A vectorial code for semantics in human hippocampus

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 23, 2025

ABSTRACT As we listen to speech, our brains actively compute the meaning of individual words. Inspired by success large language models (LLMs), hypothesized that brain employs vectorial coding principles, such is reflected in distributed activity single neurons. We recorded responses hundreds neurons human hippocampus, which has a well-established role semantic coding, while participants listened narrative speech. find encoding contextual word simultaneous whose selectivities span multiple unrelated categories. Like embedding vectors models, distance between neural population correlates with distance; however, this effect was only observed (like BERT) and reversed non-contextual Word2Vec), suggesting depends critically on contextualization. Moreover, for subset highly semantically similar words, even embedders showed an inverse correlation distances; attribute pattern noise-mitigating benefits contrastive coding. Finally, further support critical context, range covaries lexical polysemy. Ultimately, these results hypothesis hippocampus follows principles.

Language: Английский

---

A Neural Circuit Framework for Economic Choice: From Building Blocks of Valuation to Compositionality in Multitasking

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: March 13, 2025

Abstract Value-guided decisions are at the core of reinforcement learning and neuroeconomics, yet basic computations they require remain poorly understood mechanistic level. For instance, how does brain implement multiplication reward magnitude by probability to yield an expected value? Where within a neural circuit is indifference point for comparing types encoded? How do learned values generalize novel options? Here, we introduce biologically plausible model that adheres Dale’s law trained on five choice tasks, offering potential answers these questions. The captures key neurophysiological observations from orbitofrontal cortex monkeys generalizes offer values. Using single network solve diverse identified compositional representations—quantified via task variance analysis corroborated curriculum learning. This work provides testable predictions probe basis decision making its disruption in neuropsychiatric disorders.

Language: Английский

---

Prediction, inference, and generalization in orbitofrontal cortex

Current Biology, Journal Year: 2025, Volume and Issue: 35(7), P. R266 - R272

Published: April 1, 2025

Our understanding of the orbitofrontal cortex (OFC) has significantly evolved over past few decades. This prefrontal region been associated with a wide range cognitive functions, including popular view that it primarily signals expected value each possible option, allowing downstream areas to use these for decision-making. However, discovery rich, task-related information within OFC and its essential role in inference-based behaviors shifted our perspective led proposal holds map used by both humans animals making predictions inferences. Recent studies have further shown maps can be abstracted generalized, serving immediate future needs. In this review, we trace research journey leading evolving insights, discuss potential neural mechanisms supporting OFC's roles prediction, inference, generalization, compare hippocampus, another critical mapping, while also exploring interactions between two areas.

Language: Английский

---

The neural basis of swap errors in working memory

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Oct. 10, 2023

When making decisions in a cluttered world, humans and other animals often have to hold multiple items memory at once - such as the different on shopping list. Psychophysical experiments shown remembered stimuli can sometimes become confused, with participants reporting chimeric composed of features from stimuli. In particular, subjects will make "swap errors" where they misattribute feature one object belonging another object. While swap errors been described behaviorally, their neural mechanisms are unknown. Here, we elucidate these through trial-by-trial analysis population recordings posterior frontal brain regions while monkeys perform two multi-stimulus working tasks. tasks, were cued report color an item that either was previously corresponding location (requiring selection memory) or be attention position). Animals made both data, find evidence correlates emerged when correctly information is selected incorrectly memory. This led representation distractor if it target color, underlying eventual error. We did not consistent arose misinterpretation cue during encoding storage These results suggest alternative established views origins errors, highlight manipulation crucial yet surprisingly brittle processes.

Language: Английский

---

Multimodal subspace independent vector analysis captures latent subspace structures in large multimodal neuroimaging studies

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Sept. 17, 2023

A key challenge in neuroscience is to understand the structural and functional relationships of brain from high-dimensional, multimodal neuroimaging data. While conventional multivariate approaches often simplify statistical assumptions estimate one-dimensional independent sources shared across modalities, between true latent are likely more complex - dependence may exist within span one or dimensions. Here we present Multimodal Subspace Independent Vector Analysis (MSIVA), a methodology capture both joint unique vector multiple data modalities by defining cross-modal unimodal subspaces with variable In particular, MSIVA enables flexible estimation varying-size their one-to-one linkage corresponding modalities. As demonstrate, main benefit ability subject-level variability at voxel level subspaces, contrasting rigidity traditional methods that share same components subjects. We compared initialization baseline baseline, evaluated all three five candidate subspace structures on synthetic datasets. show successfully identified ground-truth datasets, while failed detect high-dimensional subspaces. then demonstrate better detected structure two large datasets including MRI (sMRI) (fMRI), baseline. From subsequent subspace-specific canonical correlation analysis, brain-phenotype prediction, voxelwise brain-age delta our findings suggest estimated optimal strongly associated various phenotype variables, age, sex, schizophrenia, lifestyle factors, cognitive functions. Further, modality- group-specific regions related measures such as age (e.g., cerebellum, precentral gyrus, cingulate gyrus sMRI; occipital lobe superior frontal fMRI), sex cerebellum sMRI, fMRI, precuneus sMRI schizophrenia temporal pole, operculum cortex lingual shedding light phenotypic neuropsychiatric biomarkers linked function.

Language: Английский

---

Modular representations emerge in neural networks trained to perform context-dependent tasks

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

Published: Oct. 1, 2024

Abstract The brain has large-scale modular structure in the form of regions, which are thought to arise from constraints on connectivity and physical geometry cortical sheet. In contrast, experimental theoretical work argued both for against existence specialized sub-populations neurons (modules) within single regions. By studying artificial neural networks, we show that this local modularity emerges support context-dependent behavior, but only when input is low-dimensional. No anatomical required. We also specialization at population level (different modules correspond orthogonal subspaces). Modularity yields abstract representations, allows rapid learning generalization novel tasks, facilitates related contexts. Non-modular representations facilitate unrelated Our findings reconcile conflicting results make predictions future experiments.

Language: Английский

---

A universal hippocampal memory code across animals and environments

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

Published: Oct. 24, 2024

How learning is affected by context a fundamental question of neuroscience, as the ability to generalize different contexts necessary for navigating world. An example swift contextual generalization observed in conditioning tasks, where performance quickly generalized from one another. A key identifying neural substrate underlying this how hippocampus (HPC) represents task-related stimuli across environments, given that HPC cells exhibit place-specific activity changes (remapping). In study, we used calcium imaging monitor hippocampal neuron rats performed task multiple spatial contexts. We investigated whether cells, which encode both locations (place cells) and information, could maintain their representation even when encoding remapped new context. To assess consistency representations, advanced dimensionality reduction techniques combined with machine develop manifold representations population level activity. The results showed remained stable place cell changed, thus demonstrating similar embedding geometries Notably, these patterns were not only consistent within same animal but also significantly animals, suggesting standardized or ‘neural syntax’ hippocampus. These findings bridge critical gap between memory navigation research, revealing maintains cognitive environments. suggest function governed framework shared an observation may have broad implications understanding memory, learning, related processes. Looking ahead, work opens avenues exploring principles strategies.

Language: Английский

---