Across-animal odor decoding by probabilistic manifold alignment

Authors: Pedro Herrero-Vidal, Dmitry Rinberg, Cristina Savin

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental When applied to recordings from the mouse olfactory bulb, our approach reveals low-dimensional population dynamics that are odor specific and have consistent structure across animals.
Researcher Affiliation Academia Pedro Herrero-Vidal Center for Neural Science Neuroscience Institute New York University pmh314@nyu.edu Dmitry Rinberg Neuroscience Institute Center for Neural Science NYU Langone Health rinberg@nyu.edu Cristina Savin Center for Neural Science Center for Data Science New York University cs5360@nyu.edu
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes Code available: github.com/pedroherrerovidal/am LDS.
Open Datasets Yes We tested our model on neural recordings from a 64-site grid-electrode stereotaxically implanted over the dorsal part of the olfactory bulb in five mice [13].
Dataset Splits Yes We used Bayesian model comparison to determine the dimensionality of the latent space from data (evaluated on a separate validation set).
Hardware Specification No The paper mentions 'on a 2.9GHz CPU' but does not specify a particular CPU model or other detailed hardware components for reproducibility.
Software Dependencies No The paper mentions 'scikit-learn' but does not specify version numbers for any software dependencies.
Experiment Setup Yes More specifically, we defined a shared low dimensional manifold (d = 3) and embedded K = 50 latent trajectories evolving over T = 41 time steps (Fig.2A). The stimulus-dependent inputs bk were constructed using a common template with stimulus-specific amplitude (individual dimensions scaled by a value drawn from N(1; 0.0004)) and rotation (evenly spaced over 170 degrees). The other latent dynamics parameters were set randomly: matrices Ak have diagonal entries drawn from N(0.4; 0.01) and off-diagonal drawn from N(0; 0.04) and the noise covariances Qk and Q0 are diagonal with variances drawn from N(0.55; 0.0025). We used d = 7 for all subsequent analyses.