Author name cluster

Jacob L. Yates

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

3 papers

1 author row

NeurIPS Conference 2024 Conference Paper

Poisson Variational Autoencoder

Hadi Vafaii
Dekel Galor
Jacob L. Yates

Variational autoencoders (VAE) employ Bayesian inference to interpret sensory inputs, mirroring processes that occur in primate vision across both ventral (Higgins et al. , 2021) and dorsal (Vafaii et al. , 2023) pathways. Despite their success, traditional VAEs rely on continuous latent variables, which significantly deviates from the discrete nature of biological neurons. Here, we developed the Poisson VAE (P-VAE), a novel architecture that combines principles of predictive coding with a VAE that encodes inputs into discrete spike counts. Combining Poisson-distributed latent variables with predictive coding introduces a metabolic cost term in the model loss function, suggesting a relationship with sparse coding which we verify empirically. Additionally, we analyze the geometry of learned representations, contrasting the P-VAE to alternative VAE models. We find that the P-VAE encodes its inputs in relatively higher dimensions, facilitating linear separability of categories in a downstream classification task with a much better (5x) sample efficiency. Our work provides an interpretable computational framework to study brain-like sensory processing and paves the way for a deeper understanding of perception as an inferential process.

PDF Details DOI

YNIMG Journal 2013 Journal Article

On determining the intracranial sources of visual evoked potentials from scalp topography: A reply to Kelly et al. (this issue)

Justin M. Ales
Jacob L. Yates
Anthony M. Norcia

The cruciform model posits that if a Visual Evoked Potential component originates in cortical area V1, then stimuli placed in the upper versus lower visual field will generate responses with opposite polarity at the scalp. In our original paper (Ales et al. , 2010b) we showed that the cruciform model provides an insufficient criterion for identifying V1 sources. This conclusion was reached on the basis of simulations that used realistic 3D models of early visual areas to simulate scalp topographies expected for stimuli of different sizes and shapes placed in different field locations. The simulations indicated that stimuli placed in the upper and lower visual field produce polarity inverting scalp topographies for activation of areas V2 and V3, but not for area V1. As a consequence of the non-uniqueness of the polarity inversion criterion, we suggested that past studies using the cruciform model had not adequately excluded contributions from sources outside V1. In their comment on our paper, Kelly et al. (this issue) raise several concerns with this suggestion. They claim that our initial results did not use the proper stimulus locations to constitute a valid test of the cruciform model. Kelly et al. , also contend that the cortical source of the initial visually evoked component (C1) can be identified based on latency and polarity criteria derived from intracranial recordings in non-human primates. In our reply we show that simulations using the suggested critical stimulus locations are consistent with our original findings and thus do not change our conclusions regarding the use of the polarity inversion criterion. We further show that the anatomical assumptions underlying the putatively optimal locations are not consistent with available V1 anatomical data. We then address the non-human primate data, describing how differences in stimuli across studies and species confound an effective utilization of the non-human primate data for interpreting human evoked potential responses. We also show that, considered more broadly, the non-human primate literature shows that multiple visual areas onset simultaneously with V1. We suggest several directions for future research that will further clarify how to make the best use of scalp data for inferring cortical sources.

Details DOI

YNIMG Journal 2010 Journal Article

V1 is not uniquely identified by polarity reversals of responses to upper and lower visual field stimuli

Justin M. Ales
Jacob L. Yates
Anthony M. Norcia

The cruciform hypothesis states that if a visual evoked potential component originates in V1, then stimuli placed in the upper versus lower visual fields will generate responses with opposite polarity at the scalp. This diagnostic has been used by many studies as a definitive marker of V1 sources. To provide an empirical test of the validity of the cruciform hypothesis, we generated forward models of cortical areas V1, V2 and V3 that were based on realistic estimates of the 3-D shape of these areas and the shape and conductivity of the brain, skull and scalp. Functional MRI was used to identify the location of early visual areas and anatomical MRI data was used to construct detailed cortical surface reconstructions and to generate boundary element method forward models of the electrical conductivity of each participant's head. These two data sets for each subject were used to generate simulated scalp activity from the dorsal and ventral subdivisions of each visual area that correspond to the lower and upper visual field representations, respectively. The predicted topographies show that sources in V1 do not fully conform to the cruciform sign-reversal. Moreover, contrary to the model, retinotopic visual areas V2 and V3 show polarity reversals for upper and lower field stimuli. The presence of a response polarity inversion for upper versus lower field stimuli is therefore an insufficient criterion for identifying responses as having originated in V1.

Details DOI