MD-SOAR Maryland Shared Open Access Repository
Not a member yet
34521 research outputs found
Sort by
Hierarchical Learning for Maze Navigation: Emergence of Mental Representations via Second-Order Learning
No full textMental representation, characterized by structured internal models mirroring external environments, is fundamental to advanced cognition but remains challenging to investigate empirically. Existing theory hypothesizes that second-order learning -- learning mechanisms that adapt first-order learning (i.e., learning about the task/domain) -- promotes the emergence of such environment-cognition isomorphism. In this paper, we empirically validate this hypothesis by proposing a hierarchical architecture comprising a Graph Convolutional Network (GCN) as a first-order learner and an MLP controller as a second-order learner. The GCN directly maps node-level features to predictions of optimal navigation paths, while the MLP dynamically adapts the GCN's parameters when confronting structurally novel maze environments. We demonstrate that second-order learning is particularly effective when the cognitive system develops an internal mental map structurally isomorphic to the environment. Quantitative and qualitative results highlight significant performance improvements and robust generalization on unseen maze tasks, providing empirical support for the pivotal role of structured mental representations in maximizing the effectiveness of second-order learning.http://arxiv.org/abs/2509.1419
Can Kink Lead to a Happier Life?
No full textOnce considered a mental disorder, today kink is having a moment. Recent movies like Babygirl tell tales of a high-powered woman exploring the world of dominance and submission. In the FX/Hulu series Dying for Sex, based on a true story, a woman who has terminal cancer is introduced to kink with different men. And then, of course, we have 2011’s hugely popular romance trilogy and movie franchise Fifty Shades of Grey. As a result, the once-fringe practice is slowly becoming normalized.https://greatergood.berkeley.edu/article/item/can_kink_lead_to_a_happier_lif
DC Mayor v AG, New Housing Targets, MDOT Capital Budget, False Viral Story
No full textWashington DC Mayor Muriel Bowser and Attorney General Brian Schwalb have responded differently to the Trump takeover. MD Governor Wes Moore orders creation of new county level housing targets and opening of state land to housing. MD Department of Transportation issues its new 2026 capital budget plan. In Montgomery County, a viral story about ICE turns out to be false. And more. Music by Dear Daria.https://open.spotify.com/episode/1MgN0dnWOCwpLJwdvhtOT
Network Traffic Classification Using Machine Learning, Transformer, and Large Language Models
No full text2025 IEEE 4th International Conference on Computing and Machine Intelligence (ICMI), April 5-6, 2025, Mount Pleasant, MI, USAThis study uses various models to address network traffic classification, categorizing traffic into web, browsing, IPSec, backup, and email. We collected a comprehensive dataset from Arbor Edge Defender (AED) devices, comprising of 30,959 observations and 19 features. Multiple models were evaluated, including Naive Bayes, Decision Tree, Random Forest, Gradient Boosting, XGBoost, Deep Neural Networks (DNN), Transformer, and two Large Language Models (LLMs) including GPT-4o and Gemini with zero- and few-shot learning. Transformer and XGBoost showed the best performance, achieving the highest accuracy of 98.95 and 97.56%, respectively. GPT-4o and Gemini showed promising results with few-shot learning, improving accuracy significantly from initial zero-shot performance. While Gemini Few-Shot and GPT-4o Few-Shot performed well in categories like Web and Email, misclassifications occurred in more complex categories like IPSec and Backup. The study highlights the importance of model selection, fine-tuning, and the balance between training data size and model complexity for achieving reliable classification results.https://ieeexplore.ieee.org/document/1114120
Lessons from the Loan Pause: More Evidence that Student Debt is Reducing Marriage and Childbearing
No full textA brief report prepared by Arielle Kuperberg, Daniel Collier, Joan Maya Mazelis, and Fenaba Addo for the Council on Contemporary Families symposium Policies Affecting Families: What We Know, and What to Expect in the Second Trump Term
In Vice President Vance’s first public address after taking office in January he laid out his priorities, stating “I want more babies in America.” Our research suggests that one way the second Trump administration can achieve this goal is by addressing the student loan crisis among young adults.
When the COVID pandemic hit in 2020, the Trump administration quickly moved to freeze loan payments and interest for public student loans. The loan freeze ultimately lasted until September 2023, with payments resuming in October 2023, giving borrowers more than a three-year break in paying off their loans. This loan pause – and the resumption of payments – has profoundly affected the nearly 43 million Americans with federal student loan debt.
Our research has examined how student loans have affected American borrowers, and how their behavior changed during and after the loan pause. We have uncovered further evidence that studentloan debt is leading to delays in family formation for many young adults, contributing to the record low childbearing rates and the record high typical age at first marriage seen in the United States today.This material is based on work supported by the National Science Foundation under grants no. 1947603 and 1947604. We thank Kalvin Benfield, Katherine Fredricks, Anurag Pant, and Jairo Rodriguez Bustamante for their research assistance.https://contemporaryfamilies.utah.edu/publications/posts/2025/march/family-policy-symposium-kuperberg-collier-mazelis-addo-student-loans.ph
Identification of the Required and Sufficient Carbohydrate-Active Enzymes (CAZymes) for Cellulose Deconstruction in Cellvibrio japonicus
No full textCellvibrio japonicus is a Gram-negative saprophytic bacterium proficient at utilizing a broad range of plant and animal polysaccharides. Previous work established that C. japonicus uses a suite of Carbohydrate-Active enZymes (CAZymes) for the efficient deconstruction of recalcitrant substrates, such as cellulose. Current genome annotation suggests that over 20 genes are predicted to encode CAZymes for cellulose degradation, with previous work suggesting that only a subset of these genes are essential. Our current work combined transcriptomic analysis, mutant strain generation coupled with growth phenotyping, and heterologous expression studies to identify the genes that encode the essential cellulose-specific CAZymes for growth using cellulose. Specifically, we found six CAZyme-encoding required for cellulose deconstruction in C. japonicus. These six gens are cel3B, cel5B, cel6A, lpmo10B, cbp2D, and cbp2E, which encode a βglucosidase, an endoglucanase, an exoglucanase, a lytic polysaccharide monooxygenase, and two carbohydrate-binding proteins, respectively. Interestingly, we found that while these CAZyme-encoding genes are necessary for growth using cellulose by C. japonicus, they are not sufficient when heterologously expressed in Escherichia coli. When expressing these genes in E. coli, we observed that E. coli was only able to utilize cellooligosaccharides and soluble forms of cellulose, but not insoluble cellulose. Additionally, we obtained evidence that suggests C. japonicus rapidly uptakes cello-oligosaccharides generated during the deconstruction of cellulose. Using a combination of thin-layer chromatography and linked enzyme assays, we observed that deconstruction of cellulose and uptake of derivative cellodextrins occurs minimally at similar rates. Our revised model of cellulose utilization by C. japonicus can be applied more broadly to ecological studies probing bacterial cellulose degradation in the context of global carbon cycling, as well as provide insights to improve industrial enzymatic conversion of lignocellulosic biomass for renewable fuels and chemicals
Translation with LLMs through Prompting with Long-Form Context
No full textACL 2023, 61st Annual Meeting of the Association for Computational Linguistics, July 9th - 14th, 2023, Toronto, CanadaStable generation of text in low-resource languages is an unsolved issue in large language models. While Large Language Models (LLMs) can often produce good translations despite not being explicitly trained for this task, this does not hold for low-resource languages. LLMs are both more likely to generate off-target text (text in another language than intended) when prompted to translate to a low-resource language, and show increased instability in translation quality across prompt templates in low-resource languages. This study implemented a prepended monolingual text prompting method in the target language and used a context-and topic-aware with few-shot machine translation (MT). We quantified these methods for low-, mid-, and high-resource languages using OpenAI GPT-4o-mini and Google Gemini-1.5-flash. Gemini results showed that the use of contexttopic-aware with few-shot MT (CTAFSMT) significantly boosted the performance for the three language categories. However, this was not consistently observed in the case of ChatGPT. It was found that the significance of the results depended on the language itself rather than the level of its resources. This study is part of Stanford University's meta-study on whether LLMs can generate novel research ideas. The code, prompts, and results of the study can be found at https://github.com/HuthaifaAshqar/Translationwith-LLMs.We would like to acknowledge that the original idea was submitted by Elizabeth Salesky from Google DeepMind for the purpose of this study, which is part of the funded Stanford University’s meta-study on whether LLMs can generate novel research ideas.https://www.authorea.com/users/884736/articles/1318311-translation-with-llms-through-prompting-with-long-form-contex
Frequency-Aware Mixture of Experts Model for Robust Multimodal Perception
No full textRobust multimodal perception is essential to understand real-world scenes, particularly under degraded, noisy, or low-visibility conditions. This dissertation introduces a Frequency-Aware Mixture-of-Experts model that combines structural features from the frequency domain with semantic and spatial representations in RGB, infrared (IR), text, and audio modalities. The work advances through a progression of perception tasks, beginning with single-modality perception, extending to vision-language modeling, and culminating in a four-modality adaptive model. We begin by addressing domain-specific perception using single-modality visual learning, which highlights the limitations of relying on a single source of information in complex environments. This motivates the integration of frequencydomain reasoning into multimodal architectures. In the next stage, we enhance vision-language modeling by introducing frequency-based low-rank features into pretrained visual encoders. These features provide noise-resilient representations while maintaining compatibility with language models, leading to improved performance in caption generation and visual question answering (VQA), particularly under visual degradation. Finally, we propose a hybrid Frequency-Aware Mixture-of-Experts (FreqMoE) model that dynamically fuses RGB and IR image features, guided by synchronized text and audio signals. A frequency domain gating mechanism that computes reliability scores from log-magnitude spectral features and a feature-wise modulation module that adapts visual features based on fused semantic embeddings. To support this four-modality setup, we extend three public RGB-IR datasets—M3FD, RoadScene, and MSRS—by adding aligned textual and audio annotations. This results in a synchronized four-modality setup that includes RGB images, IR data, captions, and audio, without requiring new data collection. Experimental results demonstrate that our method outperforms state-of-the-art baselines in both detection and fusion quality metrics. Ablation studies further validate the contributions of frequency-aware gating and semantic conditioning. Our approach offers an interpretable and adaptive solution for robust cross-modal perception under real-world constraints
New glacier thickness and bed topography maps for Svalbard
No full textKnowledge of the thickness, volume, and subglacial topography of glaciers is crucial for a range of glaciological, hydrological, and societal issues, including studies on climate-warming-induced glacier retreat and associated sea level rise. This is not in the least true for Svalbard, one of the fastest-warming places in the world. Here, we present new maps of the ice thickness and subglacial topography for every glacier on Svalbard. Using remotely sensed observations of surface height, ice velocity, rate of surface elevation change, and glacier boundaries in combination with a modelled mass balance product, we apply an inverse method that leverages state-of-the-art ice flow models to obtain the shape of the glacier bed. Specifically, we model large glaciers with the Parallel Ice Sheet Model (PISM) at 500 m resolution, while we resolve smaller mountain glaciers at 100 m resolution using the physics-informed deep-learning-based Instructed Glacier Model (IGM). Actively surging glaciers are modelled using a perfect-plasticity model. We find a total glacier volume (excluding the island Kvitøya) of 6800 ± 238 km³, corresponding to 16.3 ± 0.6 mm sea level equivalent. Validation against thickness observations shows high statistical agreement, and the combination of the three methods is found to reduce uncertainties. We discuss the remaining sources of errors, differences from previous ice thickness maps of the region, and future applications of our results.Ward van Pelt received funding from a career grant by the Swedish National Space Agency (2018-C; project no. 189/18) and a starting grant from the Swedish Research Council project no. 2020-04319). Development of PISM is supported by NASA (grant nos. 20-CRYO2020-0052 and 80NSSC22K0274) and NSF (grant no. OAC-2118285). The computations were enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS) at Chalmers partially funded by the Swedish Research Council (grant no. 2022-06725). The publication of this article was funded by the Swedish Research Council, Forte, Formas, and Vinnovahttps://tc.copernicus.org/articles/19/1/2025
How China uses second world war history in its bid to reshape the global order
No full textHistorian Meredith Oyen explains how disagreements over the history of the second world war and who fought the Japanese are central to tensions between China and Taiwan. Listen on The Conversation Weekly podcast.https://theconversation.com/how-china-uses-second-world-war-history-in-its-bid-to-reshape-the-global-order-podcast-26444