Newest Papers

Date: Newest Clear all

Advanced filtersTopic: All

Experimental

Topic

Topics are auto-detected from title, abstract, and metadata and may be imperfect.

Publication Date

Newest Oldest Clear

Average Rating

Highest Lowest

Historically Black College or University Attendance and Cognition in US Black Adults

Marilyn D. Thomas, Carol Wei, Min Hee Kim, Jennifer Manly, Suzanne E. Judd, Justin S. White, Virginia J. Howard, Christina Mangurian, Rita Hamad, M. Maria Glymour

JAMA Network Open·2026

Importance Black adults may derive long-term cognitive benefits from attendance at a historically Black college or university (HBCU) compared with a predominantly White institution (PWI). This association has not been evaluated in a nationwide sample. Objective To estimate the association between HBCU vs PWI attendance and cognitive health among later-life Black adults. Design, Setting, and Participants This cohort study used data from the Reasons for Geographic and Racial Differences in Stroke study, a prospective cohort study that recruited Black and White US adults aged 45 years and older during 2003 to 2007. The national cohort oversampled Black individuals and residents from the Stroke Belt (56%), a group of 8 Southern states defined by excess stroke mortality. The analytic sample included Black participants who attended high school in a state with an HBCU and attended college. Analysis was conducted from February to September 2025. Exposure Participants self-reported each college ever attended, classified as a PWI (reference) or HBCU. Main Outcomes and Measures Assessments of memory, language, and global cognition were conducted during follow-up every 6 months (2006-2021). Cognitive measures were standardized (ie, z -transformed). Linear regression was used to estimate the average treatment effect (ATE) using inverse of probability of treatment weighting. Additional analyses evaluated potential modification by being college-aged before 1955 (during legal racial segregation), 1955 to 1964 (during sanctioned racial discrimination), or after 1964. Results Among 1978 Black college-goers (mean [SD] age at first assessment, 61.8 [8.2] years; 1333 [67.4%] female), 699 (35.3%) attended an HBCU, 1952 (98.7%) completed a memory assessment, 1970 (99.6%) completed a language assessment, and 530 (26.8%) completed both assessments during the same follow-up visit. Compared with PWI attendance, HBCU attendance was associated with better z -scored memory (ATE, 0.13; 95% CI, 0.05-0.21), language (ATE, 0.19; 95% CI, 0.08-0.29), and global cognition (ATE, 0.22; 95% CI, 0.09-0.34). Estimates were consistent for individuals who were college-aged after 1955 but were not statistically significant among respondents who were college-aged prior to 1955. Conclusions and Relevance In this cohort study using a national dataset, HBCU attendance was associated with better cognition compared with PWI attendance for aging Black adults, which held for those attending college before and after legal racial segregation and sanctioned racial discrimination in education.

No ratings yet

View paper →

PreviousPage 3 of 534Next

Why do we do astrophysics?

David W. Hogg

arXiv·2026

At time of writing, large language models (LLMs) are beginning to obtain the ability to design, execute, write up, and referee scientific projects on the data-science side of astrophysics. What implications does this have for our profession? In this white paper, I list - and argue for - a set of facts or "points of agreement" about what astrophysics is, or should be; these include considerations of novelty, people-centrism, trust, and (the lack of) clinical value. I then list and discuss every possible benefit that astrophysics can be seen as bringing to us, and to science, and to universities, and to the world; these include considerations of love, weaponry, and personal (and personnel) development. I conclude with a discussion of two possible (extreme and bad) policy recommendations related to the use of LLMs in astrophysics, dubbed "let-them-cook" and "ban-and-punish." I argue strongly against both of these; it is not going to be easy to develop or adopt good moderate policies.

★ 5.0 (1)

View paper →

Features as Rewards: Scalable Supervision for Open-Ended Tasks via Interpretability

Aaditya Vikram Prasad, Connor Watts, Jack Merullo, Dhruvil Gala, Owen Lewis, Thomas McGrath, Ekdeep Singh Lubana

arXiv·2026

Language models trained on large-scale datasets have been shown to learn features that encode abstract concepts such as factuality or intent. Such features are traditionally used for test-time monitoring or steering. We present an alternative affordance: features as scalable supervision for open-ended tasks. We consider the case of hallucination-reduction as a desirable, yet open-ended behavior and design a reinforcement learning (RL) pipeline, titled RLFR (Reinforcement Learning from Feature Rewards), that uses features as reward functions. Grounded in a novel probing framework that identifies candidate hallucinated claims, our pipeline teaches a model to intervene and correct its completions when it is uncertain of their factuality. Furthermore, the pipeline enables scalable test-time compute, guided once more by our reward features. This end-to-end process operationalized on Gemma-3-12B-IT results in a policy that is 58% less likely to hallucinate compared to the original model (when run in tandem with our probing harness), while preserving performance on standard benchmarks. Taken together, by grounding supervision in the language of features, this paper introduces a novel paradigm in the use of interpretability for learning open-ended tasks.

No ratings yet

View paper →

Do Neural Networks Lose Plasticity in a Gradually Changing World?

Tianhui Liu, Lili Mou

arXiv·2026

Continual learning has become a trending topic in machine learning. Recent studies have discovered an interesting phenomenon called loss of plasticity, referring to neural networks gradually losing the ability to learn new tasks. However, existing plasticity research largely relies on contrived settings with abrupt task transitions, which often do not reflect real-world environments. In this paper, we propose to investigate a gradually changing environment, and we simulate this by input/output interpolation and task sampling. We perform theoretical and empirical analysis, showing that the loss of plasticity is an artifact of abrupt tasks changes in the environment and can be largely mitigated if the world changes gradually.

No ratings yet

View paper →

Learning to Drive in New Cities Without Human Demonstrations

Zilin Wang, Saeed Rahmani, Daphne Cornelisse, Bidipta Sarkar, Alexander David Goldie, Jakob Nicolaus Foerster, Shimon Whiteson

arXiv·2026

While autonomous vehicles have achieved reliable performance within specific operating regions, their deployment to new cities remains costly and slow. A key bottleneck is the need to collect many human demonstration trajectories when adapting driving policies to new cities that differ from those seen in training in terms of road geometry, traffic rules, and interaction patterns. In this paper, we show that self-play multi-agent reinforcement learning can adapt a driving policy to a substantially different target city using only the map and meta-information, without requiring any human demonstrations from that city. We introduce NO data Map-based self-play for Autonomous Driving (NOMAD), which enables policy adaptation in a simulator constructed based on the target-city map. Using a simple reward function, NOMAD substantially improves both task success rate and trajectory realism in target cities, demonstrating an effective and scalable alternative to data-intensive city-transfer methods. Project Page: https://nomaddrive.github.io/

No ratings yet

View paper →

Empowerment and individualization: online banking and household financial organization

Yang Hu, Yue Qian

European Sociological Review·2026

Abstract Bringing economic and family sociology into the digital era, this study examines and highlights the transformative potential of online banking in reconfiguring household financial organization. Using an instrumental-variable approach to analyzing data from the UK Household Longitudinal Study (2019–2023), we examine the impact of partners’ online banking use on who manages the couple’s money and controls major financial decisions. We reveal an ‘empowering effect’ whereby one’s use of online banking bolsters their control of both money management and major financial decisions, and this effect is particularly pronounced when one’s partner does not use online banking. Online banking similarly bolsters the power of women and men in money management, but subtle gender differences are observed in financial decision-making. Whereas online banking increases men’s exclusive control of major financial decisions, it bolsters women’s exclusive or joint control of such decisions. We also find an ‘individualizing effect’ whereby both women’s and men’s use of online banking increases the likelihood of partners’ independent money management, and independent money management is least likely when only one partner uses online banking. Our study provides novel insights into how digitalization, seen through online banking, redistributes power in and reconfigures the nature of couple relationships.

No ratings yet

View paper →

The Laplacian Keyboard: Beyond the Linear Span

Siddarth Chandrasekar, Marlos C. Machado

arXiv·2026

Across scientific disciplines, Laplacian eigenvectors serve as a fundamental basis for simplifying complex systems, from signal processing to quantum mechanics. In reinforcement learning (RL), these eigenvectors provide a natural basis for approximating reward functions; however, their use is typically limited to their linear span, which restricts expressivity in complex environments. We introduce the Laplacian Keyboard (LK), a hierarchical framework that goes beyond the linear span. LK constructs a task-agnostic library of options from these eigenvectors, forming a behavior basis guaranteed to contain the optimal policy for any reward within the linear span. A meta-policy learns to stitch these options dynamically, enabling efficient learning of policies outside the original linear constraints. We establish theoretical bounds on zero-shot approximation error and demonstrate empirically that LK surpasses zero-shot solutions while achieving improved sample efficiency compared to standard RL methods.

No ratings yet

View paper →

Action-to-Action Flow Matching

Jindou Jia, Gen Li, Xiangyu Chen, Tuo An, Yuxuan Hu, Jingliang Li, Xinying Guo, Jianfei Yang

arXiv·2026

Diffusion-based policies have recently achieved remarkable success in robotics by formulating action prediction as a conditional denoising process. However, the standard practice of sampling from random Gaussian noise often requires multiple iterative steps to produce clean actions, leading to high inference latency that incurs a major bottleneck for real-time control. In this paper, we challenge the necessity of uninformed noise sampling and propose Action-to-Action flow matching (A2A), a novel policy paradigm that shifts from random sampling to initialization informed by the previous action. Unlike existing methods that treat proprioceptive action feedback as static conditions, A2A leverages historical proprioceptive sequences, embedding them into a high-dimensional latent space as the starting point for action generation. This design bypasses costly iterative denoising while effectively capturing the robot's physical dynamics and temporal continuity. Extensive experiments demonstrate that A2A exhibits high training efficiency, fast inference speed, and improved generalization. Notably, A2A enables high-quality action generation in as few as a single inference step (0.56 ms latency), and exhibits superior robustness to visual perturbations and enhanced generalization to unseen configurations. Lastly, we also extend A2A to video generation, demonstrating its broader versatility in temporal modeling. Project site: https://lorenzo-0-0.github.io/A2A_Flow_Matching.

No ratings yet

View paper →

ICT Use and Mental Health Among Asian American Older Adults: Does Independent Learning Matter?

Jinyu Liu, Yifan Lou, Ethan Siu Leung Cheung

Research on Aging·2026

This study explores the relationships between ICT uses for various purposes and mental health and the moderating effect of technology learning among Chinese and Korean older Americans. The survey data were collected from 513 Chinese and Korean older Americans in New York City. Negative binomial regression was used to examine the relationships between ICT uses for social, instrumental, and health purposes, and depressive symptoms. Interaction terms between ICT use variables and technology learning styles were created for the moderation test. We found that ICT uses for social contact and health purposes were significantly associated with lower depressive symptoms among the Chinese and Korean older Americans, and the benefits were more pronounced for those who learned ICT independently. This study highlighted the importance of supporting ICT uses for social contact and health purposes and technology learning process to improve mental health of older Asian Americans.

No ratings yet

View paper →

Newest Papers

Historically Black College or University Attendance and Cognition in US Black Adults

Why do we do astrophysics?

Features as Rewards: Scalable Supervision for Open-Ended Tasks via Interpretability

What can I do if my idea has been plagiarized?

Do Neural Networks Lose Plasticity in a Gradually Changing World?

Learning to Drive in New Cities Without Human Demonstrations

Empowerment and individualization: online banking and household financial organization

The Laplacian Keyboard: Beyond the Linear Span

Experts Have World Models. LLMs Have Word Models.

Action-to-Action Flow Matching

ICT Use and Mental Health Among Asian American Older Adults: Does Independent Learning Matter?

Killing your Pretty Little Darlings - Felix Eshiet