This paper presents a novel framework for the demystification of convolutional deep learning models for time-series analysis. This is a step towards making informed/explainable decisions in the domain of time-series, powered by deep learning. There have been numerous efforts to increase the interpretability of image-centric deep neural network models, where the learned features are more intuitive to visualize. Visualization in the domain of time-series is significantly challenging as there is no direct interpretation of the filters and inputs as compared to the image modality. In addition, little or no concentration has been devoted to the development of such tools in the domain of time-series in the past. TSViz provides possibilities to explore and analyze the network from different dimensions at different levels of abstraction which includes identification of the parts of the input that were responsible for a particular prediction (including per filter saliency), importance of the different filters present in the network, notion of diversity present in the network through filter clustering, understanding of the main sources of variation learnt by the network through inverse optimization, and analysis of the network’s robustness against adversarial noise. As a sanity check for the computed influence values, we demonstrate our results on pruning of neural networks based on the computed influence information. These representations allow the user to better understand the network so that the acceptability of these deep models for time-series analysis can be enhanced. This is extremely important in domains like finance, industry 4.0, self-driving cars, health-care, counter-terrorism etc., where reasons for reaching a particular prediction are equally important as the prediction itself. We assess the proposed framework for interpretability with a set of desirable properties essential for any method in this direction.

With the advent of machine learning in applications of critical infrastructure such as healthcare and energy, privacy is a growing concern in the minds of stakeholders. It is pivotal to ensure that neither the model nor the data can be used to extract sensitive information used by attackers against individuals or to harm whole societies through the exploitation of critical infrastructure. The applicability of machine learning in these domains is mostly limited due to a lack of trust regarding the transparency and the privacy constraints. Various safety-critical use cases (mostly relying on time-series data) are currently underrepresented in privacy-related considerations. By evaluating several privacy-preserving methods regarding their applicability on time-series data, we validated the inefficacy of encryption for deep learning, the strong dataset dependence of differential privacy, and the broad applicability of federated methods.

With the rise in the employment of deep learning methods in safety-critical scenarios, interpretability is more essential than ever before. Although many different directions regarding interpretability have been explored for visual modalities, time series data has been neglected, with only a handful of methods tested due to their poor intelligibility. We approach the problem of interpretability in a novel way by proposing TSInsight, where we attach an auto-encoder to the classifier with a sparsity-inducing norm on its output and fine-tune it based on the gradients from the classifier and a reconstruction penalty. TSInsight learns to preserve features that are important for prediction by the classifier and suppresses those that are irrelevant, i.e., serves as a feature attribution method to boost the interpretability. In contrast to most other attribution frameworks, TSInsight is capable of generating both instance-based and model-based explanations. We evaluated TSInsight along with nine other commonly used attribution methods on eight different time series datasets to validate its efficacy. The evaluation results show that TSInsight naturally achieves output space contraction; therefore, it is an effective tool for the interpretability of deep time series models.

The identification and treatment of latent tuberculosis infection (LTBI) among immigrants from high-incidence regions who move to low-incidence countries is generally considered an ineffective strategy because only ≈14% of them comply with the multiple steps of the ‘cascade of care’ and complete treatment. In the Estrie region of Canada, a refugee clinic was opened in 2009. One of its goals is LTBI management. Key components of this intervention included: close collaboration with community organizations, integration within a comprehensive package of medical care for the whole family, timely delivery following arrival, shorter treatment through preferential use of rifampin, and risk-based selection of patients to be treated. Between 2009–2020, 5131 refugees were evaluated. To determine the efficacy and benefit-cost ratio of this intervention, records of refugees seen in 2010–14 (n = 1906) and 2018–19 (n = 1638) were reviewed. Cases of tuberculosis (TB) among our foreign-born population occurring before (1997–2008) and after (2009–2020) setting up the clinic were identified. All costs associated with TB or LTBI were measured. Out of 441 patients offered LTBI treatment, 374 (85%) were compliant. Adding other losses, overall compliance was 69%. To prevent one case of TB, 95.1 individuals had to be screened and 11.9 treated, at a cost of $16,056. After discounting, each case of TB averted represented $32,631, for a benefit-cost ratio of 2.03. Among nationals of the 20 countries where refugees came from, incidence of TB decreased from 68.2 (1997–2008) to 26.3 per 100,000 person-years (2009–2020). Incidence among foreign-born persons from all other countries not targeted by the intervention did not change. Among refugees settling in our region, 69% completed the LTBI cascade of care, leading to a 61% reduction in TB incidence. This intervention was cost-beneficial. Current defeatism towards LTBI management among immigrants and refugees is misguided. Compliance can be enhanced through simple measures.

Deep neural networks are one of the most successful classifiers across different domains. However, their use is limited in safety-critical areas due to their limitations concerning interpretability. The research field of explainable artificial intelligence addresses this problem. However, most interpretability methods align to the imaging modality by design. The paper introduces TimeREISE, a model agnostic attribution method that shows success in the context of time series classification. The method applies perturbations to the input and considers different attribution map characteristics such as the granularity and density of an attribution map. The approach demonstrates superior performance compared to existing methods concerning different well-established measurements. TimeREISE shows impressive results in the deletion and insertion test, Infidelity, and Sensitivity. Concerning the continuity of an explanation, it showed superior performance while preserving the correctness of the attribution map. Additional sanity checks prove the correctness of the approach and its dependency on the model parameters. TimeREISE scales well with an increasing number of channels and timesteps. TimeREISE applies to any time series classification network and does not rely on prior data knowledge. TimeREISE is suited for any usecase independent of dataset characteristics such as sequence length, channel number, and number of classes.