Deep Active Learning with Adaptive Acquisition

Authors: Manuel Haussmann, Fred Hamprecht, Melih Kandemir

IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our method on three benchmark vision data sets from different domains and complexities: MNIST for images of handwritten digits, Fashion MNIST for greyscale images of clothes, and CIFAR-10 for colored natural images. [...] We summarize the results in Table 1.
Researcher Affiliation Collaboration 1HCI/IWR, Heidelberg University, Germany 2Bosch Center for Artificial Intelligence, Renningen, Germany
Pseudocode Yes Algorithm 1: The RAL training procedure
Open Source Code Yes see github.com/manuelhaussmann/ral for a reference pytorch implementation of the proposed model.
Open Datasets Yes We evaluate our method on three benchmark vision data sets from different domains and complexities: MNIST for images of handwritten digits, Fashion MNIST for greyscale images of clothes, and CIFAR-10 for colored natural images.
Dataset Splits No The straight-forward reward would be the performance of the updated predictor on a separate validation set. This, however, clashes with the constraint imposed on us by the active learning scenario. ... Hence, we abandon this option altogether.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions 'pytorch implementation' in a footnote, implying the use of PyTorch, but does not provide specific version numbers for any software dependencies like PyTorch, Python, or CUDA.
Experiment Setup Yes To evaluate the performance of the proposed pipeline, we take as the predictor is a standard Le Net5 sized model (two convolutional layers of 20, 50 channels and two linear layers of 500, 10 neurons) and as the guide a policy net consisting of two layers with 500 hidden neurons. [...] The predictor is trained for 30 epochs between labeling rounds (labeling five points per round), while the policy net gets one update step after each round. [...] In each experiment the state is constructed by ranking the unlabeled data according to their predictive entropy and then taking every twentieth point until M = 50 points. [...] We stop after having collected 400 points starting from an initial set of 50 data points.