Walter J. Scheirer’s research while affiliated with Université Notre Dame d'Haïti and other places

Small Uncrewed Aerial Systems (sUAS) are increasingly deployed as autonomous swarms in search-and-rescue and other disaster-response scenarios. In these settings, they use computer vision (CV) to detect objects of interest and autonomously adapt their missions. However, traditional CV systems often struggle to recognize unfamiliar objects in open-world environments or to infer their relevance for mission planning. To address this, we incorporate large language models (LLMs) to reason about detected objects and their implications. While LLMs can offer valuable insights, they are also prone to hallucinations and may produce incorrect, misleading, or unsafe recommendations. To ensure safe and sensible decision-making under uncertainty, high-level decisions must be governed by cognitive guardrails. This article presents the design, simulation, and real-world integration of these guardrails for sUAS swarms in search-and-rescue missions.

Meaningful progress has been made in open world learning (OWL), enhancing the ability of agents to detect, characterize, and incrementally learn novelty in dynamic environments. However, novelty remains a persistent challenge for agents relying on state‐of‐the‐art learning algorithms. This article considers the current state of OWL, drawing on insights from a recent DARPA research program on this topic. We identify open issues that impede further advancements spanning theory, design, and evaluation. In particular, we emphasize the challenges posed by dynamic scenarios that are crucial to understand for ensuring the viability of agents designed for real‐world environments. The article provides suggestions for setting a new research agenda that effectively addresses these open issues.

Over the past few decades, machine learning has made remarkable strides, owed largely to algorithmic advancements and the abundance of high-quality, large-scale datasets. However, an equally crucial aspect in achieving optimal model performance is the fine-tuning of hyperparameters. Despite its significance, hyperparameter optimization (HPO) remains challenging due to several factors. Many existing HPO techniques rely on simplistic search methods or assume smooth and continuous loss functions, which may not always hold true. Traditional methods like grid search and Bayesian optimization often struggle to adapt swiftly and efficiently navigate the loss landscape. Moreover, the search space for HPO is frequently high-dimensional and non-convex, posing challenges in efficiently finding a global minimum. Additionally, optimal hyperparameters can vary significantly based on the dataset or task at hand, further complicating the optimization process. To address these challenges, this paper presents HomOpt, an advanced HPO methodology that integrates a surrogate model framework with homotopy optimization techniques. Unlike rigid methodologies, HomOpt offers flexibility by incorporating diverse surrogate models tailored to specific optimization tasks. Our initial investigation focuses on leveraging Generalized Additive Model (GAM) surrogates within the HomOpt framework to enhance the effectiveness of existing optimization methodologies. HomOpt's ability to expedite convergence towards optimal solutions across varied domain spaces, encompassing continuous, discrete, and categorical domains is highlighted. We conduct a comparative analysis of HomOpt applied to multiple optimization techniques (e.g., Random Search, TPE, Bayes, and SMAC), demonstrating improved objective performance on numerous standardized machine learning benchmarks and challenging open-set recognition tasks. We also integrate CatBoost within the HomOpt framework as a surrogate, showcasing its adaptability and effectiveness in handling more complex datasets. This integration facilitates an evaluation against state-of-the-art methods such as BOHB, particularly on challenging computer vision datasets like CIFAR-10 and ImageNet. Comparative analyses reveal HomOpt's competitive performance with reduced iterations and underscore potential optimizations in execution time. All the experimentation and method code can be found here: https://github.com/sabraha2/HOMOPT

HCI is increasingly taking inspiration from philosophical and religious traditions as a basis for ethical technology designs. If these values are to be incorporated into real-world designs, there may be challenges when designers work with values unfamiliar to them. Therefore, we investigate the variance in interpretations when values are translated to technology designs. To do so we identified social media designs that embodied the main principles of Catholic Social Teaching (CST). We then interviewed 24 technology experts with varying levels of familiarity with CST to assess how their understanding of how those values would manifest in a technology design. We found that familiarity with CST did not impact participant responses: there were clear patterns in how all participant responses differed from the values we determined the designs embodied. We propose that value experts be included in the design process to more effectively create designs that embody particular values.

In Natural Language Processing (NLP), semantic matching algorithms have traditionally relied on the feature of word co-occurrence to measure semantic similarity. While this feature approach has proven valuable in many contexts, its simplistic nature limits its analytical and explanatory power when used to understand literary texts. To address these limitations, we propose a more transparent approach that makes use of story structure and related elements. Using a BERT language model pipeline, we label prose and epic poetry with story element labels and perform semantic matching by only considering these labels as features. This new method, Story Grammar Semantic Matching, guides literary scholars to allusions and other semantic similarities across texts in a way that allows for characterizing patterns and literary technique.

Face parsing is a fundamental task in computer vision, enabling applications such as identity verification, facial editing, and controllable image synthesis. However, existing face parsing models often lack fairness and robustness, leading to biased segmentation across demographic groups and errors under occlusions, noise, and domain shifts. These limitations affect downstream face synthesis, where segmentation biases can degrade generative model outputs. We propose a multi-objective learning framework that optimizes accuracy, fairness, and robustness in face parsing. Our approach introduces a homotopy-based loss function that dynamically adjusts the importance of these objectives during training. To evaluate its impact, we compare multi-objective and single-objective U-Net models in a GAN-based face synthesis pipeline (Pix2PixHD). Our results show that fairness-aware and robust segmentation improves photorealism and consistency in face generation. Additionally, we conduct preliminary experiments using ControlNet, a structured conditioning model for diffusion-based synthesis, to explore how segmentation quality influences guided image generation. Our findings demonstrate that multi-objective face parsing improves demographic consistency and robustness, leading to higher-quality GAN-based synthesis.