MARS: Meta-learning as Score Matching in the Function Space

Authors: Krunoslav Lehman Pavasovic, Jonas Rothfuss, Andreas Krause

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

Reproducibility Variable Result LLM Response
Research Type Experimental In our experiments, we demonstrate that our proposed approach, called Meta-learning via Attention-based Regularised Score estimation (MARS), consistently outperforms previous meta-learners in predictive accuracy and yields significant improvements in the quality of uncertainty estimates. 6 EXPERIMENTS
Researcher Affiliation Academia Krunoslav Lehman Pavasovic ETH Zurich Switzerland klehman@ethz.ch Jonas Rothfuss ETH Zurich Switzerland rojonas@ethz.ch Andreas Krause ETH Zurich Switzerland krausea@ethz.ch
Pseudocode Yes Algorithm 1 MARS: Meta-Learning the Data-Generating Process Score
Open Source Code Yes The code scripts for reproducing the experimental results are provided in our repository2. 2https://github.com/krunolp/mars
Open Datasets Yes Swiss FEL (Milne et al., 2017), Physio Net (Silva et al., 2012), Intel Berkeley Lab temperature sensor dataset (Berkeley-Sensor) (Madden, 2004)
Dataset Splits Yes From the remaining patients, we used 100 patients for meta-training and 500 patients for meta-validation and meta-testing.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions software like Jax (Bradbury et al., 2018), Haiku (Hennigan et al., 2020), GP-Jax (Pinder & Dodd, 2022), scikit-learn (Pedregosa et al., 2011), and TensorFlow Distributions (Dillon et al., 2017), but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes We fix the number of particles to L = 10 and perform 10000 f SVGD update steps.