Transductive Active Learning: Theory and Applications

Authors: Jonas Hübotter, Bhavya , Lenart Treven, Yarden As, Andreas Krause

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate their strong sample efficiency in two key applications: active fine-tuning of large neural networks and safe Bayesian optimization, where they achieve state-of-the-art performance.
Researcher Affiliation Academia Jonas Hübotter Department of Computer Science ETH Zürich, Switzerland Bhavya Sukhija Department of Computer Science ETH Zürich, Switzerland Lenart Treven Department of Computer Science ETH Zürich, Switzerland Yarden As Department of Computer Science ETH Zürich, Switzerland Andreas Krause Department of Computer Science ETH Zürich, Switzerland
Pseudocode Yes Algorithm 1 Active Fine-Tuning of NNs
Open Source Code Yes Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification:
Open Datasets Yes We use the MNIST (Le Cun et al., 1998) and CIFAR-100 (Krizhevsky et al., 2009) datasets as testbeds.
Dataset Splits No The paper mentions training until "convergence on oracle validation accuracy" to stabilize learning curves, but does not provide specific details on the validation split or how it was formed (e.g., percentages, counts, or reference to a standard split).
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, memory specifications, or types of compute workers used for running the experiments. It only vaguely mentions "small compute resources" in the NeurIPS checklist.
Software Dependencies No The paper mentions software like "ADAM optimizer", "Efficient Net-B0", and "quadcopter simulation was adapted from Chandra (2023)", but it does not specify version numbers for any of these software components, libraries, or programming languages used.
Experiment Setup Yes Table 1: Hyperparameter summary of NN experiments. (*) we train until convergence on oracle validation accuracy. MNIST CIFAR-100 ρ 0.01 1 M 30 100 m 3 10 k 1 000 1 000 batch size b 1 10 # of epochs (*) 5 learning rate 0.001 0.001