Deep Learning for Multi-Facility Location Mechanism Design
Authors: Noah Golowich, Harikrishna Narasimhan, David C. Parkes
IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In our experiments (Section 6), Regret Net-nm learns mechanisms that are essentially strategy-proof, with social costs comparable to or lower than the best percentile and dictatorial rules. |
| Researcher Affiliation | Academia | 1Harvard University, SEAS, 33 Oxford Street, Cambridge, MA, 02138 ngolowich@college.harvard.edu, hnarasimhan@g.harvard.edu, parkes@eecs.harvard.edu |
| Pseudocode | No | The paper includes figures illustrating network architectures (Figure 1, Figure 2, Figure 3) and mathematical formulations, but no explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | The paper does not provide any statements about releasing code, nor does it include links to a source code repository. |
| Open Datasets | No | The paper describes generating synthetic data for experiments ('We consider n = 5 agents, whose peaks are distributed i.i.d. uniformly on [0, 1].'), but it does not provide access information (e.g., link, DOI, or citation to an established public dataset) for a publicly available or open dataset. |
| Dataset Splits | No | The paper mentions a 'held-out test set of 2000 examples' and states that network selection is 'based only on the training data'. It does not explicitly specify a separate validation dataset split or its size for hyperparameter tuning or early stopping, beyond what might be implicitly part of the training process. |
| Hardware Specification | Yes | Both networks take less than half hour to train on a Tesla K20Xm GPU. |
| Software Dependencies | No | The paper mentions using the 'Tensorflow library' and the 'Adam algorithm' for optimization. However, it does not provide specific version numbers for TensorFlow or any other key software components, which is required for reproducibility. |
| Experiment Setup | Yes | We implement Regret Net-nm with L = 4 hidden layers, each with 40 units. In the augmented Lagrangian solver, we performed 1000 updates on λ and 50 gradient updates on w for every one update on λ. We implement Moulin Net with L = K = 3. We use the Adam algorithm for optimizing w (with mini-batches of size 500). The learning rate in Adam was initialized to 0.005 for Regret Net-nm (and decayed by a factor 0.99 every 100 updates) and to 0.1 in Moulin Net. Weights were initialized as i.i.d. draws from N(0, 0.01). The offsets in Regret Net-nm were initialized to 0.1. |