Browsing by Author "Barredo-Arrieta, Alejandro"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Plausible Counterfactuals: Auditing Deep Learning Classifiers with Realistic Adversarial Examples(Institute of Electrical and Electronics Engineers Inc., 2020-07) Barredo-Arrieta, Alejandro; Del Ser, Javier; Tecnalia Research & Innovation; IAThe last decade has witnessed the proliferation of Deep Learning models in many applications, achieving unrivaled levels of predictive performance. Unfortunately, the black-box nature of Deep Learning models has posed unanswered questions about what they learn from data. Certain application scenarios have highlighted the importance of assessing the bounds under which Deep Learning models operate, a problem addressed by using assorted approaches aimed at audiences from different domains. However, as the focus of the application is placed more on non-expert users, it results mandatory to provide the means for him/her to trust the model, just like a human gets familiar with a system or process: by understanding the hypothetical circumstances under which it fails. This is indeed the angular stone for this research work: to undertake an adversarial analysis of a Deep Learning model. The proposed framework constructs counterfactual examples by ensuring their plausibility, e.g. there is a reasonable probability that a human could generate them without resorting to a computer program. Therefore, this work must be regarded as valuable auditing exercise of the usable bounds a certain model is constrained within, thereby allowing for a much greater understanding of the capabilities and pitfalls of a model used in a real application. To this end, a Generative Adversarial Network (GAN) and multi-objective heuristics are used to furnish a plausible attack to the audited model, efficiently trading between the confusion of this model, the intensity and plausibility of the generated counterfactual. Its utility is showcased within a human face classification task, unveiling the enormous potential of the proposed framework.Item What Lies Beneath: A Note on the Explainability of Black-box Machine Learning Models for Road Traffic Forecasting(Institute of Electrical and Electronics Engineers Inc., 2019-10) Barredo-Arrieta, Alejandro; Lana, Ibai; Del Ser, Javier; Tecnalia Research & Innovation; IATraffic flow forecasting is widely regarded as an essential gear in the complex machinery underneath Intelligent Transport Systems, being a critical component of avant-garde Automated Traffic Management Systems. Research in this area has stimulated a vibrant activity, yielding a plethora of new forecasting methods contributed to the community on a yearly basis. Efforts in this domain are mainly oriented to the development of prediction models featuring with ever-growing levels of performances and/or computational efficiency. After the swerve towards Artificial Intelligence that gradually took place in the modeling sphere of traffic forecasting, predictive schemes have ever since adopted all the benefits of applied machine learning, but have also incurred some caveats. The adoption of highly complex, black-box models has subtracted comprehensibility to forecasts: even though they perform better, they are more obscure to ITS practitioners, which hinders their practicality. In this paper we propose the adoption of explainable Artificial Intelligence (xAI) tools that are currently being used in other domains, in order to extract further knowledge from black-box traffic forecasting models. In particular we showcase the utility of xAI to unveil the knowledge extracted by Random Forests and Recurrent Neural Networks when predicting real traffic. The obtained results are insightful and suggest that the traffic forecasting model should be analyzed from more points of view beyond that of prediction accuracy or any other regression score alike, due to the treatment each algorithm gives to input variables: even with the same nominal score value, some methods can take advantage of inner knowledge that others instead disregard.