Ulisse Ferrari

This paper introduces a multi-scale representational geometry framework to analyze how neural populations encode stimulus differences across scales, showing how neurons combine to encode information and how different brain areas differ.

The paper introduces CiRF-ViT, a vision transformer with per-neuron locality priors, achieving high performance in small-scale neural prediction tasks.

The paper uses AI to analyze the structural impact of 7-mer peptide insertions in AAV capsids, identifying viable modifications and their effects on capsid stability and function.

The paper develops and evaluates two synthetic fusion promoters, Pikali and Nocchu, for enhanced and balanced transgene expression in both rods and cones of retinal organoids.

The paper shows VIP+ amacrine cells in the retina both inhibit and excite specific ganglion cells, coordinating their activity through gap junctions and GABAergic synapses.

The paper introduces two information-theoretic measures, ISSI and Iloc, to quantify neuronal sensitivity in neural systems, showing they outperform Fisher information in realistic conditions.

The paper investigates how retinal ganglion cells in mice respond to different UV-to-green light balances, finding that a specific intensity ratio is necessary for reliable responses in the ventral retina, matching natural light conditions.

The paper determines the optimal sequencing depth for accurately measuring barcode concentrations in combinatorial genetic experiments by analyzing NGS data and incorporating a mathematical model of PCR amplification.

The paper shows that direction-selective retinal ganglion cells encode motion direction uniformly across all directions, despite being tuned to cardinal axes.

The paper investigates why engineered AAV variants have lower titers and finds that DNAse digestion heat reduces their stability, leading to artificially low titers.

The paper shows that strong noise correlations improve population coding in neural populations, contradicting the common belief that they should be negative.

The paper introduces ACIDES, a method that improves performance estimation in directed evolution by combining statistical inference and in-silico simulations, tested on experimental and public datasets.

The paper develops a small-correlation expansion to compute stimulus information in large neural populations, validated on synthetic data and applied to vertebrate retinal recordings.

The paper applies a time-dependent maximum entropy model to retinal ganglion cell activity, showing it better captures collective behavior for stimuli with long-range spatial correlations than the standard pairwise Ising model.

This study investigates how retinal ganglion cells respond to natural scenes, finding that their ON-OFF selectivity varies with the image and is due to non-linear retinal circuit pathways.

This study uses optogenetics and functional ultrasound imaging to show that surface stimulation of the rat visual cortex activates a large volume of V1, including deep layers and areas beyond V1.

The paper presents a method to predict population-level neural activity from sequential single-cell recordings using copula distributions and maximum entropy modeling, showing increased synchronous activity with larger populations.

This paper introduces a two-step inference strategy for training generalized linear models of noisy interacting neurons, showing improved performance and stability across complex stimuli.

This study uses functional ultrasound imaging to map retinotopic and ocular dominance columns in the visual cortex of awake non-human primates with high spatiotemporal resolution.

The paper identifies two types of slow waves in the cerebral cortex, showing they differ in neural dynamics and sensitivity to stimuli, explaining differences in memory consolidation under sleep and anesthesia.

The paper demonstrates that rod bipolar cells provide asymmetric input to direction selective ganglion cells, contributing to their direction selectivity through all-optical manipulation and imaging.

This study demonstrates that optogenetically activated ganglion cells in mouse and macaque retinas have high spatial resolution, predicting acuity above legal blindness limits.

This paper introduces a method to separate intrinsic neuron interactions from extrinsic correlations in sensory neural networks using mean-field corrections, validated on retinal recordings.

The paper introduces a simple model that corrects Poisson statistics to better predict low variability in neural spike trains, showing improved accuracy in retinal recordings.

This study uses Maximum Entropy models to analyze neural activity in human and monkey cortex, revealing that wakefulness involves pairwise interactions while sleep involves population-wide interactions.

The paper shows that a single type of retinal ganglion cell encodes two different visual features depending on stimulus context.

The paper develops a method to estimate retinal network sensitivity to stimulus perturbations and finds a peak in sensitivity at specific frequencies, suggesting optimal information transmission.

The paper compares maximum entropy and random models for describing retinal neuron spiking activity, finding maximum entropy provides better approximations, especially in larger populations.