research goals, developing research communities, and supporting high quality mentorship in an online research environment. This method advances virtual research operations by orchestrating dynamic partnerships between master's level researchers and academic mentors, resulting in interdisciplinary publications. We then discuss the implementation of FAIR-CS in the Human-Augmented Analytics Group (HAAG), with researchers from the Georgia Tech's Online Master of Computer Science program. Through documented project records and experiences with 72 active users, we present our lessons learned and evaluate the evolution of FAIR-CS in HAAG. This paper serves as a comprehensive resource for other institutions seeking to establish similar virtual research initiatives, demonstrating how the traditional research lab environment can be effectively replicated in the virtual space while maintaining robust collaborative relationships and supporting knowledge transfer.

cs.CY↗

arXiv2022

JWST/NIRCam Coronagraphy: Commissioning and First On-Sky Results

In a cold and stable space environment, the James Webb Space Telescope (JWST or "Webb") reaches unprecedented sensitivities at wavelengths beyond 2 microns, serving most fields of astrophysics. It also extends the parameter space of high-contrast imaging in the near and mid-infrared. Launched in late 2021, JWST underwent a six month commissioning period. In this contribution we focus on the NIRCam Coronagraphy mode which was declared "science ready" on July 10 2022, the last of the 17 JWST observing modes. Essentially, this mode will allow to detect fainter/redder/colder (less massive for a given age) self-luminous exoplanets as well as other faint astrophysical signal in the vicinity of any bright object (stars or galaxies). Here we describe some of the steps and hurdles the commissioning team went through to achieve excellent performances. Specifically, we focus on the Coronagraphic Suppression Verification activity. We were able to produce firm detections at 3.35$μ$m of the white dwarf companion HD 114174 B which is at a separation of $\simeq$ 0.5" and a contrast of $\simeq$ 10 magnitudes ($10^{4}$ fainter than the K$\sim$5.3 mag host star). We compare these first on-sky images with our latest, most informed and realistic end-to-end simulations through the same pipeline. Additionally we provide information on how we succeeded with the target acquisition with all five NIRCam focal plane masks and their four corresponding wedged Lyot stops.

astro-ph.IM↗