Biology subjects

DiCarlo, J.

Publications and source records attributed to DiCarlo, J..

9 recordsLinked to original sources

bioRxiv2026

An image-computable characterization of the non-conditioned linkage of visual drive and valence in the primate amygdala

The amygdala is a key node in linking high-level representations of visual stimuli to affective state, and much research has focused on its role in learning artificial stimulus-value associations. By comparison, the primate amygdalas encoding of natural, non-conditioned visual stimuli is less well understood. Here, we report that some amygdala neurons have visual selectivity that is systematically linked to their valence tuning, without laboratory conditioning. First, electrophysiological recordings revealed that the firing rate responses of many amygdala recording sites are selective across arbitrary non-conditioned natural visual stimuli. Second, this selectivity was well-predicted by contemporary, image-computable models of the visually-driven selectivity of high-level ventral stream neurons that provide major afferents to the amygdala. Third, a subpopulation of visually selective amygdala units also coded valence, as defined in prior work. Fourth, these valence preferences were correlated with their visual tuning, in that the image-computable models predicted which visual stimuli tended to drive positive-valence sites and which tended to drive negative-valence sites. Taken together, these results suggest that the visual drive provided from the ventral visual stream into the amygdala is strong, and that it is not unlinked or randomly linked to valence coding. Furthermore, the results establish baseline image-computable models of visual encoding in the primate amygdala. SignificanceOur understanding of how the primate amygdala links natural visual input to affective state remains limited. Here, we report that visual selectivity in the amygdala is not random but is systematically coupled to valence encoding, even without explicit conditioning. Amygdala neurons exhibit reliable visual selectivity for natural stimuli, and this selectivity is predicted by image-computable models of the upstream ventral visual system. Interestingly, a subpopulation of these visually selective units also coded valence, and their valence preference correlated directly with their visual preferences. This establishes a predictive model of visual and valence encoding, suggesting that visual information is wired to influence affective processing and potentially enabling noninvasive modulation of limbic circuits.

neuroscience↗

bioRxiv2026

Predictive coding video models capture dorsal parietal representations and human judgments for surfaces defined by motion

Interacting with the world relies on knowing its three-dimensional structure, yet vision receives a stream of two-dimensional images. Recovering 3D geometry from dynamic input is therefore foundational to vision, but lacks a computational account. Here we identify one, using artificial neural networks to predict human behavior and neuronal responses from the two major visual pathways in macaques. Using videos of camouflaged objects whose shape is revealed through motion, we found both pathways carry information for motion-defined surfaces, while neurons in parietal cortex most closely track what humans perceive. Only predictive coding video models--trained to fill in missing content in natural videos--reproduced both neuronal responses and human behavior. Predictive coding thus emerges as a principle that yields a brain-like geometric world model of the physical world.

neuroscience↗

bioRxiv2022

Aligning Model and Macaque Inferior Temporal Cortex Representations Improves Model-to-Human Behavioral Alignment and Adversarial Robustness

While some state-of-the-art artificial neural network systems in computer vision are strikingly accurate models of the corresponding primate visual processing, there are still many discrepancies between these models and the behavior of primates on object recognition tasks. Many current models suffer from extreme sensitivity to adversarial attacks and often do not align well with the image-by-image behavioral error patterns observed in humans. Previous research has provided strong evidence that primate object recognition behavior can be very accurately predicted by neural population activity in the inferior temporal (IT) cortex, a brain area in the late stages of the visual processing hierarchy. Therefore, here we directly test whether making the late stage representations of models more similar to that of macaque IT produces new models that exhibit more robust, primate-like behavior. We conducted chronic, large-scale multi-electrode recordings across the IT cortex in six non-human primates (rhesus macaques). We then use these data to fine-tune (end-to-end) the model "IT" representations such that they are more aligned with the biological IT representations, while preserving accuracy on object recognition tasks. We generate a cohort of models with a range of IT similarity scores validated on held-out animals across two image sets with distinct statistics. Across a battery of optimization conditions, we observed a strong correlation between the models IT-likeness and alignment with human behavior, as well as an increase in its adversarial robustness. We further assessed the limitations of this approach and find that the improvements in behavioral alignment and adversarial robustness generalize across different image statistics, but not to object categories outside of those covered in our IT training set. Taken together, our results demonstrate that building models that are more aligned with the primate brain leads to more robust and human-like behavior, and call for larger neural data-sets to further augment these gains.

neuroscience↗

bioRxiv2020

Predicting primer and panel off-target rate in QIAseq targeted DNA panels using convolutional neural networks

In QIAseq targeted DNA panels, synthetic primers (short single-strand DNA sequences) are used for target enrichment via complementary DNA binding. Off-target priming could occur in this process when a primer binds to some loci where the DNA sequences are identical or very similar to the target template. These off-target DNA segments go through the rest of the workflow, wasting sequencing resources in unwanted regions. Off-target cannot be avoided if some segments of the target region are repetitive throughout the genome, nor can it be quantified until after sequencing. But if off-target rates can be prospectively predicted, scientists can make informed decisions about investment on high off-target panels. We developed pordle (predicting off-target rate with deep learning and epcr07), a convolutional neural network (CNN) model to predict off-target binding events of a given primer. The neural network was trained using 10 QIAseq DNA panels with 29,274 unique primers and then tested on an independent QIAseq panel with 7,576 primers. The model predicted a 10.5% off-target rate for the test panel, a -0.1% bias from the true value of 10.6%. The model successfully selected the better primer (in terms of off-target rate) for 89.2% of 3,835 pairs of close-by primers in the test panel whose off-target rates differ by at least 10%. The order-preserving property may help panel developers select the optimal primer from a group of candidates, which is a common task in panel design.

bioinformatics↗

bioRxiv2020

Unsupervised Neural Network Models of the Ventral Visual Stream

Deep neural networks currently provide the best quantitative models of the response patterns of neurons throughout the primate ventral visual stream. However, such networks have remained implausible as a model of the development of the ventral stream, in part because they are trained with supervised methods requiring many more labels than are accessible to infants during development. Here, we report that recent rapid progress in unsupervised learning has largely closed this gap. We find that neural network models learned with deep unsupervised contrastive embedding methods achieve neural prediction accuracy in multiple ventral visual cortical areas that equals or exceeds that of models derived using todays best supervised methods, and that the mapping of these neural network models hidden layers is neuroanatomically consistent across the ventral stream. Moreover, we find that these methods produce brain-like representations even when trained on noisy and limited data measured from real childrens developmental experience. We also find that semi-supervised deep contrastive embeddings can leverage small numbers of labelled examples to produce representations with substantially improved error-pattern consistency to human behavior. Taken together, these results suggest that deep contrastive embedding objectives may be a biologically-plausible computational theory of primate visual development.

neuroscience↗

bioRxiv2020

Wiring Up Vision: Minimizing Supervised Synaptic Updates Needed to Produce a Primate Ventral Stream

After training on large datasets, certain deep neural networks are surprisingly good models of the neural mechanisms of adult primate visual object recognition. Nevertheless, these models are poor models of the development of the visual system because they posit millions of sequential, precisely coordinated synaptic updates, each based on a labeled image. While ongoing research is pursuing the use of unsupervised proxies for labels, we here explore a complementary strategy of reducing the required number of supervised synaptic updates to produce an adult-like ventral visual stream (as judged by the match to V1, V2, V4, IT, and behavior). Such models might require less precise machinery and energy expenditure to coordinate these updates and would thus move us closer to viable neuroscientific hypotheses about how the visual system wires itself up. Relative to the current leading model of the adult ventral stream, we here demonstrate that the total number of supervised weight updates can be substantially reduced using three complementary strategies: First, we find that only 2% of supervised updates (epochs and images) are needed to achieve ~80% of the match to adult ventral stream. Second, by improving the random distribution of synaptic connectivity, we find that 54% of the brain match can already be achieved "at birth" (i.e. no training at all). Third, we find that, by training only ~5% of model synapses, we can still achieve nearly 80% of the match to the ventral stream. When these three strategies are applied in combination, we find that these new models achieve ~80% of a fully trained models match to the brain, while using two orders of magnitude fewer supervised synaptic updates. These results reflect first steps in modeling not just primate adult visual processing during inference, but also how the ventral visual stream might be "wired up" by evolution (a models "birth" state) and by developmental learning (a models updates based on visual experience).

neuroscience↗

bioRxiv2018

Neural Population Control via Deep Image Synthesis

Particular deep artificial neural networks (ANNs) are todays most accurate models of the primate brains ventral visual stream. Here we report that, using a targeted ANN-driven image synthesis method, new luminous power patterns (i.e. images) can be applied to the primate retinae to predictably push the spiking activity of targeted V4 neural sites beyond naturally occurring levels. More importantly, this method, while not yet perfect, already achieves unprecedented independent control of the activity state of entire populations of V4 neural sites, even those with overlapping receptive fields. These results show how the knowledge embedded in todays ANN models might be used to non-invasively set desired internal brain states at neuron-level resolution, and suggest that more accurate ANN models would produce even more accurate control.

neuroscience↗

bioRxiv2018

smCounter2: an accurate low-frequency variant caller for targeted sequencing data with unique molecular identifiers

MotivationLow-frequency DNA mutations are often confounded with technical artifacts from sample preparation and sequencing. With unique molecular identifiers (UMIs), most of the sequencing errors can be corrected. However, errors before UMI tagging, such as DNA polymerase errors during end-repair and the first PCR cycle, cannot be corrected with single-strand UMIs and impose fundamental limits to UMI-based variant calling.\n\nResultsWe developed smCounter2, a UMI-based variant caller for targeted sequencing data and an upgrade from the current version of smCounter. Compared to smCounter, smCounter2 features lower detection limit at 0.5%, better overall accuracy (particularly in non-coding regions), a consistent threshold that can be applied to both deep and shallow sequencing runs, and easier use via a Docker image and code for read pre-processing. We benchmarked smCounter2 against several state-of-the-art UMI-based variant calling methods using multiple datasets and demonstrated smCounter2s superior performance in detecting somatic variants. At the core of smCounter2 is a statistical test to determine whether the allele frequency of the putative variant is significantly above the background error rate, which was carefully modeled using an independent dataset. The improved accuracy in non-coding regions was mainly achieved using novel repetitive region filters that were specifically designed for UMI data.\n\nAvailabilityThe entire pipeline is available at https://github.com/qiaseq/qiaseq-dna under MIT license.

bioinformatics↗

bioRxiv2017

High-throughput creation and functional profiling of eukaryotic DNA sequence variant libraries using CRISPR/Cas9

Construction of genetic variant libraries with phenotypic measurement is central to advancing todays functional genomics, and remains a grand challenge. Here, we introduce a Cas9-based approach for generating pools of mutants with defined genetic alterations (deletions, substitutions and insertions), along with methods for tracking their fitness en masse. We demonstrate the utility of our approach in performing focused analysis of hundreds of mutants of a single protein and in investigating the biological function of an entire family of poorly characterized genetic elements. Our platform allows fundamental biology questions to be investigated in a quick, easy and affordable manner.

bioengineering↗