Decoding naturalistic stimuli from neural recordings is a significant challenge in systems neuroscience, primarily due to the high-dimensional and nonlinear nature of stimulus-response interactions, and is further exacerbated by the limited availability and noisiness of neural data. While contemporary approaches that incorporate generative models, such as Generative Adversarial Networks (GANs), attempt to address these issues by mapping neural responses to latent representations, they do not fully overcome these obstacles. In this work, we present a novel paradigm of differential neural decoding (dicoding) that focuses on the relative changes in response patterns, which not only expands the neural training data quadratically but also inherently denoises it. To determine the corresponding stimulus changes, this method leverages the Euclidean and feature-disentangled properties of the underlying latents through vector arithmetic. As such, we not only effectively exploit the latent space but also achieve semantically meaningful latent offsets in the context of the stimuli, resulting in improved sample efficiency. We trained a decoder to predict changes in latent vectors based on the corresponding changes in neural responses. The absolute latent vector itself was derived by vector addition of the predicted latent change (indicative of stimulus shift) to a reference latent, which was fed to the generator for the reconstruction of the perceived stimulus. Our results show that this geometrically principled approach facilitates more effective reconstruction of naturalistic stimuli from noisy and limited neural data.

To facilitate language comprehension, the brain uses contextual information and prior knowledge to predict future content. Recent breakthroughs allow us to study pre-word onset prediction during naturalistic narrative listening by mapping contextual word embeddings from Large Language Models onto ECoG data. Long-range prediction encoding has been observed in fMRI data, where including multiple upcoming word embeddings enhances the model's fit to brain data. This study examines if similar predictive information is detectable in MEG data, which offers higher temporal resolution than fMRI but lower signal-to-noise ratio than ECoG. We found that pre-onset predictive signatures are present in MEG, even in data of limited length (1 hour) and in single participants. Unlike in fMRI, adding future embeddings does not improve encoding. These findings offer a novel avenue for studying predictive processing using MEG signals and call for further investigation to explain the differences observed between fMRI and MEG approaches.

Our senses receive a continuous stream of complex information. According to Event Segmentation Theory (EST), we parse this information into meaningful events, allowing us to extract relevant information, remember it, and act upon it. Previous research has related these events to so-called ‘neural states': temporally and regionally specific stable patterns of brain activity, which tend to coincide with events in the stimulus. Here we show that these neural states additionally align with stable features in a movie stimulus that are relevant to a specific brain region. This supports the idea that many brain areas across the cortex apply event segmentation in a hierarchical manner. Using intracranial measurements, we further investigate whether neural states are present at a much smaller timescale and how their characteristics correspond to EST. Our findings provide support for the idea that neural states could underlie the cognitive skill of event segmentation.