Deep Learning Models of the Retinal Response to Natural Scenes

Authors: Lane McIntosh, Niru Maheswaranathan, Aran Nayebi, Surya Ganguli, Stephen Baccus

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We find that deep neural network models markedly outperform previous models in predicting retinal responses both for white noise and natural scenes.
Researcher Affiliation Academia Lane T. Mc Intosh 1, Niru Maheswaranathan 1, Aran Nayebi1, Surya Ganguli2,3, Stephen A. Baccus3 1Neurosciences Ph D Program, 2Department of Applied Physics, 3Neurobiology Department Stanford University {lmcintosh, nirum, anayebi, sganguli, baccus}@stanford.edu
Pseudocode No The paper contains architectural diagrams (Figure 1, Figure 2.1) but no pseudocode or explicitly labeled algorithm blocks.
Open Source Code No The paper does not provide an explicit statement about releasing source code or a link to a code repository for the described methodology.
Open Datasets No The spiking activity of a population of tiger salamander retinal ganglion cells was recorded in response to both sequences of natural images jittered with the statistics of eye movements and high resolution spatiotemporal white noise.
Dataset Splits Yes More details on the stimuli, retinal recordings, experimental structure, and division of data for training, validation, and testing are given in the Supplemental Material.
Hardware Specification Yes LM: NSF, NVIDIA Titan X Award
Software Dependencies No Optimization was performed using ADAM [20] via the Keras and Theano software libraries [21]. This lists software libraries but does not provide specific version numbers.
Experiment Setup Yes Model parameters were optimized to minimize a loss function corresponding to the negative log-likelihood under Poisson spike generation. Optimization was performed using ADAM [20] via the Keras and Theano software libraries [21]. The networks were regularized with an ℓ2 weight penalty at each layer and an ℓ1 activity penalty at the final layer, which helped maintain a baseline firing rate near 0 Hz. Models were trained over the course of 100 epochs, with early-stopping guided by a validation set.