SSP Forum: Honors Presentations 2025

The
Symbolic Systems Forum
(community sessions of SYMSYS 280 - Symbolic Systems Research Seminar)
presents

Honors Presentations 2025

Senior Honors Students
Symbolic Systems Program

Monday, June 2, 2025
12:30-2:30pm PT [Note: later ending time]
Margaret Jacks Hall (Bldg. 460), Room 126
In-person event, not recorded (see below for entry instructions if you are not an active Stanford affiliate)

Note: Lunch is provided, if pre-ordered, only for members of SYMSYS 280, but others are welcome to bring a lunch and eat during the presentation.

Schedule:

12:30 Alina Davison, "Influence of Lexical Features on Brain Activation in Dyslexic Readers" (Primary Honors Advisor: Jason Yeatman, Pediatrics and Education and Psychology; Second Reader: Jamie Mitchell, Education)
     Dyslexia is a neurodevelopmental reading disability that is often marked by difficulty with foundational reading skills, such as word decoding and phonological awareness. Neuroimaging studies have demonstrated functional differences between dyslexic and typical readers in key regions of the brain that are involved in reading, including visual processing regions and frontal regions related to semantic processing. This research investigates how specific lexical features like lexicality, string length, and bigram frequency influence neural activity in dyslexic children, and how these patterns may change as the readers complete an intensive reading intervention. Using functional MRI data collected during a lexical decision task, we assess changes across four left-hemispheric regions of the brain: the visual word form areas (VWFA1 and VWFA2), the inferior frontal sulcus (IFS), and the precentral gyrus (PrG). We observe different response patterns, both neural and behavioral, to lexical features across time, suggesting some properties of text may play a unique role in modulating brain activation in dyslexic readers and offering insight into how targeted reading intervention may influence lexical processing.

12:40 Jayson Guo, "Resting‐State EEG Biomarkers of Rapid Nitrous Oxide Treatment Response in Obsessive–Compulsive Disorder" (Primary Honors Advisor: Peter van Roessel, Psychiatry and Behavioral Sciences; Second Reader: Carolyn Rodriguez, Psychiatry and Behvaioral Sciences)
     Obsessive–compulsive disorder (OCD) is a psychiatric condition marked by intrusive, unwanted thoughts (obsessions) and repetitive behaviors or mental rituals (compulsions). Many patients remain symptomatic despite standard treatments, driving interest in rapid‐acting and efficacious alternatives. Nitrous oxide (N₂O), an NMDA receptor antagonist, has shown preliminary clinical benefit, but its neurophysiological mechanisms are not well understood. In a randomized, placebo‐controlled study, we collected eyes‐closed, resting‐state EEG immediately before and 60 minutes after inhalation of N₂O or placebo in adults with OCD. To identify potential biomarkers of the N₂O treatment response, we performed spectral analysis on this data, decomposing the EEG signal into canonical frequency bands (e.g., delta, theta, alpha, beta, gamma). Spectral analysis revealed that N₂O inhalation consistently reduced delta‐band (1–4 Hz) power and elevated low‐alpha (8–10 Hz) power relative to placebo, particularly over central and right temporoparietal electrodes. Mediation analysis indicated that suppression of central delta oscillations, particularly electrode C3, partially accounted for the treatment’s beneficial effect on OCD symptoms, whereas alpha‐band changes did not. These findings identify central delta attenuation as a potential EEG biomarker of N₂O’s rapid action in OCD, offering new insight into the disorder’s neurophysiology and treatment mechanisms.

12:50 Natalie Wang, "Refining the Relationship Between Arousal and Pupil Size: The Role of Luminance" (Primary Honors Advisor: Brian Knutson, Psychology; Second Reader: Brian Yan, Psychology)
     Pupillometry has long been considered a reliable and non-invasive method to study cognitive processes such as arousal, through observing how pupils dilate and constrict to stimuli. Historically, studies in optometry and psychology have established that luminance is one of the primary influences on pupil diameter and its subsequent changes in size. In this thesis, we investigate whether controlling for luminance improves one’s ability to infer underlying cognitive processes from observing the pupil diameter. Using an established experimental paradigm of monetary incentive processing that manipulates the participants’ arousal levels, we ask whether there is a stronger association between the participants’ self-reported arousal and pupil diameter after luminance has been controlled. In Study 1, we implemented a statistical retroactive control for luminance by using a summed convolution function to remove the luminance effect from the pupil size variations. In Study 2, we implemented a proactive control for luminance by replacing the paradigm’s original screens with isoluminant screens and stimuli. We hypothesize that by removing the luminance confound from the data, either retroactively or proactively, there should be a stronger association between pupil size and arousal. Our results showed that either retroactively statistically controlling for luminance or proactively using an isoluminant paradigm did not significantly improve the association between pupil size and arousal. These findings were discussed in the context of covariation of luminance and cognitive processes, as well as the variability of pupil size under different luminance conditions.

1:00 Tom Shahar, "Exploiting Large Language Models for Improving Online Discourse Quality" (Primary Honors Advisor: Jeff Shrager, Symbolic Systems; Second Reader: Michael Bernstein)
     Social media platforms are often reactionary, negative, and exclusionary spaces. These challenges persist due to the limitations of conventional moderation methods and the lack of meaningful investment in alternative solutions by social media companies. This project explores the potential of large language models to address these issues. Specifically, I investigate applications of LLMs in automated moderation as well as the development of conversational assistants aimed at improving the quality of online interactions.

1:10 Oliver Lee, "Acquisition asymmetries in wh-question production: Insights from K’iche’" (Primary Honors Advisor: Beth Levin, Linguistics; Second Reader Boris Harizanov, Linguistics)
     This project investigates the acquisition of two types of questions:1) subject questions (‘Who saw him?’) and 2) object questions (‘Who did he see?’). Numerous studies have attested to a ‘subject-over-object asymmetry,’ where children produce subject questions faster and with greater accuracy than object questions (Otaki et al., 2023). To assess the language-universality of this asymmetry, we assess data from the production of subject and object questions in K’iche’, an Eastern Mayan language, by children aged 4;4 to 5;10. As in other studies in question acquisition, we find that children demonstrate particular difficulty in producing which-questions as opposed to who-/what-questions. However, contrary to these studies, we find an object-over-subject asymmetry (p<0.01) in which children’s object questions are closer to the adult target than their subject questions. To explain this, I propose an analysis inspired by Norcliffe (2009). When K’iche’-speaking children produce which-questions, these questions tend to incorrectly contain what Norcliffe (2009) analyzes as a resumptive pronoun (but which has been called a Set A agreement marker traditionally); this pronoun refers to the questioned entity, as in ‘Which rabbit, he saw the cat?’ instead of ‘Which rabbit saw the cat?’. I propose that this resumptive pronoun becomes increasingly preferred in child K’iche’ question constructions that risk overloading working memory capacities.

1:20 Avery Louis, "Standing on Whose Shoulders? Success Bias, Network Topology, and the Fate of Scientific Claims" (Primary Honors Advisor: Russ Poldrack; Second Reader: Daniel McFarland)
     What makes a scientific claim reliable? This thesis explores how the social structure around a scientific idea shapes its reliability—specifically, its odds of surviving replication. We combine simulation and empirical analysis to study how patterns of influence, imitation, and visibility affect which findings endure and which fail. In the first part, we develop an agent-based model in which scientists learn from both data and one another, and show that strong success bias—copying the most celebrated peers—can reduce epistemic success. In the second part, we test these insights against real-world citation data, examining the structural features of local scientific communities around papers whose replicability has been independently assessed. We find that certain network signatures—such as high clustering and low modularity—are associated with more reliable claims. Reliability, in this view, is a property not only of methods or individuals, but of the collective structures that shape attention and belief.

1:30 Chelsea Cho, "Representational Enactivism and the Limits of Robotic Visual Perception" (Primary Honors Advisor: Josh ORourke, Philosophy; Second Reader; Thomas Icard, Philosophy)
     This thesis investigates the philosophical and computational boundaries of robotic perception through the lens of representational enactivism—a theory that reconciles embodied sensorimotor interaction with internal representation. Challenging both the Cartesian model of perception and orthodox enactivist rejections of representation, the work proposes a hybrid framework wherein perception is both enacted through bodily engagement and scaffolded by internal cues and anticipatory structures. Using this framework, the thesis evaluates four state-of-the-art robotic vision models—OpenVLA, LAPA, UVA, and KAGI—assessing their alignment with enactivist criteria: sensorimotor knowledge, joint modeling of vision and action, interactive learning, and enactive representation.

1:40 Kevin Kai Wen Chen, "Impacts of Short-Form Media Use on Sustained Attention and Memory: A Behavioral Study of TikTok Consumption in Young Adults" (Primary Honors Advisor: Anthony Wagner, Psychology; Second Readers: Baba Shiv, Business, and Brian Knutson, Psychology)
     Short-form media platforms like TikTok are widely used by young adults. Despite its widespread use, there is limited research on the direct cognitive impacts of TikTok consumption. In this online study of 338 participants aged 18–24, I explored how TikTok usage relates to attention and memory performance. Participants completed the Sustained Attention to Response Task with a surprise memory component (SART-M). Heavier TikTok use was associated with greater attentional fluctuation and reduced memory accuracy. In addition, self-reported TikTok addiction, impulsivity, and media multitasking were all positively correlated with usage. These findings raise important questions about the cognitive costs of frequent short-form media use, with potential implications for learning, digital well-being, and attention hygiene in educational environments.

1:50 Vernita Zhai, "Eternalising the Apocalypse: Blake's Cartography of (Bio/Onto)logical Becoming" (Primary Honors Advisor: Denise Gigante, English; Second Reader: Alexander Nemerov, Art & Art History)
     Unfolding across an expansive and often convoluted set of prophetic books, the personal mythology of Romantic poet, painter, and engraver William Blake (1757–1827) narrates the fragmentation of the universal man Albion into four principles of human nature (in The First Book of Urizen, 1794, and Milton, c. 1804-11), and his eventual regeneration into a primordial organic and spiritual wholeness (in Jerusalem, c. 1804-20). This thesis advances a phenomenological reading of Blake’s mythos as an expression of what Italian philosopher and semiologist Paolo Virno terms the “biological invariant”: the acultural and ahistorical core of the human defined positively as the absence of a determinate nature. Abounding with scenes of cosmogenesis, ontogenesis, and morphogenesis, and conveyed through etchings that refuse to stabilise into a single interpretation, Blake’s universe—both narratively and formally—evokes the nature of the human animal as a being whose potential always exceeds that which can be materially realised. As a counterpoint to the methods of the metaphysical and positivist sciences, this thesis asks what we might discover about the ontology of Being when we set aside preconceived notions of human nature and attend instead to phenomenological and aesthetic experience as a genuine form of knowledge. Ultimately, in engaging these questions with and through Blake, this thesis argues for the importance of a phenomenological ontology as both an epistemic and ethical imperative.

2:00 Sarah Rose Smale, "Home tonometry and glaucoma – patient and provider perspectives on indications, benefits, and barriers to wider implementation" (Primary Honors Advisor: Michael Snyder, Genetics; Second Reader: Sharon F. Freedman, Ophthalmology and Pediatrics, Duke University School of Medicine)
     Glaucoma is the second leading cause of blindness globally. It is a degenerative condition that commonly results in elevated intraocular pressure (IOP), which, in turn, damages the eye's optic nerve. Glaucoma is an irreversible condition with no cure but there are treatments available to manage and monitor glaucoma. Treatments mainly focus on reducing IOP. Therefore, monitoring IOP in eyes with suspected and documented glaucoma is pivotal in the management of this disease. Tonometry (the measurement of IOP) is currently performed almost exclusively in the clinic setting, periodically, for patients with glaucoma. However, IOP is known to fluctuate widely in some patients with glaucoma, often showing IOP peaks outside traditional clinical office hours. Home tonometry, which enables patients to self-monitor IOP, is a more recently introduced practice than home blood pressure monitoring or home blood glucose monitoring. The FDA approved the first tonometer for home use in 2017. Despite its adoption by some eye care physicians and their patients, much remains to be learned regarding what factors lead physicians to be more likely to prescribe home tonometry for their glaucoma patients. Past research has found that home tonometry helps make more informed treatment decisions, particularly with regards to medical treatment (usually eyedrops) and surgery. However, home tonometry is still used by a minority of glaucoma patients. There remain real and perceived barriers to wider spread use of home tonometry, both on the provider and the patient side of the equation. This study aims to better understand the factors which influence the practice of home tonometry in glaucoma management. It takes a survey-based approach towards understanding attitudes and perspectives of both glaucoma patients and glaucoma specialists. Survey questions target perspectives on home tonometry, as well as barriers to its use. The survey results offer insights on home tonometry in adult glaucoma management, by querying both those who prescribe home tonometry for their patients, and those patients who have used home tonometry themselves. This project will hopefully lend enthusiasm to home tonometry as it may benefit both glaucoma patients and their glaucoma specialists in managing this prevalent and chronic disease.

2:10 Gene Sung-Ho Kim, "Beyond Being Accurate: Implications for Leveraging Natural Language to Support Nonvisual 3D Shape Understanding" (Primary Honors Advisor: Sean Follmer, Mechanical Engineering); Second Reader: Hari Subramonyam, Education)
     Generative AI can translate complex visual information into interactive natural language, which presents an exciting opportunity to proliferate visual descriptions for blind and low vision individuals. Much prior work has focused on 2D visualizations, graphics, and design tools. However, when it comes to more complex digital 3D information, it is still unclear if natural language alone can afford a sufficiently deep geometric understanding to support screen reader accessible 3D visualization and design. This work investigates the potential and limitations of natural language to support nonvisual 3D modeling through a series of qualitative studies. I articulate design and technical implications for using natural language in future accessible 3D modeling tools in addition to general implications of using language to represent complex 3D information.

After last speaker: Dessert for all

Notes on entry to the meeting room:

Entry to the building is open to anyone with an active Stanford ID via the card readers next to each door. If you do not have a Stanford ID, you can gain entry between 12:15 and 12:30pm ONLY by knocking on the exterior windows of room 126. These windows are to the left of the west side exterior door on the first floor of Margaret Jacks Hall, which faces the back east side of Building 420. Please do not knock on these windows after 12:30pm when the talk has started. We will not be able to come out and open the door for you at that point.