Publications

Abstract:

Nowadays, biometric systems are essential for iden- tification and authentication. This study proposes deep learning models for user identification using electrocardiomatrices (ECMs) as input. The proposed models are evaluated from different perspectives to assess their efficiency and performance for user identification. One aspect involves comparing convolutional and recurrent neural models. In contrast, another element consists of testing the models with different numbers of ECMs and beats per frame for generating the ECMs. The results are based on an analysis of three different electrocardiogram databases.

Abstract:

The widespread adoption of IoT devices has made the production of low-cost systems a priority. Since construction costs are generally directly related to the complexity of security methods, researchers are exploring methods that provide acceptable security with minimal hardware complexity. One such method is the use of permutation functions in ultra-lightweight authentication protocols that employ simple operators such as XOR and Shift. This paper demonstrates the critical importance of the internal structure of a permutation function in ensuring system security. This implies that even if a protocol is designed securely and efficiently, structural weaknesses in the function can render the protocol vulnerable. To illustrate this, we examine a recently published protocol named ULBRAP for supply chain management systems and reveal its security flaws, including secret disclosure and traceability attacks. We also demonstrate the attack step-by-step on Raspberry Pi devices, publishing the details on GitHub and presenting them in a video. The attack method requires 1,710,947 hash calculations, which takes approximately five minutes in our experiments. Finally, we propose a solution to address the issues associated with these functions.

Abstract:

Biometric systems are an important technique for user identification in today's world, which have been welcomed due to their non-invasive nature and high resistance to forgery and fraud. Physiological and behavioral biomarkers are two main types of biometric identifiers. Behavioral identifiers, such as voice recognition, are based on human or even animal actions. Physiological biometrics, such as fingerprints and facial recognition, which have been used in our daily lives in the past years, are based on the physical characteristics of the human body. One of the various biometrics that have been investigated in studies in this field is the heart signal, which has been well used in authentication and identification systems due to its simple acquisition process compared to biomarkers such as the brain signal. In addition, there are valid databases on heart signal data, which the researchers of this issue refer to evaluate their systems. In this study, the analysis, analysis, and comparison of different authentication methods using heart signal biometrics have been studied. Also, in the following, the advantages and disadvantages of deep learning methods and models proposed in this field have been examined. In the final part, firstly, the implementation of the method presented in Fuster and Lopez's research is discussed, and then, to evaluate, we present the tests designed using the network created in this study, and after that, concluding based on the results.

Brief Description:

Introducing a distance-bounding protocol using ECG signals and deep learning to authenticate implantable medical device (IMD) owners and thwart replay attacks.

Brief Description:

Applying diverse machine learning and deep learning models to PUF datasets to identify vulnerabilities in the underlying concepts.

Brief Description:

Developing sequence graph-based embeddings to capture dependencies in genomic and protein sequences, aiming to classify healthy and patient users and identify causative genes using machine learning and deep learning techniques.

Brief Description:

Utilizing sensor data from Karun’s Bridge in Iran during pre-construction, construction, and post-construction phases to train deep learning models, identifying the optimal time for detecting potential bridge failures.