Constrained Robust Submodular Partitioning

Authors: Shengjie Wang, Tianyi Zhou, Chandrashekhar Lavania, Jeff A Bilmes

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Lastly, we apply the algorithms on a real-world machine learning data partitioning problem showing good results. We empirically test Algs. 3 and. 7 on the CIFAR-10 training set [19] (|V| = 50000). We test the algorithms and compare their objective values (Eq. (4)) with a matroid constraint... In Fig. 1, we report the results for different matroid constraints with various block sizes. In Fig. 2, we use the partitioned blocks as minibatches to train a Res Net-9 model and compare their performance on the test set.
Researcher Affiliation Academia University of Washington, Seattle1; University of Maryland, College Park2
Pseudocode Yes Algorithm 1: Min-Block Streaming... Algorithm 2: Min-Block Greedy... Algorithm 3: Constrained Min-Block Greedy... Algorithm 4: Constrained Submodular Greedy Max... Algorithm 5: Cardinality Round-Robin Greedy... Algorithm 6: Round-Robin Greedy Iterations (RR(f, R, m0, M, J))... Algorithm 7: Matroid Round-Robin Greedy...
Open Source Code No The paper does not provide any statement about making its source code available or include a link to a code repository.
Open Datasets Yes We empirically test Algs. 3 and. 7 on the CIFAR-10 training set [19] (|V| = 50000).
Dataset Splits No The paper mentions using the CIFAR-10 training set and evaluating on the test set, but it does not specify a separate validation set or the percentages/counts for a train/validation/test split.
Hardware Specification No The paper does not explicitly describe the specific hardware used for running the experiments (e.g., GPU/CPU models, memory specifications).
Software Dependencies No The paper mentions training a “Res Net-9 model” and cites “Adam: A method for stochastic optimization” [15]. It also refers to a “Gaussian kernel” and “L2 distances”. However, it does not provide specific version numbers for any software libraries (e.g., PyTorch, TensorFlow, scikit-learn) or other computational tools used.
Experiment Setup No The paper describes the setup for the data partitioning experiments (e.g., number of samples, number of blocks, bandwidth of the kernel). However, it does not provide specific hyperparameters for training the Res Net-9 model, such as learning rate, batch size, optimizer settings beyond just naming ADAM, or detailed training schedules. The parameters mentioned in Figure 2 are for the matroid constraint, not the training itself.