Towards modular and programmable architecture search

Authors: Renato Negrinho, Matthew Gormley, Geoffrey J. Gordon, Darshan Patil, Nghia Le, Daniel Ferreira

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

Reproducibility Variable Result LLM Response
Research Type Experimental 7 Experiments. We showcase the modularity and programmability of our language by running experiments that rely on decoupled of search spaces and search algorithms. Table 1: Test results for search space experiments. Table 2: Test results for search algorithm experiments.
Researcher Affiliation Collaboration Carnegie Mellon University1, TU Wien2, Microsoft Research Montreal3
Pseudocode Yes Algorithm 1: Transition. Algorithm 2: Ordered Hyperps. Algorithm 3: Random search.
Open Source Code Yes We release an implementation of our language with this paper2. 2Visit https://github.com/negrinho/deep_architect for code and documentation.
Open Datasets Yes We refer to the search spaces we consider as Nasbench [27], Nasnet [28], Flat [15], and Genetic [26].
Dataset Splits Yes The test results for the fully trained architecture with the best validation accuracy are reported in Table 1.
Hardware Specification Yes We thank Google for generous TPU and GCP grants.
Software Dependencies No The paper mentions software like TensorFlow [24], PyTorch [25], and Scikit-Learn [22] but does not provide specific version numbers for any of them.
Experiment Setup Yes For the search phase, we randomly sample 128 architectures from each search space and train them for 25 epochs with Adam with a learning rate of 0.001.