Key-value memory in the brain

Neuron, Journal Year: 2025, Volume and Issue: unknown

Published: March 1, 2025

Language: Английский

---

Input-driven dynamics for robust memory retrieval in Hopfield networks

Science Advances, Journal Year: 2025, Volume and Issue: 11(17)

Published: April 23, 2025

The Hopfield model provides a mathematical framework for understanding the mechanisms of memory storage and retrieval in human brain. This has inspired decades research on learning dynamics, capacity estimates, sequential transitions among memories. Notably, role external inputs been largely underexplored, from their effects neural dynamics to how they facilitate effective retrieval. To bridge this gap, we propose dynamical system which input directly influences synapses shapes energy landscape model. plasticity-based mechanism clear energetic interpretation process proves at correctly classifying mixed inputs. Furthermore, integrate within modern architectures elucidate current past information are combined during process. Last, embed both classic proposed an environment disrupted by noise compare robustness

Language: Английский

---

Robust Retrieval of Dynamic Sequences through Interaction Modulation

PRX Life, Journal Year: 2023, Volume and Issue: 1(2)

Published: Dec. 22, 2023

In this study, interaction modulation is proposed as a new paradigm for controlling transitions between functional configurations of complex biological systems, in contrast to traditional approaches based on input modulation.

Language: Английский

---

Short-term Hebbian learning can implement transformer-like attention

PLoS Computational Biology, Journal Year: 2024, Volume and Issue: 20(1), P. e1011843 - e1011843

Published: Jan. 26, 2024

Transformers have revolutionized machine learning models of language and vision, but their connection with neuroscience remains tenuous. Built from attention layers, they require a mass comparison queries keys that is difficult to perform using traditional neural circuits. Here, we show neurons can implement attention-like computations short-term, Hebbian synaptic potentiation. We call our mechanism the match-and-control principle it proposes when activity in an axon synchronous, or matched, somatic neuron synapses onto, synapse be briefly strongly potentiated, allowing take over, control, downstream for short time. In scheme, are represented as spike trains comparisons between two performed individual spines hundreds key per query roughly many there network.

Language: Английский

---

The role of astrocytic γ-aminobutyric acid in the action of inhalational anesthetics

European Journal of Pharmacology, Journal Year: 2024, Volume and Issue: 970, P. 176494 - 176494

Published: March 12, 2024

Language: Английский

---

Astrocytes as a mechanism for contextually-guided network dynamics and function

PLoS Computational Biology, Journal Year: 2024, Volume and Issue: 20(5), P. e1012186 - e1012186

Published: May 31, 2024

Astrocytes are a ubiquitous and enigmatic type of non-neuronal cell found in the brain all vertebrates. While traditionally viewed as being supportive neurons, it is increasingly recognized that astrocytes play more direct active role function neural computation. On account their sensitivity to host physiological covariates ability modulate neuronal activity connectivity on slower time scales, may be particularly well poised dynamics circuits functionally salient ways. In current paper, we seek capture these features via actionable abstractions within computational models neuron-astrocyte interaction. Specifically, engage how nested feedback loops interaction, acting over separated time-scales, endow with capability enable learning context-dependent settings, where fluctuations task parameters occur much slowly than within-task requirements. We pose general model neuron-synapse-astrocyte interaction use formal analysis characterize astrocytic modulation constitute form meta-plasticity, altering ways which synapses neurons adapt time. then embed this bandit-based reinforcement environment, show presence time-scale enables multiple fluctuating contexts. Indeed, networks learn far reliably compared dynamically homogeneous conventional non-network-based bandit algorithms. Our results fuel notion interactions benefit different time-scales conveyance task-relevant contextual information onto circuit dynamics.

Language: Английский

---

Barcode activity in a recurrent network model of the hippocampus enables efficient memory binding

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 13, 2024

Abstract Forming an episodic memory requires binding together disparate elements that co-occur in a single experience. One model of this process is neurons representing different components bind to “index” — subset unique memory. Evidence for has recently been found chickadees, which use hippocampal store and recall locations cached food. Chickadee hippocampus produces sparse, high-dimensional patterns (“barcodes”) uniquely specify each caching event. Unexpectedly, the same participate barcodes also exhibit conventional place tuning. It unknown how barcode activity generated, what role it plays formation retrieval. unclear index (e.g. barcodes) could function neural population represents content place). Here, we design biologically plausible generates uses them experiential content. Our from inputs through chaotic dynamics recurrent network Hebbian plasticity as attractor states. The matches experimental observations indices (barcodes) signals (place tuning) are randomly intermixed neurons. We demonstrate reduce interference between correlated experiences. show tuning complementary barcodes, enabling flexible, contextually-appropriate Finally, our compatible with previous models generating predictive map. Distinct indexing functions achieved via adjustment global gain. results suggest may resolve fundamental tensions specificity (pattern separation) flexible completion) general systems.

Language: Английский

---

Deep-prior ODEs augment fluorescence imaging with chemical sensors

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: Oct. 24, 2024

To study biological signalling, great effort goes into designing sensors whose fluorescence follows the concentration of chemical messengers as closely possible. However, binding kinetics are often overlooked when interpreting cell signals from resulting measurements. We propose a method to reconstruct spatiotemporal underlying in consideration process. Our fits data under constraint corresponding reactions and with help deep-neural-network prior. test it on several GCaMP calcium sensors. The recovered concentrations concur common temporal waveform regardless sensor kinetics, whereas assuming equilibrium introduces artifacts. also show that our can reveal distinct events distribution single neurons. work augments current highlights importance incorporating physical constraints computational imaging.

Language: Английский

---

Scalable bio-inspired training of deep neural networks with FastHebb

Neurocomputing, Journal Year: 2024, Volume and Issue: 595, P. 127867 - 127867

Published: May 24, 2024

Recent work on sample efficient training of Deep Neural Networks (DNNs) proposed a semi-supervised methodology based biologically inspired Hebbian learning, combined with traditional backprop-based training. Promising results were achieved various computer vision benchmarks, in scenarios scarce labeled data availability. However, current learning solutions can hardly address large-scale due to their demanding computational cost. In order tackle this limitation, contribution, we investigate novel solution, named FastHebb (FH), the reformulation rules terms matrix multiplications, which be executed more efficiently GPU. Starting from Soft-Winner-Takes-All (SWTA) and Principal Component Analysis (HPCA) rules, formulate improved FH versions: SWTA-FH HPCA-FH. We experimentally show that approach accelerates speed up 70 times, allowing us gracefully scale experiments large datasets network architectures such as ImageNet VGG.

Language: Английский

---

Beyond neurons: computer vision methods for analysis of morphologically complex astrocytes

Frontiers in Computer Science, Journal Year: 2024, Volume and Issue: 6

Published: Sept. 25, 2024

The study of the geometric organization biological tissues has a rich history in literature. However, geometry and architecture individual cells within traditionally relied upon manual or indirect measures shape. Such rudimentary are largely result challenges associated with acquiring high resolution images cellular components, as well lack computational approaches to analyze large volumes high-resolution data. This is especially true brain tissue, which composed complex array cells. Here we review tools that have been applied unravel nanoarchitecture astrocytes, type cell increasingly being shown be essential for function. Astrocytes among most structurally functionally diverse mammalian body partner neurons. Light microscopy does not allow adequate astrocyte morphology, however, large-scale serial electron data, provides nanometer 3D models, enabling visualization fine, convoluted structure astrocytes. Application computer vision methods resulting nanoscale models helping reveal organizing principles but complete understanding its functional implications will require further adaptation existing tools, development new approaches.

Language: Английский

---