Haonan (Holland) Sun

Attention · Memory · Neuroimaging

How a learned readiness to shift attention shapes what we perceive and remember.

Shift cue · 250 ms Attending right · cue says L
The letter-cue RSVP task from Experiment 1 of my current project, adapted from Sali, Anderson and Yantis (2015). Two digit streams update every 250 ms. A letter in the attended stream (dashed outline) names the side to attend next: the same side means hold, the other side means shift. The participant then judges whether the digit in that stream is odd or even. Blocks that are mostly-shift or mostly-hold teach people how ready to be to move attention. The demo runs at half speed.

I am a Master's student in Psychology at Wake Forest University, advised by Dr. Anthony Sali. I study attentional control: how people learn how ready to be to shift attention, and what that learned state changes. I ask two questions about it. Is memory better when the state at test matches the state at study? And which stage of a shift does the state change? So far, memory was better when the state matched, but only for pictures first seen early in learning. My thesis takes on the second question. With Dr. Stefon van Noordt, I am also analyzing EEG to see how medial-frontal theta during cognitive control differs in early-phase psychosis.

Before Wake Forest, I spent three years at the Sleep and NeuroImaging Center at Southwest University in Chongqing, China (PI: Prof. Xu Lei), first as an undergraduate researcher and then as lab coordinator. There I used structural and functional MRI, machine learning and brain transcriptomics to identify biologically grounded subtypes of insomnia disorder, and ran data collection at four hospitals for a national insomnia project.

  • Attentional flexibility
  • Encoding specificity
  • Spatial attention
  • Cognitive control
  • Insomnia heterogeneity
  • MRI & machine learning

Research

Four lines of work

At Wake Forest I study attentional control: how people learn how ready to be to shift attention, what that learned state changes, and how control shows up in the EEG. Before that, at Southwest University, I used MRI and machine learning to study why patients with the same insomnia diagnosis differ.

Now · Wake Forest University

  1. 1
    Learned attentional flexibilityWith Dr. Anthony Sali · two projects, including my thesis
  2. 2
    Cognitive control in the EEGWith Dr. Stefon van Noordt · one project

Before · Southwest University

  1. 3
    Heterogeneity in insomnia disorderWith Prof. Xu Lei · four papers and a study I led
  2. 4
    Other projectsA study I ran as student PI, and a metabolomics project

Now · Wake Forest University

1Learned attentional flexibility

2025–present · with Dr. Anthony Sali

People learn, without being told, how often a setting will ask them to move their attention (Sali, Anderson & Yantis, 2015). In mostly-shift blocks, shifting gets cheaper; in mostly-hold blocks, it stays costly. This shows up in the shift cost: how much slower people are when a cue tells them to move than when it tells them to stay.

I ask two questions about this learned state. Is it part of the context a memory is stored in, so that memory is better when the state matches? And which stage of a shift does it change? My first project takes on the first question, and my thesis takes on the second.

Project leadSep 2025–present

Attentional flexibility as an internal encoding context for explicit and implicit memory

Question 1Is memory better when the attentional state at test matches the one at study?

Memory is better when the context at retrieval matches the context at encoding (Tulving & Thomson, 1973). This holds for external settings, such as learning on land or underwater, and for internal states, such as being sober or intoxicated. I ask whether a learned attentional state can serve as that context: is memory better when the shift-or-hold state at retrieval matches the state at encoding?

I designed both experiments, programmed them in MATLAB with Psychtoolbox, obtained IRB approval, ran the participants, and analyzed the data in R with (generalized) linear mixed-effects models and repeated-measures ANOVA.

Experiment 1 · Explicit memoryN = 21 · presented at NCCC, March 2026
Task
The letter-cue RSVP task at the top of this page. Four blocks of 60 trials, two mostly-shift and two mostly-hold. On 15 trials per block a word followed the attention trial. Words were studied in the first two blocks and tested with a two-choice recognition test in the last two.
Design
Mixed. Participants were tested either in the same block type they had studied in (congruent group) or in the other one (incongruent group).
Result
People learned the block regularity. In mostly-hold blocks, shift trials were about 200 ms slower than hold trials; in mostly-shift blocks the two did not differ. Recognition accuracy did not depend on whether the state matched.
Next
Few memory trials per condition survived screening, so a small effect could have been missed. Experiment 2 moved to a within-subject design and a continuous measure.
Experiment 2 · Implicit memoryN = 42 · to be presented at OPAM, November 2026
Task
The cue became a picture. Both streams showed phase-scrambled noise, and the cue was a degraded object image (20% image, 80% noise) in the attended stream: an animal meant shift, an inanimate object meant hold. Images came from the Konkle and Caramazza (2013) animacy-by-size set. Four blocks of 80 trials; mostly-shift blocks had 70% shift trials and mostly-hold blocks had 70% hold trials.
Design
Fully within-subject. The first two blocks, one of each type, were for encoding. 96 of their pictures came back as cues in the last two blocks, with the same meaning as before. Memory was measured implicitly, as response time on trials cued by a repeated picture.
S = mostly-shift block, H = mostly-hold block. A repeated picture keeps its cue meaning; only the block type around it can change.
Retrieval in SRetrieval in H
Encoding in SCongruentIncongruent
Encoding in HIncongruentCongruent
Result
The shift cost was again smaller in mostly-shift blocks (about 60 vs 140 ms at encoding), and the difference held at retrieval. Repeated pictures were handled faster when the block type matched. The benefit came from pictures first seen early in an encoding block, before performance leveled off, where it was about 45 ms. Pictures first seen late in a block showed none.
Next
Two accounts fit this pattern. The state may bind more strongly to items met while the task is still being learned. Or an abstract context cue may help more after a longer delay, and early items waited longer before they came back. In this design, learning stage and delay move together, so the next step is to pull them apart.
Experiment code on GitHub
Master's thesis · in progressJun 2026–present

The temporal dynamics of learned attentional flexibility: Investigating the role of disengagement

Committee: Dr. Anthony Sali, Dr. Janine Jennings, Dr. Stefon van Noordt

Question 2Which stage of an attention shift does the learned state change?

Learned flexibility shows up as a smaller shift cost, but a difference in response time cannot say which part of a shift changed, or when. A spatial shift has three parts: disengaging from the current location, moving, and engaging at the new one (Posner et al., 1984). My thesis tests whether the learned state works through the first.

  1. DisengageLower the priority of the location attention is on, so another location can take over. This is the step the thesis tests.
  2. MoveCarry attention toward the new location.
  3. EngageSettle attention on what is at the destination.

I treat disengagement in priority-map terms: it reduces the weight on the currently prioritized location. That way the account does not depend on attention sitting at only one place at a time.

1 · Start location · 500 ms
2 · Go, then a 1,350–1,950 ms gap
3 · Color cue and Landolt Cs
4 · Report "up" by 2,000 ms after go
One shift trial with incongruent targets, for a participant whose shift color is blue. The cue appears at the starting location on the right, so attention moves left, where the C opens up. The C on the right opens down and pulls toward the wrong answer.
Task
A compelled-response task, borrowed from studies of urgent saccadic choice (Shankar et al., 2011). A thick outline marks the starting location, carried over from the previous trial. When it disappears, the clock starts, and the response has to get under way before the participant knows what it should be. After 1,350–1,950 ms the starting circle turns blue or red, meaning shift or hold (counterbalanced), and a Landolt C appears in each circle. The participant reports whether the C at the to-be-attended location opens up or down, by 2,000 ms after the go signal. That leaves 50–650 ms after the cue. An eye tracker checks central fixation.
Design
Context × trial type × congruency, all within subject. Four mostly-shift blocks (70% shift) and four mostly-hold blocks (70% hold), 80 trials each: 640 trials in about 1.5 hours. On congruent trials the two Cs open the same way; on incongruent trials the C at the location to ignore points to the wrong answer. Planned N = 40, so that about 34 usable participants give 80% power for the key interaction.
Measure
Processing time is response time minus the gap: how long the cue had been on screen when the response came. Plotting accuracy against processing time gives a tachometric curve that climbs from chance to a ceiling. Its center point and rise time show when the cued C takes over the response, and how fast, separately for congruent and incongruent trials.
Prediction
If a mostly-shift context speeds disengagement, the C at the old location should stop interfering sooner on shift trials, so the congruency effect shrinks. On hold trials the same change should backfire: readiness to leave hands priority to the other location before the cue, so its C interferes more and the congruency effect grows. The key test is the context × trial type × congruency interaction on center point and rise time.

2Cognitive control in the EEG

With Dr. Stefon van Noordt

In progressCollaboration

Medial-frontal theta during cognitive control in early-phase psychosis

QuestionDoes early-phase psychosis change medial-frontal theta during a cognitive-control task, and does a dose of caffeine change it?

Medial-frontal theta (4–7 Hz) is an EEG index of cognitive-control processing. This project re-analyzes EEG from Francis et al. (2025), who studied the P300 in the same recordings. Here the focus moves from that evoked response to theta-band activity.

Data
14 healthy controls and 11 people with early-phase psychosis, each recorded in a caffeine session and a placebo session.
Task
The AX-CPT. A cue letter is followed by a probe letter, and A followed by X is the frequent target. The other pairs (AY, BX, BY) are rarer. On BX trials the habitual response to X has to be withheld, so they matter most if enough clean trials remain.
Analysis
Wavelet time–frequency decomposition, then theta power over a medial-frontal cluster centered on Fz, measured after the cue and after the probe. The time window is set from the task timing and prior work, not from the group differences. Mixed-effects models test group, drug and their interaction, with age as a covariate.
Then
Exploratory tests of whether theta, or its change with caffeine, tracks accuracy and response time.

My partThe analysis, from checking the cleaned and segmented recordings through the time–frequency maps and scalp distributions to the statistical models.

Before · Southwest University, Chongqing

3Heterogeneity in insomnia disorder

2022–2025 · with Prof. Xu Lei

Case–control MRI studies of insomnia often disagree. One reason is that the diagnosis groups together people whose brains differ in different ways. With colleagues at the Sleep and NeuroImaging Center, I first looked for data-driven subtypes, then asked what biology and behavior go with insomnia-related brain differences. My part ranged from collecting data in hospitals to preprocessing, machine learning and imaging transcriptomics. I made the figures shown here for the papers; select one to see it full size.

Finding subtypes

  1. 2022–2023 · Biological Psychology, 2023

    Subtypes from gray matter volume

    Patients varied more from one another than controls did. HYDRA clustering of T1-weighted MRI in 56 patients and 73 controls found two reliable subtypes. One had widespread gray matter reductions tied to daytime functioning. The other had a focal increase in the right superior temporal gyrus tied to sleep disturbance.

    Paper and my part
    Brain surface maps for two insomnia subtypes. Subtype 1 shows blue patches of reduced gray matter across frontal, temporal and occipital cortex. Subtype 2 shows one red-yellow patch of increased gray matter in the right superior temporal gyrus.
    Gray matter differences from controls: widespread decreases in subtype 1 (blue), a focal increase in subtype 2 (red-yellow). Fig. 3 in the paper. © 2023 Elsevier B.V.

  2. 2024 · Human Brain Mapping, 2025

    Subtypes from morphometric similarity networks

    We described each person's cortex as a network across 308 regions, built from five surface features. In 144 patients and 124 controls, these networks separated two subtypes. One slept objectively short on wearable sleep EEG and showed impaired cognition. The other slept a normal length of time but reported poor sleep.

    Paper and my part
    Cortical maps of morphometric similarity network strength for controls and two insomnia subtypes, case-control t-maps for each subtype, and two scatterplots relating each subtype's t-map to network strength in controls.
    Network strength in controls and each subtype (a), each subtype against controls (b, c), and how those changes relate to the healthy network (d). Fig. 3 in the paper. © 2025 The Authors, CC BY-NC-ND.

  3. 2024–2025 · internal lab report

    Subtypes with progression stages Project lead

    Applied SuStaIn to multimodal MRI (MIND networks, fALFF, ReHo, functional connectivity) to recover both subtypes and their disease stages, and mapped each trajectory to a sleep-EEG phenotype from portable forehead recordings.

    My partDesigned the study and did all of the preprocessing. Built MIND networks from T1w MRI and the resting-state measures used as SuStaIn input, and wrote my own processing for the raw forehead EEG.

What goes with the brain differences

  1. 2023–2024 · PNP&BP, 2025

    Brain function, not structure, tracks gene expression

    We linked case–control brain maps to Allen Human Brain Atlas gene expression, in a discovery sample (264 patients, 129 controls) and a validation sample (115, 160). Changes in resting-state function tracked gene expression; gray matter changes did not. 677 genes correlated positively and 582 negatively with the main functional component. They pointed to HPA-axis regulation, ion transport and inhibitory neuron pathways.

    Paper and my part
    A cortical map of the first gene expression component; scatterplots showing it correlates with the functional component but not with gray matter differences; a list of the genes most strongly correlated with the functional component, with example scatterplots and expression maps.
    The main gene expression component (A) tracks the functional changes (B) but not gray matter (C). D lists the genes most strongly linked to the functional changes. Fig. 3 in the paper. © 2024 Elsevier Inc.

  2. 2023–2024 · J. Psychiatric Research, 2024

    Response inhibition and frontal–striatal connectivity

    Young adults with insomnia (115) stopped more slowly than controls (160) on the stop-signal task. They also differed in frontal–striatal and motor-area connectivity. Within patients, stronger right superior frontal–putamen connectivity went with slower stopping.

    Paper and my part
    Glass-brain views with labeled nodes in the left and right supplementary motor area, left paracentral lobule, right superior frontal gyrus, right orbital middle frontal cortex and right putamen, joined by red and blue edges.
    Connections that differed between patients and controls: red edges stronger in patients, blue edges weaker. SFG, superior frontal gyrus; OMFC, orbital middle frontal cortex; SMA, supplementary motor area; PCL, paracentral lobule. Fig. 5 in the paper. © 2023 Elsevier Ltd.

4Other projects

2021–2024

  1. 2021–2023

    Sleep, anxiety and depression during COVID-19 Student PI

    A funded longitudinal study of how anxiety and depression shape sleep quality over time, with rumination as a mediator and negative cognitive bias as a moderator. Advised by Dr. Yuan Liu.

    My partWrote the proposal and won national undergraduate funding through a three-round review with an oral defense. Led four research assistants and fit longitudinal moderated mediation models in Mplus.

  2. 2024

    Predicting sleep patterns from metabolomics and metagenomics

    Third-party metabolomic and metagenomic data from more than 1,500 participants, linked to sleep phenotypes.

    My partCurated the data, ran dimensionality reduction and clustering, and wrote R scripts to speed up processing for the lab (code).

Data collection for two cohort studies, the Behavioral Brain Research Project of Chinese Personality and a national study of brain–mind development in children, is listed under Positions.

Publications

Peer-reviewed articles

All four come from the Sleep and NeuroImaging Center at Southwest University. I am second author on each and made the figures. The line under each paper says what else I did; a summary and one figure from each are under Research.

  1. 2025

    Gene expression is associated with brain function of insomnia disorder, rather than brain structure

    Zhang, H., Sun, H., Li, J., Lv, Z., Tian, Y., & Lei, X.

    Progress in Neuro-Psychopharmacology and Biological Psychiatry, 136, 111209

    doi:10.1016/j.pnpbp.2024.111209Summary and figure

    My partComputed GMV, ALFF and ReHo maps and effect sizes, processed AHBA expression data across several atlases, ran the spatial permutation and bootstrap tests and the GO/KEGG, tissue and cell-type enrichment analyses, and wrote parts of the Methods, Results and Discussion.

  2. 2025

    Subtypes of insomnia disorder identified by cortical morphometric similarity network

    Zhang, H., Sun, H., Li, J., & Lei, X.

    Human Brain Mapping, 46(1), e70119

    doi:10.1002/hbm.70119Open accessSummary and figure

    My partRan FreeSurfer, built the individual networks, mapped AHBA expression onto the same 308 regions (AHBA_SurfRecon), and extracted sleep latency, total sleep, NREM and REM time, REM cycles and wake time from the forehead EEG. Helped answer reviewers on the methods.

  3. 2024

    Response inhibition impairment related to altered frontal-striatal functional connectivity in insomnia disorder: A pilot and non-clinical study

    Zhang, H., Sun, H., Li, J., Yang, J., Fan, Y., Jülich, S. T., & Lei, X.

    Journal of Psychiatric Research, 170, 138–146

    doi:10.1016/j.jpsychires.2023.12.023Summary and figure

    My partScored an executive-function battery (stop-signal, digit span, task switching) for about 120 participants, ran CAT12 VBM and SPM12 resting-state preprocessing, helped with seed-based connectivity, and worked on the Methods and revision.

  4. 2023

    Subtypes of insomnia revealed by the heterogeneity of neuroanatomical patterns: A structural MRI study

    Zhang, H., Sun, H., Li, J., Fan, Y., Jülich, S. T., & Lei, X.

    Biological Psychology, 180, 108591

    doi:10.1016/j.biopsycho.2023.108591Summary and figure

    My partPooled MRI from five earlier lab projects, preprocessed the T1w and DTI data in SPM and FSL, ran kernel SVM classification and HYDRA in MATLAB, and drafted parts of the Methods and Discussion.

Presentations

Posters

  1. Nov 2026Upcoming

    Attentional flexibility as an internal context: A temporally graded context-dependent priming effect

    Sun, H., Mu, B., & Sali, A. W. · 34th Object Perception, Attention, & Memory (OPAM), San Diego, CA

    ShowsExperiment 2: the match benefit for pictures first seen early in a block. Details

  2. Apr 2026

    Attentional states as psychological context for both explicit and implicit memory: An examination of encoding specificity

    Sun, H., Mu, B., & Sali, A. W. · Wake Forest University Psychology Department Poster Session, Winston-Salem, NC

    ShowsExperiments 1 and 2 together.

  3. Mar 2026

    Attentional states as psychological context for long-term memory: An examination of encoding specificity

    Sun, H., Mu, B., & Sali, A. W. · North Carolina Conference on Cognition (NCCC), Duke University, Durham, NC

    ShowsExperiment 1: learned shift costs and the recognition test. Details

  4. Jun 2024

    Brain gene expression of insomnia disorder is associated with brain function, rather than structure

    Zhang, H., Sun, H., Li, J., Lv, Z., & Lei, X. · 30th Annual Meeting of the Organization for Human Brain Mapping (OHBM), Seoul, South Korea

    ShowsEarly version of the gene expression paper. Paper

  5. Aug 2022

    Negative cognitive bias moderates the mediating effect of rumination on the relationship between anxiety/depression and sleep quality trajectories during COVID-19 lockdown: Preliminary evidence

    Sun, H., Wu, H., Liu, H., Yang, R., & Liu, Y. · National College Students' Innovative Training Program mid-term review, Southwest University, Chongqing

    ShowsPreliminary data from my student-PI project.

Curriculum vitae

Background

Full CV (PDF)

Education

Teaching assistant

  • PSY 151 · Introductory PsychologyDr. Stefon van NoordtSpring 2026
  • PSY 220 · BiopsychologyDr. Melissa MaffeoFall 2025
  • PSY 246 · Culture and PsychologyDr. Merve Balkaya-InceFall 2025

Positions

  • Lab Coordinator, Sleep and NeuroImaging Center, Southwest University PI: Prof. Xu Lei. Main project: a National Key R&D Program subproject on the cognitive neuroscience of insomnia subtypes and common comorbidities.
    • Led off-site data collection at four hospitals in Chongqing: scheduling, liaison with physicians, and the travel and equipment budget.
    • Led data collection for about 150 insomnia patients, out of about 600 participants recruited by the team: portable sleep EEG, actigraphy, cognitive tasks and questionnaires.
    • Organized more than 140 structural, resting-state and task MRI scans with two imaging centers, and assisted with overnight polysomnography.
    • Ran batch fMRIPrep preprocessing and quality control on the project's multimodal data, and merged earlier lab projects into one dataset of 90 patients and 100 controls.
    • Maintained the EEG and actigraphy devices, trained new research assistants, and completed metagenomic and metabolomic analysis training at BGI, Shenzhen.
    Apr 2023 – Feb 2025
  • Research Assistant, Behavioral Brain Research Project of Chinese PersonalityProf. Xu Lei. Testing sessions, questionnaire scoring and actigraphy for a longitudinal multimodal cohort.May 2022 – May 2023
  • Research Assistant, national cohort study on brain–mind development of school-age childrenSouthwest University site, Prof. Qinghua He. EEG recording with children and mock MRI sessions.May 2023 – Sep 2023
  • Student PI, National College Students' Innovative Training ProgramAdvised by Dr. Yuan Liu. Longitudinal study of anxiety, depression and sleep quality during COVID-19; led a team of four.Sep 2021 – Sep 2023

Honors

  • Dean's Scholarship, Wake Forest University2025, 2026
  • Second Class Scholarship (top 10%), Southwest University2022
  • National College Students' Innovative Training Program (top 1%)2021, 2022
  • Winner (top 1%), China College Students "Internet+" Innovation Competition2022
  • Bronze Award (top 30%), "Hanhong Cup" Innovation Competition2022

Service

  • President, Psychology Student Union, Southwest UniversityOrganized the department's first Student Representative Conference and acted on over half of its proposals. Restructured the union and added five new events.2021–2022
  • Volunteer, group counseling programChongqing No. 122 Middle School. Designed and ran five supervised group sessions for 21 students.2021–2022

Methods & tools

Neuroimaging
Structural (T1w, DTI) and functional (resting-state, task) MRI in surface and voxel space; surface-based morphometry and morphometric similarity networks (MSN, MIND); SPM, CAT12, FreeSurfer, FSL, ANTs, fMRIPrep
EEG & sleep
EEGLAB; portable forehead sleep EEG, including custom processing of raw recordings; actigraphy; overnight polysomnography support
Machine learning
HYDRA, SuStaIn, kernel SVM, dimensionality reduction and clustering
Genes & omics
Imaging transcriptomics with the Allen Human Brain Atlas (spatial permutation tests, GO/KEGG enrichment); metabolomic and metagenomic data
Statistics
Linear and generalized linear mixed-effects models, repeated-measures ANOVA, SEM and longitudinal mediation; R, Mplus, SPSS
Experiments
MATLAB with Psychtoolbox (RSVP and compelled-response tasks); Python and R for analysis pipelines
Languages
Mandarin (native), English (fluent)