Bayesian target optimisation for high-precision holographic optogenetics

Authors: Marcus Triplett, Marta Gajowa, Hillel Adesnik, Liam Paninski

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our approach in simulations and using data from in vitro experiments, showing that Bayesian target optimisation considerably reduces OTS across all conditions we test.
Researcher Affiliation Academia 1Department of Statistics, Columbia University 2Zuckerman Mind Brain Behavior Institute, Columbia University 3Department of Molecular and Cell Biology, UC Berkeley
Pseudocode Yes Algorithm 1: Bayesian target optimisation (Bataro).; Algorithm 2: Projected Monte Carlo gradient descent algorithm for optimising single-target holograms
Open Source Code Yes An open-source implementation of Bayesian target optimisation is available in Python at https: //github.com/marcustriplett/bataro.
Open Datasets No The paper mentions using 'detailed cell-attached recordings in slice [42]' to create synthetic optogenetics experiments, but reference [42] is another paper (bioRxiv), not a direct link or citation to a publicly available dataset.
Dataset Splits Yes We selected the parameters of the GP covariance kernel using 5-fold cross-validation on a separate set of recordings that were made on the same set of four cells, ensuring the hyperparameter selection was using out-of-sample data... For each hyperparameter combination θ and for each cell, we used Newton s method to fit the GP-Bernoulli model to 80% of the trials... On the remaining 20% of the trials (denoted as Theld-out), we evaluated the log-likelihood
Hardware Specification Yes All computational procedures were performed either on a desktop workstation running Ubuntu 18.04 with an Intel Xeon E5-2620 v4 CPU, four GTX 1080 Ti GPUs, and 112GB RAM, or on the Axon computer cluster based at the Zuckerman Institute (Columbia University) using nodes comprised of two Xeon E5-2660 v4 CPUs, eight GTX 1080 Ti GPUs, and 125GB RAM.
Software Dependencies No The paper mentions 'Ubuntu 18.04' which is an operating system, and 'Python' for implementation, but does not provide specific version numbers for other key software components, libraries, or frameworks used (e.g., PyTorch, NumPy).
Experiment Setup Yes Table S1: Parameters used for simulations and generating synthetic optogenetics experiments. This table provides specific hyperparameters such as 'Kernel radial lengthscale', 'Kernel power lengthscale', 'Kernel amplitude', and 'Learning rate for spike thresholds'.