Online Hyperparameter Optimization for Class-Incremental Learning
Authors: Yaoyao Liu, Yingying Li, Bernt Schiele, Qianru Sun
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, we find our method performs well consistently. We conduct extensive CIL experiments by plugging our method into three top-performing methods (LU-CIR, AANets, and RMM) and testing them on three benchmarks (i.e., CIFAR-100, Image Net-Subset, and Image Net Full). |
| Researcher Affiliation | Academia | 1Max Planck Institute for Informatics, Saarland Informatics Campus 2Department of Computer Science, Johns Hopkins University 3Computing and Mathematical Sciences, California Institute of Technology 4School of Computing and Information Systems, Singapore Management University |
| Pseudocode | No | The paper describes the algorithm steps in paragraph form under "Policy Learning" but does not present a structured pseudocode or algorithm block. |
| Open Source Code | Yes | Code is provided at https://class-il.mpi-inf.mpg.de/online/ |
| Open Datasets | Yes | Datasets. We employ CIFAR-100 (Krizhevsky et al. 2009), Image Net-Subset (Rebuffi et al. 2017) (100 classes), and Image Net-Full (Russakovsky et al. 2015) (1000 classes) as the benchmarks. |
| Dataset Splits | Yes | We use the same data splits and class orders as the related work (Rebuffi et al. 2017; Liu et al. 2021a,b). ... We rebuild the local training and validation sets, and obtain the local environment hi = ((E0:i 1 Di) \ B0:i, B0:i). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or processor types) used for running experiments. |
| Software Dependencies | No | The paper mentions network architectures and baselines, but does not specify software dependencies with version numbers (e.g., PyTorch version, Python version, CUDA version). |
| Experiment Setup | Yes | Configurations. We discretize the hyperparameter search space into 50 actions, i.e., card(A)=50. We update the policy for 25 iterations in each phase, i.e., T=25. |