FLEX: Faithful Linguistic Explanations for Neural Net Based Model Decisions
Authors: Sandareka Wickramanayake, Wynne Hsu, Mong Li Lee2539-2546
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiment results on two benchmark datasets demonstrate that the proposed framework can generate discriminative and faithful explanations compared to state-of-the-art explanation generators. We also show how FLEX can generate explanations for images of unseen classes as well as automatically annotate objects in images. We evaluate the effectiveness of the proposed framework on CUB (Wah et al. 2011) and MPII (Andriluka et al. 2014) datasets. |
| Researcher Affiliation | Academia | Sandareka Wickramanayake, Wynne Hsu, Mong Li Lee School of Computing, National University of Singapore {sandaw, whsu, leeml}@comp.nus.edu.sg |
| Pseudocode | No | The paper describes methods through text and equations but does not present a formal pseudocode or algorithm block. |
| Open Source Code | No | Not found. The paper mentions using 'publicly available GVE and MME codes' which refers to third-party code, not the code for FLEX itself. |
| Open Datasets | Yes | CUB. This is the Caltech UCSD Birds dataset containing 11,788 images of birds belonging to 200 classes (Wah et al. 2011). MPII. This dataset has 25k images of human poses for different activities (Andriluka et al. 2014). |
| Dataset Splits | No | The paper uses 'CUB' and 'MPII' datasets and mentions 'training dataset' and 'test sets' (e.g., 'CUB and MPII test sets') but does not specify the explicit percentages or sample counts for training, validation, and testing splits, nor does it cite specific pre-defined splits with sufficient detail for reproducibility. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory, or processing power) used for running the experiments. |
| Software Dependencies | No | The paper mentions software components like 'LSTM' and 'Res Net-50' as part of the model architecture, but does not list specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x) required for replication. |
| Experiment Setup | Yes | In our experiments, we set η to 80%. In FLEX, visual features are embedded to 512-dimensional space whereas words are embedded to 1000-dimensional space. LSTM hidden state dimension is 1000. We use λ = 0.1 for CUB, and λ = 0.001 for MPII for subsequent experiments as these values give us the best DREL results. |