Daria Kussovska
@dariakussovska.bsky.social
PhD student at Columbia University | Biomedical Engineering | Human Single-Neuron Computation
9/ I’m incredibly grateful to my PI @nuttidanuttida.bsky.social and to my coauthor Robert Kim for their support throughout this project. Thank you also to Jan Kamiński, Ueli Rutishauser, and Florian Mormann for invaluable feedback.
December 31, 2025 at 9:32 AM
9/ I’m incredibly grateful to my PI @nuttidanuttida.bsky.social and to my coauthor Robert Kim for their support throughout this project. Thank you also to Jan Kamiński, Ueli Rutishauser, and Florian Mormann for invaluable feedback.
See the preprint for more! www.biorxiv.org/content/10.1...
Neural heterogeneity shapes the temporal structure of human working memory
Working memory (WM) enables temporary retention of information essential for flexible cognition. Although persistent population activity has long been regarded as a principal mechanism of memory maint...
www.biorxiv.org
December 31, 2025 at 9:32 AM
See the preprint for more! www.biorxiv.org/content/10.1...
8/ Overall, our results suggest that human WM can be supported by cell-type-specific, temporally structured dynamics across heterogeneous neurons. We are very excited about these results, and continuing to study how neuronal heterogeneity supports WM computations.
December 31, 2025 at 9:31 AM
8/ Overall, our results suggest that human WM can be supported by cell-type-specific, temporally structured dynamics across heterogeneous neurons. We are very excited about these results, and continuing to study how neuronal heterogeneity supports WM computations.
7/ Mapping these effects to putative cell types, we find distinct contributions. Putative pyramidal neurons express content in burst-aligned events during maintenance, whereas putative interneurons are strongly modulated by memory load and behavior.
December 31, 2025 at 9:28 AM
7/ Mapping these effects to putative cell types, we find distinct contributions. Putative pyramidal neurons express content in burst-aligned events during maintenance, whereas putative interneurons are strongly modulated by memory load and behavior.
6/ This is where bursting comes in! We examined burst structure and found that maintenance is supported by transient, burst-like episodes of coordinated activity.
December 31, 2025 at 9:28 AM
6/ This is where bursting comes in! We examined burst structure and found that maintenance is supported by transient, burst-like episodes of coordinated activity.
5/ To look at population dynamics more directly, we applied dPCA to concept-cell activity. During maintenance, stimulus trajectories evolve over time. Separation is stronger earlier and weaker later in the maintenance period, consistent with a code that transforms across time.
December 31, 2025 at 9:27 AM
5/ To look at population dynamics more directly, we applied dPCA to concept-cell activity. During maintenance, stimulus trajectories evolve over time. Separation is stronger earlier and weaker later in the maintenance period, consistent with a code that transforms across time.
4/ Stimulus information is not stably expressed across the full maintenance. A key clue for us was that decoding accuracy increases with wider temporal windows. This suggests apparent persistence can emerge from temporally interleaved activity across neurons.
December 31, 2025 at 9:26 AM
4/ Stimulus information is not stably expressed across the full maintenance. A key clue for us was that decoding accuracy increases with wider temporal windows. This suggests apparent persistence can emerge from temporally interleaved activity across neurons.
3/ Interesting: at the single-neuron level, most concept cells do not show sustained elevation throughout maintenance. Yet when pooled across neurons, population activity can still look persistent.
December 31, 2025 at 9:24 AM
3/ Interesting: at the single-neuron level, most concept cells do not show sustained elevation throughout maintenance. Yet when pooled across neurons, population activity can still look persistent.
2/ We analyzed an open human single-unit dataset (Kyzar et al., 2024): 902 neurons from 21 neurosurgical participants performing a Sternberg WM task. We focused on stimulus-selective “concept cells.”
Thank you to the Rutishauser lab and the original authors for sharing these invaluable recordings!
Thank you to the Rutishauser lab and the original authors for sharing these invaluable recordings!
December 31, 2025 at 9:24 AM
2/ We analyzed an open human single-unit dataset (Kyzar et al., 2024): 902 neurons from 21 neurosurgical participants performing a Sternberg WM task. We focused on stimulus-selective “concept cells.”
Thank you to the Rutishauser lab and the original authors for sharing these invaluable recordings!
Thank you to the Rutishauser lab and the original authors for sharing these invaluable recordings!
1/ Persistent activity is the classic working memory (WM) account, but continuous single-neuron firing is metabolically expensive and hard to reconcile with human cortical heterogeneity. So, we asked: can human WM be maintained without sustained single-neuron persistence?
December 31, 2025 at 9:23 AM
1/ Persistent activity is the classic working memory (WM) account, but continuous single-neuron firing is metabolically expensive and hard to reconcile with human cortical heterogeneity. So, we asked: can human WM be maintained without sustained single-neuron persistence?
10/ I’m incredibly grateful to my PI @nuttidanuttida.bsky.social and to my coauthor Robert Kim for their support throughout this project. Thank you also to Jan Kamiński, Ueli Rutishauser, and Florian Mormann for invaluable feedback.
December 31, 2025 at 8:26 AM
10/ I’m incredibly grateful to my PI @nuttidanuttida.bsky.social and to my coauthor Robert Kim for their support throughout this project. Thank you also to Jan Kamiński, Ueli Rutishauser, and Florian Mormann for invaluable feedback.
9/ Overall, our results suggest that human WM can be supported by cell-type-specific, temporally structured dynamics across heterogeneous neurons. We are very excited about these results, and continuing to study how neuronal heterogeneity supports WM computations.
December 31, 2025 at 8:24 AM
9/ Overall, our results suggest that human WM can be supported by cell-type-specific, temporally structured dynamics across heterogeneous neurons. We are very excited about these results, and continuing to study how neuronal heterogeneity supports WM computations.
8/ Mapping these effects to putative cell types, we find distinct contributions. Putative pyramidal neurons express content in burst-aligned events during maintenance, whereas putative interneurons are strongly modulated by memory load and behavior.
December 31, 2025 at 8:22 AM
8/ Mapping these effects to putative cell types, we find distinct contributions. Putative pyramidal neurons express content in burst-aligned events during maintenance, whereas putative interneurons are strongly modulated by memory load and behavior.
7/ This is where bursting comes in! We examined burst structure and found that maintenance is supported by transient, burst-like episodes of coordinated activity.
December 31, 2025 at 8:21 AM
7/ This is where bursting comes in! We examined burst structure and found that maintenance is supported by transient, burst-like episodes of coordinated activity.
6/ To look at population dynamics more directly, we applied dPCA to concept-cell activity. During maintenance, stimulus trajectories evolve over time. Separation is stronger earlier and weaker later in the maintenance period, consistent with a code that transforms across time.
December 31, 2025 at 8:18 AM
6/ To look at population dynamics more directly, we applied dPCA to concept-cell activity. During maintenance, stimulus trajectories evolve over time. Separation is stronger earlier and weaker later in the maintenance period, consistent with a code that transforms across time.
5/ Stimulus information is not stably expressed across the full delay. A key clue for us was that decoding accuracy increases with wider temporal windows. This suggests apparent persistence can emerge from temporally interleaved activity across neurons.
December 31, 2025 at 8:16 AM
5/ Stimulus information is not stably expressed across the full delay. A key clue for us was that decoding accuracy increases with wider temporal windows. This suggests apparent persistence can emerge from temporally interleaved activity across neurons.
4/ Interesting: at the single-neuron level, most concept cells do not show sustained elevation throughout maintenance. Yet when pooled across neurons, population activity can still look persistent.
December 31, 2025 at 8:13 AM
4/ Interesting: at the single-neuron level, most concept cells do not show sustained elevation throughout maintenance. Yet when pooled across neurons, population activity can still look persistent.
3/ We analyzed an open human single-unit dataset (Kyzar et al., 2024): 902 neurons from 21 neurosurgical participants performing a Sternberg WM task. We focused on stimulus-selective “concept cells.”
Thank you to the Rutishauser lab and the original authors for sharing these invaluable recordings!
Thank you to the Rutishauser lab and the original authors for sharing these invaluable recordings!
December 31, 2025 at 8:11 AM
3/ We analyzed an open human single-unit dataset (Kyzar et al., 2024): 902 neurons from 21 neurosurgical participants performing a Sternberg WM task. We focused on stimulus-selective “concept cells.”
Thank you to the Rutishauser lab and the original authors for sharing these invaluable recordings!
Thank you to the Rutishauser lab and the original authors for sharing these invaluable recordings!
2/ Persistent activity is the classic WM account, but continuous single-neuron firing is metabolically expensive and hard to reconcile with human cortical heterogeneity. So, we asked: can human WM be maintained without sustained single-neuron persistence?
December 31, 2025 at 8:11 AM
2/ Persistent activity is the classic WM account, but continuous single-neuron firing is metabolically expensive and hard to reconcile with human cortical heterogeneity. So, we asked: can human WM be maintained without sustained single-neuron persistence?
1/ Persistent activity is the classic WM account, but continuous single-neuron firing is metabolically expensive and hard to reconcile with human cortical heterogeneity. So, we asked: can human WM be maintained without sustained single-neuron persistence?
December 31, 2025 at 8:09 AM
1/ Persistent activity is the classic WM account, but continuous single-neuron firing is metabolically expensive and hard to reconcile with human cortical heterogeneity. So, we asked: can human WM be maintained without sustained single-neuron persistence?