Rehearsal Learning for Avoiding Undesired Future
Authors: Tian Qin, Tian-Zuo Wang, Zhi-Hua Zhou
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments validate the effectiveness of the proposed rehearsal learning framework and the informativeness of the bound. We evaluate the proposed approach on two datasets. |
| Researcher Affiliation | Academia | Tian Qin, Tian-Zuo Wang, Zhi-Hua Zhou National Key Laboratory for Novel Software Technology Nanjing University, Nanjing, 210023, China {qint, wangtz, zhouzh}@lamda.nju.edu.cn |
| Pseudocode | Yes | Algorithm 1 Finding candidate alterations of size 1, Algorithm 3 Find candidate single alterations, Algorithm 4 Mutual information estimation. |
| Open Source Code | No | The paper does not provide an explicit statement or link indicating the availability of open-source code for the described methodology. |
| Open Datasets | Yes | Bermuda Data. We take an example from ecology, where environment variables in Bermuda are recorded [57] and the variable generation order is available [58]. Andreas Andersson and Nicholas Bates. In situ measurements used for coral and reef-scale calcification structural equation modeling including environmental and chemical measurements, and coral calcification rates in Bermuda from 2010 to 2012 (BEACON project), 2018. http://lod.bco-dmo.org/id/dataset/720788. |
| Dataset Splits | Yes | 1. Sample two sets of i.i.d. parameters, the training set Str = { Gi, θi, εi }n i=1 and the validation set Sval = { Gi, θi, εi }2n i=n+1, from P(G, θ | Dt) and P(ε | θ). ... 3. Validate alterations in C by rehearsals on Sval and remove those with ˆpξt(Sval) < τ from C. |
| Hardware Specification | No | The paper does not specify any hardware details (e.g., CPU, GPU models, memory, or cluster specifications) used for running the experiments. |
| Software Dependencies | No | The paper mentions methods like 'Bayesian ridge regression' and 'Bayesian optimization' but does not list specific software, libraries, or their version numbers. |
| Experiment Setup | Yes | For each dataset, we alter one variable in each round and repeat experiments with 100 rounds 20 times. The graph prior is initialized with samples from the learnable graph equivalence class instead of learning from data as well-established rehearsal graph learning methods are still in the process of development. The success probability is estimated with 1,000 samples from the true SRM. We compute the PAC bound with 1,000 samples from the posterior with δ = 0.05. The observational dataset size is set to 10. Table 1 shows the average probabilities of successfully avoiding the undesired future of several RL methods and the proposed rehearsal learning method with τ = 0.7. |