Iterative Teacher-Aware Learning
Authors: Luyao Yuan, Dongruo Zhou, Junhong Shen, Jingdong Gao, Jeffrey L Chen, Quanquan Gu, Ying Nian Wu, Song-Chun Zhu
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We then validate our algorithms with extensive experiments on various tasks including regression, classification, and inverse reinforcement learning using synthetic and real data. We also show the advantage of modeling teacher-awareness when agents are learning from human teachers. |
| Researcher Affiliation | Academia | Luyao Yuan1 yuanluyao@ucla.edu Dongruo Zhou1 drzhou@cs.ucla.edu Juhong Shen2 jhshen@ucla.edu Jingdong Gao1 mxuan@ucla.edu Jeffrey L. Chen1 jlchen0@ucla.edu Quanquan Gu1 qgu@cs.ucla.edu Ying Nian Wu3 ywu@stat.ucla.edu Song-Chun Zhu1,3,4 sczhu@stat.ucla.edu 1Department of Computer Science, 2Department of Mathematics, 3Department of Statistics University of California, Los Angeles 4Beijing Institute for General Artificial Intelligence (BIGAI) |
| Pseudocode | Yes | Algorithm 1: Iterative Teacher-Aware Learning Input: Data distribution D, teacher parameter ω , learning rate ηt, teacher estimation scale βt Result: ν(T ) |
| Open Source Code | Yes | Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] Details are in Section 5, Section B and supplementary codes. |
| Open Datasets | Yes | Linear Classifiers on Natural Image Datasets: We further evaluated our teacher-aware learner on image datasets, CIFAR-10 [37] and Tiny Image Net [1] (an adaptation of Image Net [16] used in Stanford 231n with 200 classes and 500 images in each class). |
| Dataset Splits | No | The paper discusses the use of a 'test set' and mini-batch sampling, but does not specify explicit training/validation/test dataset splits with percentages, sample counts, or references to predefined splits. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments within the main text. |
| Software Dependencies | No | The paper mentions 'Scikit-learn' but does not provide specific version numbers for software dependencies, which are required for reproducible descriptions. |
| Experiment Setup | Yes | The mini-batch Dt is randomly sampled at every step with batch size 20. The learning rate is 1e-3 for all the experiments. βt is in the scale of 1e4, varying for different settings. We grid search βt starting from 1e4 and use the largest one inducing Eq. (4) that is no longer a delta function. |