Uncertainty-Aware Hierarchical Refinement for Incremental Implicitly-Refined Classification

Authors: Jian Yang, Kai Zhu, Kecheng Zheng, Yang Cao

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on widely used benchmarks (i.e., IIRC-CIFAR, IIRC-Image Net-lite, IIRC-Image Net-Subset, and IIRC-Image Net-full) demonstrate the superiority of our proposed method over the state-of-the-art approaches.
Researcher Affiliation Collaboration Jian Yang1 Kai Zhu1, , Kecheng Zheng2 Yang Cao1,3, 1 University of Science and Technology of China 2 Ant Group 3 Institute of Artificial Intelligence, Hefei Comprehensive National Science Center
Pseudocode Yes Algorithm 1 Acquisition of Hierarchical Semantic Relationship
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] See Supplementary Materials.
Open Datasets Yes According to the semantic relevance among labels, CIFAR100 [13] and Image Net [6] datasets are rearranged to form the two-level hierarchy datasets [1].
Dataset Splits Yes The validation set also follows the incomplete information setting.
Hardware Specification No The paper states, 'Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] See Supplementary Materials.' However, the main paper itself does not explicitly describe the hardware used for experiments.
Software Dependencies No The paper mentions methods and components like 'RBF kernel' and 'BCEWith Logits Loss', but it does not specify software dependencies (e.g., libraries, frameworks) with version numbers.
Experiment Setup Yes Specifically, we use the RBF distance [19] to calculate the representation extension loss. ... γ denotes hyper-parameters for balancing the losses. Moreover, in our experiments, γ is set as 10.0. A detailed description of the hyperparameter selection is shown in supplementary material B.2. ... IIRC-CIFAR. Ten superclasses are set up, each with about 4 to 8 subclasses. In incremental phases, each new phase introduces five classes. IIRC-CIFAR involves 22 phases with ten preset class orders called phase configuration for multiple tests.