pooyan safari

Mr. pooyan safari

orcid

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
9 July 2019 in Applied Sciences
Verified
Journal Article
Open access
PDF file
UKUmair Khan
Umair Khan
PSpooyan safari
pooyan safari
Javier HernandoJavier Hernando
Javier Hernando
Restricted Boltzmann Machines (RBMs) have shown success in both the front-end and backend of speaker verification systems. In this paper, we propose applying RBMs to the front-end for the tasks of speaker clustering and speaker tracking in TV broadcast shows. RBMs are trained to transform utterances into a vector based representation. Because of the lack of data for a test speaker, we propose RBM adaptation to a global model. First, the global model—which is referred to as universal RBM—is trained with all the available background data. Then an adapted RBM model is trained with the data of each test speaker. The visible to hidden weight matrices of the adapted models are concatenated along with the bias vectors and are whitened to generate the vector representation of speakers. These vectors, referred to as RBM vectors, were shown to preserve speaker-specific information and are used in the tasks of speaker clustering and speaker tracking. The evaluation was performed on the audio recordings of Catalan TV Broadcast shows. The experimental results show that our proposed speaker clustering system gained up to 12% relative improvement, in terms of Equal Impurity (EI), over the baseline system. On the other hand, in the task of speaker tracking, our system has a relative improvement of 11% and 7% compared to the baseline system using cosine and Probabilistic Linear Discriminant Analysis (PLDA) scoring, respectively.

3 October 2024 in Journal of Optical Communications and Networking
Verified
Journal Article
Closed access
CSCaio Santos
Caio Santos
AMAbdelrahmane Moawad
Abdelrahmane Moawad
BSBehnam Shariati
Behnam Shariati
RERobert Emmerich
Robert Emmerich
PSpooyan safari
pooyan safari
JFJohannes K. Fischer
Johannes K. Fischer
Due to the scarcity of diverse and well-organized public datasets, individual research organizations are often forced to develop and utilize their own datasets. However, the utilization of machine learning (ML) models in optical communications and networks heavily depends on the existence of high-quality datasets, especially covering the various parameters to be optimized in wavelength-division multiplexing (WDM) systems. In this work, we present a public dataset for developing and testing ML models. The dataset is developed in a laboratory setting and includes 12,672 samples including data points with different modulation formats, symbol rates, distances, WDM channel allocation profiles, etc. Each data point offers more than 60 features, revealing almost every aspect of the transmission setup. Moreover, we provide optical spectra of the entire C-band as well as a constellation diagram of the channel under test for all the data points. The diversity and extensiveness of the dataset alongside a well-structured document would allow plenty of use-cases and studies to be carried out covering quality of transmission (QoT) studies, optical spectrum analysis, constellation diagram modeling, digital twin evaluation, etc. Similar to our previous efforts, the current dataset aims to facilitate collaboration by offering a way for fair comparison of research outcomes in data analysis within the domain of optical communication systems.

23 October 2025 in Journal of Optical Communications and Networking
Verified
Journal Article
Closed access
AMAngela Mitrovska
Angela Mitrovska
BSBehnam Shariati
Behnam Shariati
AJAydin Jafari
Aydin Jafari
PSpooyan safari
pooyan safari
JFJohannes Karl Fischer
Johannes Karl Fischer
RFRonald Freund
Ronald Freund
The telecommunications industry is undergoing a paradigm shift toward open, disaggregated, and automated networks, necessitating a secure, regulated, and sovereign approach to telemetry data sharing among stakeholders. This paper introduces a pioneering governance framework that leverages the Eclipse Dataspace Components Connector to enforce policy-driven telemetry data exchange within the multi-stakeholder telco ecosystem. The proposed framework offers advanced anonymization mechanisms and dynamic policy-enforcement controls, including stakeholder-specific, time-based, and location-aware access restrictions, ensuring compliance with privacy regulations. In this regard, we propose two novel, to our knowledge, data model vocabularies for modeling the telemetry data sharing problem according to the principles of the International Data Spaces Association, enabling seamless integration and valid application of data sovereignty principles. We experimentally validate the proposed framework through four use-cases designed based on real-world scenarios from operational settings, which address various stakeholder-specific data exchange scenarios, over the Fraunhofer HHI’s software-defined-networking-enabled photonics testbed. We present the policy enforcement capabilities of the framework through various experiments. Additionally, we report an in-depth performance analysis to reveal the latency and communication overhead of the proposed framework compared to conventional telemetry sharing solutions that do not comply with data sovereignty principles. The work in this paper demonstrates innovative contributions that enable data governance within optical networks, driving forward compliance, innovation, and stakeholder collaboration.

29 May 2025 in Journal of Optical Communications and Networking
Verified
Journal Article
Closed access
HAHassan Akbari
Hassan Akbari
BSBehnam Shariati
Behnam Shariati
JMJuan L. Moreno Morrone
Juan L. Moreno Morrone
PSpooyan safari
pooyan safari
JFJohannes K. Fischer
Johannes K. Fischer
RFRonald Freund
Ronald Freund
The advancement of machine learning (ML)-assisted solutions for monitoring and performance analysis in space-division multiplexing (SDM) networks has been significantly constrained by a shortage of large, well-structured, and publicly available datasets. As a result, researchers often rely on custom-built datasets, making reproducibility difficult and complicating cross-comparisons of proposed methods. To address this gap, we introduce 18 novel, to our knowledge, and publicly available quality of transmission (QoT) datasets, designed specifically for SDM networks. These datasets cover a wide range of SDM configurations, incorporating multiple fiber types, switching strategies, and two distinct network topologies. By offering consistent benchmarks, these datasets aim to support the development of ML-driven automation in SDM networks, making studies more efficient, reliable, and comparable. To demonstrate practical applications, we developed and evaluated two ML models—a classification model and a regression model—using these datasets to predict QoT outcomes. The findings illustrate the importance of the proposed datasets in benchmarking various ML-based approaches, facilitating comparison and validation across studies, and advancing ML-based network automation. This work accelerates progress toward more operationally efficient and high-performing SDM networks.

24 May 2023 in Applied Sciences
Verified
Journal Article
Open access
PDF file
PSpooyan safari
pooyan safari
MIMiquel India
Miquel India
Javier HernandoJavier Hernando
Javier Hernando
Recently, there has been a significant surge of interest in Self-Attention Networks (SANs) based on the Transformer architecture. This can be attributed to their notable ability for parallelization and their impressive performance across various Natural Language Processing applications. On the other hand, the utilization of large-scale, multi-purpose language models trained through self-supervision is progressively more prevalent, for tasks like speech recognition. In this context, the pre-trained model, which has been trained on extensive speech data, can be fine-tuned for particular downstream tasks like speaker verification. These massive models typically rely on SANs as their foundational architecture. Therefore, studying the potential capabilities and training challenges of such models is of utmost importance for the future generation of speaker verification systems. In this direction, we propose a speaker embedding extractor based on SANs to obtain a discriminative speaker representation given non-fixed length speech utterances. With the advancements suggested in this work, we could achieve up to 41% relative performance improvement in terms of EER compared to the naive SAN which was proposed in our previous work. Moreover, we empirically show the training instability in such architectures in terms of rank collapse and further investigate the potential solutions to alleviate this shortcoming.

23 January 2025 in Journal of Lightwave Technology
Verified
Journal Article
Closed access
GHGuillermo von Hünefeld
Guillermo von Hünefeld
MKMahdi Kaveh
Mahdi Kaveh
JHJoseph Hopfmüller
Joseph Hopfmüller
PSpooyan safari
pooyan safari
EŞEnes Şeker
Enes Şeker
RFRonald Freund
Ronald Freund
This paper builds upon our previous work on photonic reservoir computing, presenting an enhanced artificial neural network (NN) integrated on a photonic integrated circuit (PIC) for the intelligent monitoring of optical signals. The NN employs the reservoir computing paradigm on optical signals, with a photonic reservoir and a digital readout for modulation format identification (MFI). Our focus is on the identification of dual-polarized wavelength-division multiplexing (WDM) signals with symbol rates between 32 and 35 GBd, covering modulation formats including 4QAM, and 16QAM. Experimental results demonstrate the system's robust MFI capabilities, achieving high accuracy across various signal formats. These findings highlight the potential of integrated photonic NNs in enhancing the performance and efficiency of optical communication systems.