Background/Purpose: College students face rising rates of anxiety, depression, and stress, worsened by academic pressure, social isolation, and inequities in care access. Despite campus services, many remain underserved due to stigma, fragmented support, and limited community integration. Goals: This study explores barriers to mental health care for Sacramento County college students and identifies community-informed strategies to improve access, engagement, and equity. Methods: Using a participatory qualitative approach, semi-structured interviews were conducted with 22 stakeholders, including campus counselors (n=6), peer mentors (n=6), faculty advisors (n=5), and nonprofit coordinators (n=5). Interviews examined challenges, service gaps, and potential interventions and were thematically analyzed. Results: Stakeholders reported fragmented referrals, low resource awareness, and stigma. Peer mentors noted preference for informal support; faculty cited inconsistent training. Structural barriers, such as cost, availability, transportation, further limited access. Suggested strategies included embedding mental health support in academic and social structures, expanding peer-led programs, and improving campus-community coordination. Conclusions: Disparities reflect structural, social, and cultural factors. Community-informed, integrated, peer-supported interventions can reduce barriers and enhance equity. Context-specific programs leveraging campus-community networks are critical to improving access and reducing stigma.

Modern vision–language models (VLMs) are artificial neural networks capable of complex mathematical reasoning, yet they lack the ability to robustly count elements within an image. Their apparent “number sense” emerges from visual feature patterns in training data that covary with numerosity, such as density and surface area. These approximations collapse unpredictably and are systematically different than human enumeration, marking a systematic difference in numerical reasoning of biological and artificial intelligences. Which numerical strategies do VLMs use, and why do shortcuts cause systematic failures? We adapted empirical paradigms from animal psychology to study numerosity in VLMs. Stimuli containing 1–111 elements were engineered in either a Numerosity-Only experimental condition or five control conditions which systematically introduced covariates of number––grid alignment, size uniformity, surface area, density, and spread. ChatGPT-4.1 was then tasked with counting the number of elements in 17,982 stimuli. Results demonstrated condition-specific rates of exponential decay in accuracy with numeric increase, along with differential patterns in accuracy, bias, and entropy. These findings motivated an attractor hypothesis, in which responses trade off between numerical resolution and collapse toward attractor numbers. Our experiment establishes a cross-species framework for evaluating number representations in AI with implications for cognitive alignment.

Major depressive disorder (MDD) and generalized anxiety disorder (GAD) are two of the most common mental health conditions. Previous studies have associated lower heart rate variability (HRV) with both anxiety and depression, but nearly all studies have utilized research-grade instruments in controlled laboratory conditions, which potentially limits the generalizability of these findings to real-world patient settings. Consumer wearable devices are widely available and can potentially enable entirely automated, longitudinal collection of physiological signals related to mental health. After providing informed consent, 4470 participants ranging from healthy to severe depression wore Apple Watch for up to 12 months and completed self-reported depression (Patient Health Questionnaire-8; PHQ-8) and anxiety (Generalized Anxiety Disorder Scale-7; GAD-7) questionnaires every 14 days. Using mixed effects linear regression, we identified significant negative associations between mean HRV and PHQ-8 and GAD-7 scores (βPHQ-8 = -0.084; p < 0.001, βGAD-7 = -0.087; p < 0.001). The results validate previous findings that lower average activity of the parasympathetic nervous system is associated with anxiety and depression. Further, they demonstrate that consumer wearable devices can contribute to the understanding of the physiological components of depression and anxiety.

This abstract has been withheld from publication.