Towards Backdoor Attack on Deep Learning based Time Series Classification

Explainable Neural Dynamics Models for Motor Temperature Prediction

“…The reported average fooling rate is meticulously computed by aggregating outcomes from comprehensive experiments, encompassing all datasets explored and the application of algorithms across diverse DL models. The findings indicate that TimeTrojanDE [13] achieved a notable average fooling rate of 92.5%. TSBA [12] displayed a higher efficacy with a fooling rate of 99.07%, and TrojanFlow [11] was similarly effective, achieving a fooling rate of 99.65%.…”

“…Even state-of-the-art computational configurations, bolstered by multiple GPU resources, experienced memory overflows when employing this approach for trigger generation on large and complex datasets, according to our rigorous empirical investigation. In [13], the authors proposed the TimeTrojan-DE algorithm, which also employs a generative approach to produce the trojan trigger for a white-box attack and incorporates an evolutionary algorithm aiming to decrease the number of iterations required to generate optimized triggers. Nevertheless, the authors highlighted that as the search space expands, the efficiency of the algorithm diminishes.…”

“…However, there is no existing study on trojan attacks specifically targeting DL models working with time series data (TSD) in the SG. Despite the sparse research on trojan attacks in TSD unrelated to the complex event spectrum of SG, our comprehensive empirical analysis reveals that the methodologies proposed in these studies [11], [12] strain computational resources and cause memory overflows when applied to datasets with large numbers of samples and time steps as well as become less efficient as the search space widens [13]. Therefore, this paper aims to address this gap by proposing a novel approach for generating highly effective trojan attacks on DL models that use TSD in the SG.…”

Deep Learning Based Model for Predicting the Contribution of SMEs to the Economy

“…LSTM, zaman serileri, ses ve metin gibi sıralı verileri işlemek için özel olarak tasarlanmış tekrarlı sinir ağı modelidir [42]. LSTM, sıralı verilerdeki uzun vadeli bağımlılıkları öğrenme yeteneğine sahiptir.…”

S3: Sneaky Spectral Strike Trojan Attacks on Deep Learning-based Time Series Smart Grid Models

“…However, there is no existing study on trojan attacks specifically targeting DL models working with time series data (TSD) in the SG. Despite the sparse research on trojan attacks in TSD unrelated to the complex event spectrum of SG, our comprehensive empirical analysis reveals that the methodologies proposed in these studies [11], [12] strain computational resources and cause memory overflows when applied to datasets with large numbers of samples and time steps as well as become less efficient as the search space widens [13]. Therefore, this paper aims to address this gap by proposing a novel approach for generating highly effective trojan attacks on DL models that use TSD in the SG.…”