Online Learning with Transductive Regret
Authors: Mehryar Mohri, Scott Yang
NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we present some toy experiments illustrating the effectiveness of the Reduced Power Method for approximating the stationary distribution in FASTSWAP. We considered n base learners, where n 2 {40, 80, 120, 160, 200}, each using the weighted-majority algorithm [Littlestone and Warmuth, 1994]. We generated losses as i.i.d. normal random variables with means in (0.1, 0.9) (chosen randomly) and standard deviation equal to 0.1. We capped the losses above and below to remain in [0, 1]. We ran FASTSWAP for 10,000 rounds in each simulation and repeated each simulation 16 times. |
| Researcher Affiliation | Collaboration | Mehryar Mohri Courant Institute and Google Research New York, NY mohri@cims.nyu.edu Scott Yang D. E. Shaw & Co. New York, NY yangs@cims.nyu.edu |
| Pseudocode | Yes | Algorithm 1: FASTSWAP; {Ai}N i=1 are external regret minimization algorithms. Algorithm: FASTSWAP((Ai)N i=1); Algorithm 2: FASTTIMESELECTTRANSDUCE; AI, (AI,u,i) external regret algorithms. |
| Open Source Code | No | The paper does not provide any explicit statements about releasing source code or links to a code repository. |
| Open Datasets | No | The paper describes generating synthetic data for its toy experiments ("We generated losses as i.i.d. normal random variables"), but it does not specify the use of a publicly available dataset or provide access information for the generated data. |
| Dataset Splits | No | The paper describes running simulations for a fixed number of rounds (10,000 rounds) but does not mention dataset splits for training, validation, or testing in the traditional sense, as it generates i.i.d. random variables for losses. Thus, it does not specify exact split percentages or counts. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU/CPU models or memory used for running the experiments. |
| Software Dependencies | No | The paper mentions using "the weighted-majority algorithm [Littlestone and Warmuth, 1994]" as base learners but does not specify any software names with version numbers for implementation or analysis. |
| Experiment Setup | No | The paper describes the setup of the toy experiments (e.g., number of base learners, loss generation method, number of rounds), but it does not provide specific hyperparameter values for the FASTSWAP or other algorithms, nor does it detail system-level training settings. |