Sleep for Enhanced Learning

Overseen By Rishi Krishnamurthy, BA

Age: 18 - 65

Sex: Any

Trial Phase: Academic

Sponsor: University of Pennsylvania

Must not be taking: Antidepressants, Sedatives

Disqualifiers: MR contraindications, Claustrophobia, Pregnancy, others

No Placebo Group All trial participants will receive the active study treatment (no placebo)

Compensation: Varies

Number of Visits: 3

105 Participants Needed

Trial Summary

Will I have to stop taking my current medications?

If you are currently taking antidepressants or sedatives, you will need to stop taking them to participate in this trial.

What data supports the effectiveness of the treatment Associative inference, Category learning, Sleep for enhanced learning?

Research shows that sleep helps consolidate memories and improve learning by replaying experiences and generating new insights. Studies indicate that sleep can enhance the ability to learn new vocabulary and improve memory performance, suggesting that sleep is crucial for associative learning and category formation. ^{1 2 3 4 5}

How does the Sleep for Enhanced Learning treatment differ from other treatments for learning enhancement?

This treatment is unique because it leverages sleep to enhance learning by reinforcing memory consolidation and forming new associations, unlike other methods that focus on active learning during wakefulness. Sleep's recurring brain oscillations help process information hierarchically, improving generalization and adaptive behavior. ^{5 6 7 8 9}

What is the purpose of this trial?

Acting adaptively requires quickly picking up on structure in the environment and storing the acquired knowledge for effective future use. Dominant theories of the hippocampus have focused on its ability to encode individual snapshots of experience, but the investigators and others have found evidence that it is also crucial for finding structure across experiences. The mechanisms of this essential form of learning have not been established. The investigators have developed a neural network model of the hippocampus instantiating the theory that one of its subfields can quickly encode structure using distributed representations, a powerful form of representation in which populations of neurons become responsive to multiple related features of the environment. The first aim of this project is to test predictions of this model using high resolution functional magnetic resonance imaging (fMRI) in paradigms requiring integration of information across experiences. The results will clarify fundamental mechanisms of how humans learn novel structure, adjudicating between existing models of this process, and informing further model development. There are also competing theories as to the eventual fate of new hippocampal representations. One view posits that during sleep, the hippocampus replays recent information to build longer-term distributed representations in neocortex. Another view claims that memories are directly and independently formed and consolidated within the hippocampus and neocortex. The second aim of this project is to test between these theories. The investigators will assess changes in hippocampal and cortical representations over time by re-scanning participants and tracking changes in memory at a one-week delay. Any observed changes in the brain and behavior across time, however, may be due to generic effects of time or to active processing during sleep. The third aim is thus to assess the specific causal contributions of sleep to the consolidation of structured information. The investigators will use real-time sleep electroencephalography to play sound cues to bias memory reactivation. The investigators expect that this work will clarify the anatomical substrates and, critically, the nature of the representations that support encoding and consolidation of novel structure in the environment.

Research Team

Anna C Schapiro, PhD

Principal Investigator

University of Pennsylvania

Eligibility Criteria

This trial is for healthy adults aged 18-35 with normal or corrected vision, normal hearing, fluent English skills, and no history of major psychiatric/neurological disorders. It excludes vulnerable populations, those on antidepressants/sedatives (for MRI), anyone with neurological disorders (for EEG), MR contraindications like metal implants, claustrophobia (MRI-specific), and pregnant women.

Inclusion Criteria

I am not currently taking any antidepressants or sedatives.

Not a member of a vulnerable population

My vision is normal or corrected to normal.

Exclusion Criteria

Claustrophobia (Aims 1 and 2; MRI-specific)

Individuals with MR contraindications such as non-removable biomedical devices or metal in or on the body (Aims 1 and 2; MRI-specific)

Pregnant women will also be excluded from neuroimaging, as the effects of MR on pregnancy are not fully understood (Aims 1 and 2; MRI-specific)

Timeline

Screening

Participants are screened for eligibility to participate in the trial

1-2 weeks

Initial Learning and fMRI Assessment

Participants undergo fMRI to assess neural representations during learning tasks involving object associations and category learning.

2-3 hours

1 visit (in-person)

Sleep EEG and Memory Reactivation

Participants' memory is tested before and after a nap, with EEG monitoring and auditory cues to encourage memory reactivation.

4-5 hours

1 visit (in-person)

Follow-up fMRI Assessment

Participants return for a second fMRI scan to evaluate changes in neural substrates and memory consolidation after one week.

1-2 hours

1 visit (in-person)

Follow-up

Participants are monitored for changes in memory and neural representations over time.

1 week

Treatment Details

Interventions

• Associative inference
• Category learning
• Sleep

Trial Overview The study investigates how humans learn structured information over time and the role of sleep in this process. Using high-resolution fMRI to track brain activity and real-time sleep EEG to influence memory reactivation, it aims to understand hippocampal functions in learning and memory consolidation.

Participant Groups

Group I: Manipulating replay during sleep using real-time EEG Experimental Treatment 2 Interventions

In the proposed electroencephalography (EEG) study, all participants will undergo the same procedure. Participants will learn the visual features and spoken names associated with three categories of novel objects. Participants' memory for these objects and the objects' parts will be tested before and after a nap. The investigators will monitor brain activity during the nap in real time and, at optimal moments, quietly play the spoken names of the objects to encourage reactivation of particular objects in particular orders. The investigators will assess how this manipulation impacts memory for these objects.

Group II: Learning and consolidation in category learning Experimental Treatment 1 Intervention

The proposed functional magnetic resonance imaging study assesses the neural representations contributing to humans' ability to learn new categories of objects. All participants will undergo the same procedure. Participants will learn about novel objects, each with several colored parts. Some parts are unique to individual objects and others are shared among the members of the category. The investigators will assess how different regions of the brain contribute to learning and remembering these different kinds of parts, and how the resulting representations support category understanding. Participants will be brought back one week later for a second scan, to evaluate how the neural substrates of these processes change with consolidation.

Group III: Learning and consolidation in Associative Inference Experimental Treatment 1 Intervention

The proposed functional magnetic resonance imaging study assesses the neural representations contributing to humans' ability to associate objects in the support of simple inferences and generalization. All participants will undergo the same procedure. Participants will learn about pairs of objects and then be asked to make judgments and inferences about the relationships between the objects. The order of presentation of the objects will be manipulated within subjects, as different learning theories make different predictions about how learning will unfold under different orderings. Participants will be brought back one week later for a second scan, to evaluate how the neural substrates of these processes change with consolidation.

Who Is Running the Clinical Trial?

University of Pennsylvania

Lead Sponsor

Trials

2,118

Recruited

45,270,000+

National Institute of Mental Health (NIMH)

Collaborator

Trials

3,007

Recruited

2,852,000+

Findings from Research

The study presents an unsupervised learning algorithm for a multilayer network of stochastic neurons that uses both bottom-up recognition and top-down generative connections to process and reconstruct data representations.

The algorithm operates in two phases: a 'wake' phase where recognition connections help interpret input data, and a 'sleep' phase where generative connections refine the model, enhancing its ability to accurately reconstruct data across layers.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

The "wake-sleep" algorithm for unsupervised neural networks. Hinton, GE., Dayan, P., Frey, BJ., et al.[2019]

The study proposes that different sleep states, particularly REM and NREM, play distinct roles in learning: REM sleep facilitates the generation of new sensory inputs, enhancing the extraction of semantic concepts, while NREM sleep focuses on replaying memories to strengthen learned representations.

Using a model inspired by generative adversarial networks (GANs), the research demonstrates that these offline states optimize learning processes, suggesting a novel computational framework for understanding how sleep contributes to memory and concept formation.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Learning cortical representations through perturbed and adversarial dreaming. Deperrois, N., Petrovici, MA., Senn, W., et al.[2022]

Participants who heard unknown Japanese words during sleep were able to identify images related to those words after waking, demonstrating that new associative memories can form during sleep (N = 22).

Electroencephalography revealed that specific brain responses during sleep predicted how well participants would remember the information upon waking, indicating that slow-wave activity plays a crucial role in memory formation.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Learning New Vocabulary Implicitly During Sleep Transfers With Cross-Modal Generalization Into Wakefulness. Koroma, M., Elbaz, M., Léger, D., et al.[2023]

The study developed a thalamo-cortical model that mimics brain states of wakefulness and deep sleep, demonstrating the brain's ability to learn quickly from limited and noisy examples.

The model showed that sleep enhances visual classification abilities by promoting synaptic homeostasis and helping to associate similar memories, highlighting the importance of sleep in cognitive functions.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Thalamo-cortical spiking model of incremental learning combining perception, context and NREM-sleep. Golosio, B., De Luca, C., Capone, C., et al.[2021]

Sleep plays a crucial role in enhancing learning and memory by facilitating the reactivation of neural patterns in the hippocampus, which is essential for long-term memory consolidation.

The study suggests that sleep's nested oscillations support hierarchical processing of information, allowing the brain to better infer causal relationships and improve predictive abilities, ultimately leading to more adaptive behaviors.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Oscillations, neural computations and learning during wake and sleep. Penagos, H., Varela, C., Wilson, MA.[2018]

Humans can learn new information during sleep, as shown by brain activity changes during auditory-olfactory associative learning.

Different brain wave patterns are involved at various stages of sleep-learning: delta and sigma waves are linked to the early formation of new associations, while theta waves are associated with later stages of the learning process.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Neural Dynamics of Associative Learning during Human Sleep. Canales-Johnson, A., Merlo, E., Bekinschtein, TA., et al.[2021]

The proposed model extends the standard Hopfield network by incorporating both on-line learning (for storing new information) and off-line unlearning (for removing errors), achieving the theoretical maximum storage capacity of α=1, which is a significant improvement over the traditional model's capacity of α∼0.14.

The model demonstrates robustness against thermal noise and provides explicit guidelines for synaptic plasticity and the necessary 'sleep rate' to ensure effective learning and pattern reinforcement, validated through extensive numerical simulations.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Dreaming neural networks: Forgetting spurious memories and reinforcing pure ones. Fachechi, A., Agliari, E., Barra, A.[2019]

The study presents a linear network model for factor analysis that uses hidden variables to understand correlations between visible variables, which can be learned through a simple wake-sleep learning method.

This approach offers a straightforward alternative to Hebbian learning for modeling how the brain adapts and changes based on activity, suggesting it could effectively represent cortical plasticity.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Factor analysis using delta-rule wake-sleep learning. Neal, RM., Dayan, P.[2019]

Healthy adults trained on a probabilistic category-learning task showed improved performance after a night of sleep, indicating that sleep enhances learning.

The study compared two training modes (feedback and observational) but found that both types benefited from sleep, suggesting a universal mechanism for sleep-related learning enhancement.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Sleep enhances category learning. Djonlagic, I., Rosenfeld, A., Shohamy, D., et al.[2022]

References

1. United States pubmed.ncbi.nlm.nih.gov

The "wake-sleep" algorithm for unsupervised neural networks. [2019]

2. England pubmed.ncbi.nlm.nih.gov

Learning cortical representations through perturbed and adversarial dreaming. [2022]

3. Switzerland pubmed.ncbi.nlm.nih.gov

Learning New Vocabulary Implicitly During Sleep Transfers With Cross-Modal Generalization Into Wakefulness. [2023]

4. United States pubmed.ncbi.nlm.nih.gov

Thalamo-cortical spiking model of incremental learning combining perception, context and NREM-sleep. [2021]

5. England pubmed.ncbi.nlm.nih.gov

Oscillations, neural computations and learning during wake and sleep. [2018]

6. United States pubmed.ncbi.nlm.nih.gov

Neural Dynamics of Associative Learning during Human Sleep. [2021]

7. United States pubmed.ncbi.nlm.nih.gov

Dreaming neural networks: Forgetting spurious memories and reinforcing pure ones. [2019]

8. United States pubmed.ncbi.nlm.nih.gov

Factor analysis using delta-rule wake-sleep learning. [2019]

9. United States pubmed.ncbi.nlm.nih.gov

Sleep enhances category learning. [2022]