Eric C. Larson

In this work, we generate Cayley graphs to obtain spectra corresponding to orthogonal eigenvectors that efficiently decorrelate multidimensional signals. Optimal signal decorrelation information is commonly obtained from the Karhunen-Loeve transform. However, whereas the traditional KL-transform autocovariance matrix is generated using multiple products between signal components, generating the Cayley graph only requires memory access operations for assigning weights to the graph edges. The graph is generated by coloring the edges between different vertices, where the vertices represent symmetry group elements, and the weight values are accessed from the discrete multidimensional functions. The efficiency from avoiding product calculations allows the use of spectral information for a wide variety of applications, as the sparsity of the spectra indicates the degree to which a function can be maximally decorrelated. For real-world applications, we begin by demonstrating how Cayley graph spectra can be used for one-dimensional signal classification. We extend our theory to multiple dimensions, allowing us to develop higher-dimensional applications. We obtain the Cayley graph representation of multiple two-dimensional image patches, which demonstrates that the spectra at each window patch can be used to match key points between transformed images. As another multidimensional example, we apply the method to a three-dimensional image patch; this is then used for boundary edge detection of three-dimensional scans. Thus, our method is useful for extracting valuable decorrelation information from signals of any number of dimensions, and it can outperform common signal processing techniques for a diverse set of tasks.

Assessing the quality of random bitstreams used to support cryptographic and other applications is crucially important as any detectable deviations from true randomness can introduce exploitable vulnerabilities resulting in a loss of security. Existing randomness tests, such as those recommended by the U.S. National Institute of Standards and Technology (NIST), examine statistical characteristics of bitstreams such as bit distributions, periodicity, cumulative sums and other properties. However, these methods do not explicitly test for generalized causality within a bitstream – instances where earlier sequences influence the likelihood of later sequences that are undetectable through correlation-based analyses. To address this gap, we propose a new approach, the Granger-inspired Test for Randomness (GTR), that applies principles of Granger causality to detect causal relationships within a single bitstream. To validate our approach, we conduct experiments using a 10-million-bit sample acquired from the NIST Randomness Beacon as a baseline case. We compare GTR to other methods that assess random bitstream quality. Our findings suggest that GTR outperforms many randomness tests, identifying subtle structural dependencies in bitstreams that are not detected with many current tests.

We present an in-situ, exploratory multi-case study demonstrating how in-service teachers can engage in participatory design of an educational video game to enhance middle school students’ computational thinking. Through participatory exercises, we explored how in-service teachers think about educational video games that supplement classroom instruction. Data were collected from eight in-service teachers at urban middle schools using interviews, focus groups, and field notes. The data indicated that during conceptualization of game design, 82% of teacher comments were related to learning mechanisms, and only 14% were related to gameplay. Additionally, participating teachers stressed the importance of embedding relevant pedagogical content knowledge in the game design processes, including knowledge of content and students, teaching, and curriculum. This work illustrates that participatory game design with teachers has a potentially significant positive impact, resulting in practical, student-centered mechanics for game design.

This study investigates how natural language processing (NLP) can support the assessment and learning of science vocabulary among multilingual and multicultural learners, drawing on data from two federally funded studies in the United States. Students define and use target vocabulary in a sentence, with responses transcribed and scored using NLP tools. Employing a mixed-methods design and guided by established socioecological theoretical frameworks, we examine how students’ sociocultural contexts and background knowledge influence their understanding of science word knowledge and applicability. Our findings highlight both the potential and challenges of using AI tools in equitable and culturally responsive ways, offering insights to improve NPL-based assessment tools that support literacy teaching and learning in diverse student populations.

Traditional statistical optimization-based state estimation (DSSE) algorithms rely on detailed grid parameters and mathematical assumptions of all possible uncertainties. Furthermore, random data missing due to communication failures, congestion, and cyberattacks, makes these methods easily infeasible. Inspired by recent advances in digital twins (DTs), this paper proposes an interactive attention-based DSSE model for robust grid monitoring by integrating three core components: physical entities, virtual modeling, and data fusion. To enable robustness against various data missing in heterogeneous measurements, we first propose physics-informed data augmentation and transfer. Moreover, a state-of-the-art attention-based spatiotemporal feature learning is proposed, followed by a novel cross-interaction feature fusion for robust voltage estimation. A case study in a real-world unbalanced 84-bus distribution system with raw data validates the accuracy and robustness of the proposed DT model in estimating voltage states, with random locational, arbitrary ratios (up to 40% of total measurements) of data missing.

Since its introduction, softmax attention has become the backbone of modern transformer architectures due to its expressiveness and scalability across a wide range of tasks. However, the main drawback of softmax attention is the quadratic memory requirement and computational complexity with respect to the sequence length. By replacing the softmax nonlinearity, linear attention and similar methods have been introduced to avoid the quadratic bottleneck of softmax attention. Despite these linear forms of attention being derived from the original softmax formulation, they typically lag in terms of downstream accuracy. While strong intuition of the softmax nonlinearity on the query and key inner product suggests that it has desirable properties compared to other nonlinearities, the question of why this discrepancy exists still remains unanswered. This work demonstrates that linear attention is an approximation of softmax attention by deriving the recurrent form of softmax attention. Using this form, each part of softmax attention can be described in the language of recurrent neural networks (RNNs). Describing softmax attention as an RNN allows for the ablation of the components of softmax attention to understand the importance of each part and how they interact. In this way, our work helps explain why softmax attention is more expressive than its counterparts.

Electronic Health Record (EHR) data encompass diverse modalities -- text, images, and medical codes -- that are vital for clinical decision-making. To process these complex data, multimodal AI (MAI) has emerged as a powerful approach for fusing such information. However, most existing MAI models optimize for better prediction performance, potentially reinforcing biases across patient subgroups. Although bias-reduction techniques for multimodal models have been proposed, the individual strengths of each modality and their interplay in both reducing bias and optimizing performance remain underexplored. In this work, we introduce FAME (Fairness-Aware Multimodal Embeddings), a framework that explicitly weights each modality according to its fairness contribution. FAME optimizes both performance and fairness by incorporating a combined loss function. We leverage the Error Distribution Disparity Index (EDDI) to measure fairness across subgroups and propose a sign-agnostic aggregation method to balance fairness across subgroups, ensuring equitable model outcomes. We evaluate FAME with BEHRT and BioClinicalBERT, combining structured and unstructured EHR data, and demonstrate its effectiveness in terms of performance and fairness compared with other baselines across multiple EHR prediction tasks.