Rising Stars Session

Saturday, March 7, 2026, 1:00 – 3:00 pm PST, Salon EF

Talk 1, 1:00 pm

Default mode event representations and memory

Alexander Barnett¹; ¹McGill University

People construct and use mental "event models" by combining incoming information with prior knowledge to comprehend, interpret and predict ongoing events. Multivariate representations with default mode network (DMN) areas appear to carry event model information. Here we examined how events are stored into memory and how event representations shape what is remembered. Healthy adults were scanned with fMRI as they encoded and recalled cartoon movies. We observed that functional coupling between the DMN and hippocampus during encoding at critical moments called "event boundaries" was related to subsequent retrieval of events. Next we sought to see whether multivariate event patterns shaped recall content. Using topic modeling, we were able to examine how similarly pairs of participants recalled events from the cartoon movies. To do this, we transformed recall transcripts into "embeddings" and calculated the similarity between embedding vectors between participants. We observed that participants who recalled events more similarly also had similar multivariate event patterns in DMN regions at encoding. Together, these findings suggest that event representations in DMN regions at encoding shape the features recalled at retrieval and the success of retrieval may be dependent on hippocampal-DMN interactions.

Talk 2, 1:15 pm

Distraction disrupts control over the encoding and maintenance of information in visual working memory

Blaire Dube¹ (bdube@mun.ca), Elissa Orbasli¹; ¹Memorial University of Newfoundland

Careful control over visual working memory (VWM) allows it to effectively support behaviour: A filter governs which information gains access to VWM, and concurrently maintained representations are organized into distinct functional states. During visual search, for instance, a target is held in an active state that biases attention toward matching input, while prospectively relevant information can be maintained in an accessory state shielded from attention to prevent interference. Recently, the Filter Disruption Theory proposed that salient distraction disrupts control over access to VWM, allowing irrelevant information to enter. Does salient distraction also disrupt control over the internal organization of VWM, causing accessory representations to incidentally guide attention? Across two experiments, participants maintained two shapes in VWM: an active item serving as the search target, and an accessory memory item (AMI) that was relevant for a subsequent memory test. Participants first completed a visual search among either five (Experiment 1) or two (Experiment 2) non-targets, one of which occasionally matched the AMI, before performing the memory task. On some trials, a salient distractor—a briefly flashing border surrounding a non-target—appeared during search and sometimes coincided with the AMI’s appearance in the search array. When no distractor was present, the AMI did not influence attention: search response times were unaffected by its presence. In contrast, when a salient distractor appeared, search was reliably slowed by the presence of the AMI. These findings suggest that salient distraction disrupts control over the functional organization of VWM, allowing accessory representations to transiently assume an attention-guiding role.

Talk 3, 1:30 pm

Dual-format of preparatory attention in human visual cortex

Mengyuan Gong¹ (gongmy426@zju.edu.cn); ¹Zhejiang University

To manage overwhelming sensory inputs, the brain must prioritize task-relevant information through the allocation of attention. However, the neural representational format of attentional templates during preparation remains elusive. While a prevailing view suggests that target templates reflect veridical sensory features, recent theories propose that non-veridical representations are often “good-enough” for attentional guidance. Here, we investigated how the brain coordinates these seemingly conflicting views. In an fMRI study using a cuing paradigm, we show that preparatory activity patterns for a cued target distinctly differ from those evoked by actual sensory perception, indicating a "non-sensory" representation. Crucially, by perturbing neural activity using a visual impulse ("pinging" technique), we uncovered a latent activity pattern in the visual cortex that resembled the veridical sensory template. The emergence of this sensory-like format coincided with enhanced functional connectivity between V1 and frontoparietal regions and predicted improved behavioral performance. Together, these findings reveal a dual-format mechanism for preparatory attention: the brain concurrently maintains an active, non-sensory representation alongside a latent, sensory-like template. This representational flexibility may provide a significant advantage for adaptive attentional control in complex environments.

Talk 4, 1:45 pm

Inferring Self-schemas with Sequential Sampling Models

Peter Hitchcock¹ (peter.hitchcock@emory.edu), Sandarsh Pandey¹; ¹Emory University

Cognitive theories of depression assert that maladaptive self-schemas are the taproot of this disorder. Yet it has been difficult to measures self-schemas, which are typically defined as latent representations that guide cognition outside of conscious awareness. Self-reports, reflecting patent self-beliefs, are clearly inadequate measures of self-schemas by this definition. Researchers have thus turned to behavioral tasks in which participants make positive or negative self-judgments. These tasks are thought to capture distinct information from self-report, yet it has not been possible to verify this in past tasks. We developed a novel task and deployed it in two studies oversampling for depression (N=1008; median PHQ-9=12). Explicit self-beliefs collected before the task exerted a marked influence on self-judgments in the task itself—a source of variance that past studies have neglected. Yet there were also distinct effects of valenced processing during the task itself, as captured by a regressor on the drift rate of a drift-diffusion model and corresponding to distinct choice and RT signatures. Moreover, although, as expected, more (vs. less) depressed individuals reported starkly more negative (and less positive) explicit self-beliefs, they also showed distinct valenced processing—relatively more negative (than positive) evidence accumulation—during the task itself. Conversely, although we replicated a common finding that more depressed individuals subsequently remembered relatively more negative (of all total) words, this pattern was explained simply by negative words being more self-descriptive for them. Our results show how decision-science methods can be translated to gain insights into self-schemas theorized to be at the root of depression.

Talk 5, 2:00 pm

Neural algorithms of human language

Laura Gwilliams¹ (laura.gwilliams@stanford.edu); ¹Stanford University

The goal of my research is to develop a theoretically grounded, biologically constrained and computationally explicit account of how the human brain achieves language comprehension. In my talk, I will present a series of studies that examine neural responses at different spatial scales: From population ensembles using magnetoencephalography and electrocorticography, to the encoding of speech properties in individual neurons across the cortical depth using Neuropixels probes and microelectrodes in humans. The results provide insight into (i) what auditory and linguistic representations serve to bridge between sound and meaning; (ii) what operations reconcile auditory input speed with neural processing time; (iii) how information at different timescales is nested, in time and in space, to allow information exchange across hierarchical structures.

Talk 6, 2:15 pm

Stubborn and flexible predictions in perception and belief updating

Daniel Yon¹ (d.yon@bbk.ac.uk); ¹Birkbeck, University of London, ²All Souls College, University of Oxford

Thinking about the human mind has been transformed by the idea that the brain may be ‘Bayesian’ – relying on predictions and prior beliefs to make sense of a noisy, ambiguous and unstable world. A key problem that predictive brains need to solve is setting the ‘balance’ between stubbornness and flexibility -determining whether to rely on old predictions in new contexts, or to form new hypotheses as environments change around us. In this talk I will describe a mixture of experiments from my lab which reveal that perceptual brain circuits can rely too stubbornly on ‘outdated’ sensory predictions, biasing our representation of the present. However, flexibility might not be all it is cracked up to be – as alternate work from the group suggests that overestimating the volatility of our environment can also lead to unstable patterns of prediction – including in pathological states like paranoid delusions. Understanding how the brain should and does set this balance is important for understanding how perception and cognition become (mis)attuned to the extracranial world we find ourselves inhabiting.

Talk 7, 2:30 pm

Working memory beyond the cortex

Anastasia Kiyonaga¹ (akiyonaga@ucsd.edu); ¹University of California, San Diego

The core role of working memory (WM) is to sustain short-term goal content in mind for long enough to guide behavior. Yet the nervous system may engage myriad functions to accomplish that feat. WM representations are distributed across the brain and can now be detected in evolutionarily earlier structures than previously thought—like the cerebellum, thalamus, and superior colliculus. Mounting evidence also shows that peripheral oculomotor and physiological signals can carry WM feature content, raising the question of just how far WM signals go. Here, in three studies, we test the idea that visual WM content is adaptively distributed across the nervous system according to behavioral demands. We examine how WM content is expressed in activity patterns across the eyes and hands, and whether the distribution of such peripheral motor activity shifts with the task context. We find feature-specific WM content signals in pupil size modulations (Expt. 1), spatio-temporal gaze sequences (Expt. 2), and systematic manual gestures (Expt. 3). In all experiments, we manipulate which stimulus dimensions are most relevant for the WM test (e.g., visual detail or semantic category) and find that peripheral WM signals are modulated accordingly. These findings converge on the idea that WM may recruit the most primary structures in sensorimotor processing. Rather than indiscriminately spread throughout the brain and body, however, WM content information is efficiently allocated across the sensory and motor structures that are best-positioned to support current needs.