Deep Bayesian Active Learning with Image Data

Authors: Yarin Gal, Riashat Islam, Zoubin Ghahramani

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this paper we combine recent advances in Bayesian deep learning into the active learning framework in a practical way. We develop an active learning framework for high dimensional data, a task which has been extremely challenging so far, with very sparse existing literature. Taking advantage of specialised models such as Bayesian convolutional neural networks, we demonstrate our active learning techniques with image data, obtaining a significant improvement on existing active learning approaches.
Researcher Affiliation Collaboration 1University of Cambridge, UK 2The Alan Turing Institute, UK 3Uber AI Labs.
Pseudocode No The paper describes algorithms and approximations mathematically but does not include structured pseudocode or an algorithm block.
Open Source Code Yes The code for these experiments is available at http://mlg.eng.cam.ac.uk/yarin/publications. html#Gal2016Active.
Open Datasets Yes We demonstrate this on both the MNIST dataset, as well as for skin cancer diagnosis from lesion images (ISIC2016 task).
Dataset Splits Yes All models are trained on the MNIST dataset with a (random but balanced) initial training set of 20 data points, and a validation set of 100 points on which we optimise the weight decay (this is a realistic validation set size, in comparison to the standard validation set size of 5K used in similar applications such as semi-supervised learning on MNIST).
Hardware Specification No The paper does not specify the exact hardware (e.g., GPU model, CPU, memory) used for running the experiments.
Software Dependencies No The paper mentions "following the example Keras MNIST CNN implementation (fchollet, 2015)" but does not provide specific version numbers for Keras or other software dependencies.
Experiment Setup Yes All models are trained on the MNIST dataset with a (random but balanced) initial training set of 20 data points, and a validation set of 100 points... The test error of each model and each acquisition function was assessed after each acquisition... We repeated the acquisition process 100 times, each time acquiring the 10 points that maximised the acquisition function over the pool set. Each experiment was repeated three times and the results averaged... Model structure: convolution-relu-convolution-relu-max pooling-dropout-dense-relu-dropout-dense-softmax, with 32 convolution kernels, 4x4 kernel size, 2x2 pooling, dense layer with 128 units, and dropout probabilities 0.25 and 0.5. For melanoma diagnosis: We begin by creating an initial training set of 80 negative examples and 20 positive examples... We use batch size 8 and weight decay set by (1 p)l2/N, where N is the number of training points, p = 0.5 is the dropout probability, and the length-scale squared l2 is set to 0.5.