Visual Concept-Metaconcept Learning
Authors: Chi Han, Jiayuan Mao, Chuang Gan, Josh Tenenbaum, Jiajun Wu
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Evaluation on both synthetic and real-world datasets validates our claims. We present a systematic evaluation on both synthetic and real-world images, with a focus on learning efficiency and strong generalization. The experiment section is organized as follows. In Section 4.1 and Section 4.2, we introduce datasets we used and the baselines we compare our model with, respectively. And then we evaluate the generalization performance of various models from two perspectives. |
| Researcher Affiliation | Collaboration | Chi Han MIT CSAIL and IIIS, Tsinghua University Jiayuan Mao MIT CSAIL Chuang Gan MIT-IBM Watson AI Lab Joshua B. Tenenbaum MIT BCS, CBMM, CSAIL Jiajun Wu MIT CSAIL |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Project Page: http://vcml.csail.mit.edu. (On the first page, which links to a GitHub repository.) |
| Open Datasets | Yes | We evaluate different models on both synthetic (CLEVR [Johnson et al., 2017]) and natural image (GQA [Hudson and Manning, 2019], CUB [Wah et al., 2011]) datasets. We use a subset of 70K images from the CLEVR training split for learning visual concepts and metaconcepts. |
| Dataset Splits | Yes | 15K images from the validation split are used during test. (referring to CLEVR) and We also evaluate all trained models on a validation set which has the same data distribution as the training set. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments, only mentioning general model architectures like ResNet-34. |
| Software Dependencies | No | The paper mentions general software components like Mask R-CNN, ResNet-34, and Adam optimizer, but does not specify their version numbers or other software dependencies with versions. |
| Experiment Setup | Yes | Unless otherwise stated, the concepts and metaconcepts are learned with an Adam optimizer [Kingma and Ba, 2015] based on learning signals back-propagated from the neuro-symbolic reasoning module. We use a learning rate of 0.001 and a batch size of 10. |