On Smoother Attributions using Neural Stochastic Differential Equations
Authors: Sumit Jha, Rickard Ewetz, Alvaro Velasquez, Susmit Jha
IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | This paper uses the recently identified connection between dynamical systems and residual neural networks to show that the attributions computed over neural stochastic differential equations (SDEs) are less noisy, visually sharper, and quantitatively more robust. Using dynamical systems theory, we theoretically analyze the robustness of these attributions. We also experimentally demonstrate the efficacy of our approach in providing smoother, visually sharper and quantitatively robust attributions by computing attributions for Image Net images using Res Net-50, Wide Res Net-101 models and Res Ne Xt-101 models. |
| Researcher Affiliation | Collaboration | Sumit Jha1 , Rickard Ewetz2 , Alvaro Velasquez3 and Susmit Jha4 1University of Texas at San Antonio, San Antonio, TX 2University of Central Florida, Orlando, FL 3Air Force Research Laboratory, Rome, NY 4SRI International, Menlo Park, CA sumit.jha@utsa.edu, rickard.ewetz@ucf.edu, alvaro.velasquez.1@us.af.mil, susmit.jha@sri.com |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or link for the open-source code of the methodology described in the paper. |
| Open Datasets | Yes | Our attribution analysis and model training is performed on a system with four NVIDIA V100 32GB GPUs using the Res Net-50, Wide Res Net-101-2 and Res Ne Xt-101 models [He et al., 2016] on the Image Net benchmark. |
| Dataset Splits | Yes | We analyzed attributions for 100 images from the Image Net validation data set using both Neural ODEs and Neural SDEs for Res Net-50, Wide Res Net-101-2 and Res Ne Xt-101 models, and repeated the analysis 5 times to eliminate any significant statistical variations in our quantitative results. |
| Hardware Specification | Yes | Our attribution analysis and model training is performed on a system with four NVIDIA V100 32GB GPUs |
| Software Dependencies | No | We use the Captum tool [Kokhlikyan et al., 2020] for computing attributions and use a variety of different attribution methods including integrated gradients, Smoothgrad, Deep LIFT and Deep LIFT Shap to demonstrate the generality of our result. The Adam optimizer with a learning rate of 0.0001 and 100 epochs is used to train stochastic residual network models with a normalized injected noise of magnitudes 0.1, 0.25, 0.5 and 0.75. Attributions are computed on inferences without any noise injection using the Captum library [Kokhlikyan et al., 2020]. |
| Experiment Setup | Yes | The Adam optimizer with a learning rate of 0.0001 and 100 epochs is used to train stochastic residual network models with a normalized injected noise of magnitudes 0.1, 0.25, 0.5 and 0.75. |