Axiomatic Attribution for Deep Networks
Authors: Mukund Sundararajan, Ankur Taly, Qiqi Yan
ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In Section 6, we demonstrate the ease of applicability over several deep networks, including two images networks, two text processing networks, and a chemistry network. These applications demonstrate the use of our technique in either improving our understanding of the network, performing debugging, performing rule extraction, or aiding an end user in understanding the network s prediction. |
| Researcher Affiliation | Industry | 1Google Inc., Mountain View, USA. Correspondence to: Mukund Sundararajan <mukunds@google.com>, Ankur Taly <ataly@google.com>. |
| Pseudocode | No | The paper describes the approximation of integrated gradients mathematically and explains the process in text, but it does not provide a formally structured pseudocode or algorithm block. |
| Open Source Code | Yes | More examples can be found at https://github.com/ ankurtaly/Attributions |
| Open Datasets | Yes | Image Net object recognition dataset (Russakovsky et al., 2015), Wiki Table Questions dataset (Pasupat & Liang, 2015) |
| Dataset Splits | No | The paper uses standard datasets like ImageNet and Wiki Table Questions, but it does not explicitly provide specific train/validation/test split percentages, sample counts, or detailed splitting methodologies. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory, or cloud instance types) used for running the experiments. |
| Software Dependencies | No | The paper mentions 'TensorFlow' but does not specify any version numbers for software dependencies or libraries. |
| Experiment Setup | No | The paper describes aspects of how to compute Integrated Gradients (e.g., number of steps for integral approximation), but it does not provide specific hyperparameter values, training configurations, or system-level settings for the deep learning models used in the experiments. |