Search

How sensory systems rapidly adapt to changing stimulus statistics remains unclear. Here the authors show that gain adaptation in recurrent networks can implement fast efficient coding, unifying prior attraction and adapter repulsion, and supporting adaptive behavior.

Whether large language models (LLMs) process language in a human-like manner remains unclear. Here, the authors re-examine prior LLM-to-brain mapping results, demonstrating that LLM-brain alignment can result from non-robust methods.

Perseveration – repeating one choice when others would generate larger rewards – is a common behavior, but neither its purpose nor neuronal mechanisms are understood. Here the authors demonstrate a neural correlate and causal role of dorsal prefrontal cortex, specifically anterior supplementary motor cortex, in perseveration in mice performing a dynamic reward learning task.

Representational similarity analysis of human brain fMRI during natural scene viewing reveals two cortical routes: a ventromedial pathway for scene context and layout, and a lateral occipitotemporal pathway tuned to animate content.

Decision-making is influenced by subjective reference points. Here, choices are framed as potential gains or losses relative to a cumulative token count, revealing a gain-related substrate of reference-dependency in dorsal anterior cingulate cortex.

During rapid behavioural switches in flying bats, hippocampal neurons can rapidly switch their core computation to represent the relevant behavioural variables, supporting behavioural flexibility.

Whether orientation-selectivity is discernable via fMRI remains unclear. Here, by analyzing a public dataset of responses to natural scenes using neurally-inspired image-computable models, the authors isolate and characterize a coarse-scale orientation map and demonstrate that orientation-selective BOLD responses reflect multiple distinct computations at a range of spatial scales.

It has been proposed that language meaning is represented throughout the cerebral cortex in a distributed ‘semantic system’, but little is known about the details of this network; here, voxel-wise modelling of functional MRI data collected while subjects listened to natural stories is used to create a detailed atlas that maps representations of word meaning in the human brain.

Face cells in the macaque inferotemporal cortex are initially able to detect faces and then rapidly switch to a face-specific neural code to discriminate between different face identities.

This study uses the inception loop framework to map neuronal invariances in mouse V1, revealing a bipartite receptive-field organization linked to segmentation and a synaptic-level hierarchy of increasing invariance supported by the MICrONS dataset.

The brain generates high-dimensional representations of complex sensory environments and concurrently predicts expected stimuli. Here the authors show that neural circuits that perform these computations exhibit desegregated representations of sensory stimuli and prediction errors.

Doerig, Kietzmann and colleagues show that the brain’s response to visual scenes can be modelled using language-based AI representations. By linking brain activity to caption-based embeddings from large language models, the study reveals a way to quantify complex visual understanding.

Speech is encoded by the firing patterns of speech-controlling neurons in different regions of the brain, which Tankus and colleagues analyse in this study. They find highly specific encoding of vowels in medial–frontal neurons and nonspecific tuning in superior temporal gyrus neurons.

Wingert, Parida and colleagues measured tuning subspaces from deep-learning models trained on single neurons in auditory cortex. They show that subspaces distinguish functional properties between neuronal subtypes and describe a framework for sparse, efficient coding of natural sounds.

Keshishian, Mischler et al. report that a recurrent automatic speech recognition system aligns closely with brain organization: model layers map to distinct cortical regions and naturally learn to encode a parallel progression from acoustic to phonetic, lexical and semantic content.

The authors analytically determine how neuronal correlations and geometry collectively determine readout generalization across tasks and show how these geometric features follow distinct trajectories over the course of learning.

In this study, the authors present that expected-sound omissions in mouse auditory cortex evoked distinct, time-locked activity in layers 1–4 of the Temporal Association Area, suggesting a higher-order, integrated prediction error.

How the pulvinar represents complex visual information, its functional topography, and its relationship to cortical processing of visually presented objects remains unclear. Here authors show that responses to natural scenes in the human pulvinar reveal organized spatial maps for both low-level visual features, such as local contrast, as well as high-level visual features, such as bodies and faces.

EEG-aligned fine-tuning makes artificial vision models more brain-like, enhancing model-human representational similarity across EEG, fMRI, and behavior.

How the brain processes unexpected events is not fully understood. Here authors show few unexpected spikes in one cortical neuron ripple across the local network and beyond, broadcasting surprise. This scale-free spread, with a reliable source trace, reflects critical brain dynamics that help cortex quickly detect novel events.

How neuron-level interactions produce complex cognitive behavior remains unclear. Here, the authors develop a brain circuit mechanistic model based on physiological computation, that uncovers an unexpected neural code, subsequently validated by empirical data.

Aligning electrocorticography data into a shared space improves how large language models predict brain activity during language comprehension, enhancing encoding accuracy, cross-participant generalization and denoising—especially in language-selective regions.

The way in which the brain processes language from a collection of sounds to meaningful concepts remains poorly understood. Here, the authors show that the brain’s temporal responses to speech closely follow the layer-by-layer progression of LLMs, revealing shared computational principles.

Aoi et al. used a new dimensionality-reduction method to disentangle the contributions of different task variables to neural population activity, which revealed rotational dynamics in monkey PFC during context-dependent decision-making.

A shared neural feature space encoding self-generated autobiographical imagery and externally driven sentence semantics is revealed by decoding imagined autobiographical content from fMRI data with a model trained on semantic feature representations.

Learning shifts multi-task representations from compositional to conjunctive formats. Cortical conjunctions uniquely associate with effects of practice, and index switch costs. Subcortex is critical for cross-region binding of task rule information.

The human superior temporal gyrus processes acoustic–phonetic properties of speech regardless of whether the language is familiar to the listener, but only encodes word boundaries and language-specific sound sequences if the language is known.

This study presents a brain rhythm-based inference model (BRyBI) for speech processing in the auditory cortex. BRyBI shows how rhythmic neural activity enables robust speech processing by dynamically predicting context and elucidates mechanistic principles that allow robust speech parsing in the brain.

Temporal integration throughout the human auditory cortex is predominantly locked to absolute time and does not vary with the duration of speech structures such as phonemes or words.

Appeals to representation are widespread, despite neuroscientists’ uncertainty about what kind of findings count as evidence for such claims. In this Perspective, Pohl and colleagues develop a unified framework that distinguishes four conceptual dimensions relevant to representation, illustrating them in information-theoretic terms to explicitly characterize representation in neuroscience.

Species-specific gaze shifts with natural stimuli drive correlated spiking in retinal ganglion cells.

Separating firing rate from spiking irregularity is a key challenge in analyzing neural activity. Here, the authors present a mathematical model and inference method that capture diverse spike patterns across neurons, cortical areas, and cognitive states.

Using in silico neuroscience, Gifford et al. developed a neural control algorithm to modulate the representational relationships between visual cortical areas, revealing how these areas jointly represent the world as an interconnected network.

Yamashita et al. explore how conversational content is represented in the brain, revealing shared and distinct brain activity patterns for speech production and comprehension, with contrasting timescale properties between the two processes.

Multimodal large language models are shown to develop object concept representations similar to those of humans. These representations closely align with neural activity in brain regions involved in object recognition, revealing similarities between artificial intelligence and human cognition.

Here, the authors show that Intracortical recordings in human motor cortex during attempted multi-finger movements reveal a largely linear code with two violations: magnitude normalization and greater shifts in tuning direction for weakly represented fingers.

Correlated neural variability is prominent in the visual cortex, yet its role in neural coding is debated. Here, the authors show that covariability is modulated by spatial context and that it encodes uncertainty in a probabilistic representation of natural scenes.

The role of brain interactions in encoding goal-directed learning signals remains unclear. Here, the authors show that information gain is encoded through synergistic and higher-order interactions and is broadcast to the prefrontal reward circuitry.

Whether temporal code and rate code have different rates of representational drift over extended periods is not fully understood. Using ultraflexible electrodes, here authors show that temporal codes extracted from fast spiking patterns reduce visual representational drift compared to firing rates over 15 consecutive days in mice.

How natural scenes are represented by the neuronal populations of a specific visual area such as V4 remain not fully understood. The authors produced a dataset of widefield calcium imaging of macaque V4 responses to a large set of natural images, and used deep learning techniques to elucidate how natural image features are encoded and topologically organized in V4.

A population code for the dynamics of choice formation in the primate premotor cortex is revealed, with diverse single-neuron tuning to a shared decision variable.

This study reveals how the human brain integrates contextual information differently from Large Language Models. A model that combines short-term and long-term context is introduced, improving predictions of neural activity in higher-order brain regions.

People are capable of making near-optimal decisions in volatile, changing environments. Here, the authors show how two neural decision signals encode distinct aspects of the belief updating process underpinning such adaptive decisions.

In decision-making, choice stochasticity can arise at multiple stages in the brain. Here, the authors show that noise arising early and late relative to a nonlinear neural computation leads to opposing contextual effects on choice accuracy.

Accurate motion perception depends on accurate estimation of retinal motion speed. Here, from natural image movies, the authors derive the optimal computational rules for estimating speed, and show that these computations predict both human speed discrimination performance and the tuning of speed-selective neurons.

The visual signals transmitted by the retina to the brain are affected by random drift in eye position, but the impact of this on visual capabilities is not clear. Here, the authors show that the decoding of images from evoked spike trains recorded in the macaque retina improves with fixational eye movements, even when the eye position is unknown.