Recognizing retinal ganglion cells in the dark
Authors: Emile Richard, Georges A. Goetz, E.J. Chichilnisky
NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In the present paper, we introduce two novel and efficient computational methods for cell type identification in a neural circuit, using spatiotemporal voltage signals produced by spiking cells recorded with a high-density, large-scale electrode array [14]. We describe the data we used for our study in Section 2, and we show how the raw descriptors used by our classifiers are extracted from voltage recordings of a primate retina. We then introduce a classifier that leverages both handspecified and random-projection based features of the electrical signatures of unique RGCs, as well as large unlabeled data sets, to identify cell types (Section 3). We evaluate its performance for distinguishing between midget, parasol and small bistratified cells on manually annotated datasets. Then, in Section 4, we show how matrix completion techniques can be used to identify populations of unique cell types, and assess the accuracy of our algorithm by predicting the polarity (ON or OFF) of RGCs on datasets where a ground truth is available. Section 5 is devoted to numerical experiments that we designed to test our modeling choices. |
| Researcher Affiliation | Academia | Emile Richard Stanford University emileric@stanford.edu Georges Goetz Stanford University ggoetz@stanford.edu E.J. Chichilnisky Stanford University ej@stanford.edu |
| Pseudocode | Yes | Algorithm 1 Polarity matrix completion |
| Open Source Code | No | The paper does not provide a direct link or explicit statement about the availability of the source code for the methodology described. |
| Open Datasets | No | The labeled data consists of 436 OFF midget, 652 OFF parasol, 964 ON midget, 607 ON parasol and 169 small bistratified cells assembled from 10 distinct recordings. |
| Dataset Splits | Yes | Our numerical experiment consists in hiding one out of 10 labeled recordings, learning cell classifiers on the 9 others and testing the classifier on the hidden recording. |
| Hardware Specification | Yes | Using a NVIDIA Tesla K40 GPU drastically accelerated these steps, allowing us to scale up our experiments. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers, such as Python versions or library versions (e.g., PyTorch, TensorFlow). |
| Experiment Setup | Yes | In Table 1 we report classification accuracies on 3 different classification tasks: 1. Cell type identification (T): midget vs. parasol vs. small bistratified cells; 2. Polarity identification (P): ON versus OFF cells; 3. Cell type and polarity (T+P): ON-midget vs. ON-parasol vs. OFF-midget vs. OFF-parasol vs. small bistratified. Each row of the table contains the data used as input. The first column represents the results for the method where the dictionary learning step is performed with k = 30, and EIs are recorded within a radius of 125 µm from the central electrode (19 electrodes on our array). Figure 3 middle panel illustrates the impact of EI diameter on classification accuracy. |