PhD AI based Domain Adaptation for telecom data in RAN

Job title: PhD AI based Domain Adaptation for telecom data in RAN

Company: Ericsson

Job description: Job Description:About this opportunityWhile 6G is still in the first stages of research and development, it is expected to bring several advancements over 5G technology: better throughput, lower latency, automation of the network management. In recent years, Artificial Intelligence (AI) has emerged as a pivotal catalyst for innovation within the telecommunications sector as an integral part of making the networks smarter and meet the new challenges to deliver the expected performance that far exceed networks of today.Today, the efficient deployment of AI in this dynamic sector is intrinsically tied to the ability to exploit and interpret huge time series data generated by a diverse range of network components and devices, including a wide array of devices, IoT sensors, and network elements. Time Series Classification is the task of classifying the correct label of a time series given or not a set of labeled example time series. In order to create and extract value from this data in real time, we use many different ML/AI techniques and technologies that maximize the output for each function in our highly distributed and decentralized networks making them more resilient, adaptive and optimized than has ever been possible before. However, the most common setup of these ML/AI technologies is to be trained and set for a fixed scenario, so it needs adaptation or retraining in case of any change. Thus, we are interested in adaptation techniques that can lead to a better and faster generalization of the models.While some research has delved into unsupervised domain adaptation for time series, existing algorithms often exhibit limited generalization across diverse use-cases. Furthermore, a considerable research gap persists in identifying optimal source-target domain pairs based on time series data similarity and in developing robust approaches for multi-source to multi-target adaptation scenarios. One direction toward the best selection of the pairs is to estimate the shift between the source and the target domains. This direction holds the potential to offer valuable insights into the following open questions: i) when does domain adaptation prove beneficial; ii) which algorithm is most effective depending on the shift level; and iii) which source(s) and target(s) to choose in the case of the multi-source multi-target adaptation. Another important challenge in Telecommunications is that the shift between training data and deployment data can also arise in the label distribution (different probabilities for each label, or new labels). It occurs when there is a significant change in the distribution of labeled outcomes, often associated with network performance metrics or user behavior, between different time periods or geographical regions.Join our TeamYou’ll get the opportunity toYou will embark on a research journey starting with an in-depth exploration of the literature, conducting a comprehensive review of transfer learning and domain adaptation techniques. Your initial focus will be on studying the generalization of existing unsupervised domain adaptation solutions for time series data. The primary emphasis will be on developing metrics and methodologies to quantify domain shifts and explain the drop in accuracy when adapting models from source to target domains. This analysis will help identify the challenges and limitations of current techniques.Following this, you will delve into advanced techniques tailored to complex scenarios such as:Selecting the best source-target scenario in cases of multiple adaptations, including multi-source and multi-target adaptation.
Managing label shifts between source and target domains.
Addressing non-homogeneous adaptation challenges.
You will propose, select, and rigorously evaluate predictive models specifically designed for unsupervised domain adaptation in time series. The ultimate goal is to develop innovative solutions that effectively address the challenges associated with adapting to constantly evolving data environments. Practical experiments will be conducted on various datasets, providing a detailed assessment of the technical performance.Key skills

• Hold a master’s degree or equivalent in computer science, applied mathematics, electrical engineering, or related disciplines.
• Good understanding of ML/AI techniques/concepts such as SVM, Bayesian models, neural networks, time series analysis, random forests, gradient boosting, and hyper-parameter optimization techniques.
• Familiar with ML packages and frameworks such as PyTorch, Scikit-learn, and NumPy.
• Possess strong practical programming skills and are comfortable reading theoretical research articles.
• Are proficient in Python programming and experienced in using scientific Python packages.
• Have good communication skills in written and spoken English.
• Demonstrate creativity and the ability to independently formulate and solve problems.
• (Optional but recommended) Have experience with stack technologies such as Kubernetes, Docker, and Git.

What happens once you apply?to find all you need to know about what our typical hiring process looks like. This may vary depending on the location and role.Encouraging a diverse and inclusive organization is core to our values at Ericsson, that’s why we nurture it in everything we do. We truly believe that by collaborating with people with different experiences we drive innovation, which is essential for our future growth. We encourage people from all backgrounds to apply and realize their full potential as part of our Ericsson team.Ericsson is proud to be an Equal Opportunity and Affirmative Action employer,

Expected salary:

Location: Massy, Essonne – Palaiseau, Essonne

Job date: Fri, 23 Aug 2024 22:11:17 GMT

Apply for the job now!

Submit your Resume!