Schema Inference for Interpretable Image Classification
Authors: Haofei Zhang, Mengqi Xue, Xiaokang Liu, Kaixuan Chen, Jie Song, Mingli Song
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Both the theoretical analysis and the experimental results on several benchmarks demonstrate that the proposed schema inference achieves encouraging performance and meanwhile yields a clear picture of the deductive process leading to the predictions. |
| Researcher Affiliation | Academia | Haofei Zhang , Mengqi Xue , Xiaokang Liu, Kaixuan Chen & Jie Song Zhejiang University {haofeizhang,mqxue,yijiyeah,chenkx,sjie}@zju.edu.cn Mingli Song Shanghai Institute for Advanced Study, Zhejiang University brooksong@zju.edu.cn |
| Pseudocode | Yes | Algorithm 1 Schema Net optimizer with initialization and sparsification. |
| Open Source Code | Yes | Our code is available at https://github.com/zhfeing/Schema Net-Py Torch. |
| Open Datasets | Yes | Datasets. We evaluate our method on CIFAR-10/100 (Krizhevsky et al., 2009), Caltech-101 (Li et al., 2022), and Image Net (Deng et al., 2009). |
| Dataset Splits | No | The paper does not explicitly provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) for training, validation, and testing across all datasets. For Caltech-101, it only specifies a training set and a test set, without a validation set. |
| Hardware Specification | Yes | We implement our method with Pytorch (Paszke et al., 2019) and train all the settings for 50 epochs with the batch size of 64 on one NVIDIA Tesla A100 GPU. |
| Software Dependencies | Yes | We implement our method with Pytorch (Paszke et al., 2019) and train all the settings for 50 epochs with the batch size of 64 on one NVIDIA Tesla A100 GPU. |
| Experiment Setup | Yes | The matcher and IR-Atlas in our method are optimized by Adam W (Loshchilov & Hutter, 2019) with a learning rate of 10-3, weight decay of 5e-4, and cosine annealing as the learning rate decay schedule. We implement our method with Pytorch (Paszke et al., 2019) and train all the settings for 50 epochs with the batch size of 64 on one NVIDIA Tesla A100 GPU. All input images are resized to 224 224 pixels before feeding to our Schema Net. We adopt Res Net-style data augmentation strategies: random-sized cropping and random horizontal flipping. |