Bayesian Functional Optimization
Authors: Ngo Anh Vien, Heiko Zimmermann, Marc Toussaint
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate BFO in three typical functional optimization tasks: i) a synthetic functional optimization problem, ii) optimizing activation functions for a multi-layer perceptron neural network, and iii) a reinforcement learning task whose policies are modeled in RKHS. |
| Researcher Affiliation | Academia | Ngo Anh Vien EEECS/ECIT, Queen s University Belfast Heiko Zimmermann MLR, University of Stuttgart Marc Toussaint MLR, University of Stuttgart |
| Pseudocode | Yes | Algorithm 1 RKHS-REMBO; Algorithm 2 The BFO framework |
| Open Source Code | No | The paper does not provide an explicit statement or link to open-source code for the described methodology. |
| Open Datasets | Yes | The MNIST database consists of labeled 28x28 pixel greyscale images of handwritten digits. It contains a test data set of 10.000 data tuples and a training data set of 60.000 data tuples. |
| Dataset Splits | Yes | We selected the objective functional for Bayesian functional optimization as the cross entropy of the validation data set obtained by training the MLP model with the query activation function. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for the experiments (e.g., CPU, GPU models, memory). |
| Software Dependencies | No | The paper mentions using 'Tensor Flow with the ADAM optimizer' but does not specify version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | For all optimizers (except the functional gradient method), we use the same number N = 2 of features... The bandwidth σ1 is set equal to the bandwidth of the Gaussians... The network is trained using the cross entropy loss and stochastic minibatch gradient descent with batches of size 100, using Tensor Flow with the ADAM optimizer... We use three centers for the parametric methods, and also sparsify the functions in BFO to three basis functions... We use N = 16 centres... and set discout factor γ = 0.99 and a horizon H = 400. |