MSc thesis: Memory Augmented Neural Nets for Timeseries Modelling

As the tech firm that created the mobile world, and with more than 54,000 patents to our name, we’ve made it our business to make a mark. When joining our team at Ericsson you are empowered to learn, lead and perform at your best, shaping the future of technology. This is a place where you're welcomed as your own perfectly unique self, and celebrated for the skills, talent, and perspective you bring to the team. Are you in?

Come, and be where it begins.

In the Research Area Artificial Intelligence, which is part of Ericsson Research, we are pushing the technology frontiers in AI, combining machine learning and reasoning methods, tools, and techniques to drive intelligent autonomous operations in large complex telecom systems.

We are now looking for a talented and motivated student to join us for a study on Memory Augmented Neural Nets for Timeseries Modelling.

The success of a machine learning (ML) model in modelling of temporal data largely depends how well it can learn from the past. For that the ideal ML model should meet the balance in learning from the past while not being fixated in it.

Recurrent neural nets (RNNs) have shown promising performance in modelling temporal data where short-term dependencies are of greater importance than the long-term dependencies, such as hand-writing recognition. RNNs are less effective in capturing long-term dependencies such as in question answering problems, and where a large memory is needed. Recently there has been attempts in extending RNNs by separating the computing unit from the memory unit. This family of techniques is known as memory-augmented neural nets which share conceptual similarities to the Turing machine or Von Neumann architecture. Well-known examples are the neural Turing machine (NTM) and its extension the differentiable neural computers (DNC). DNCs extend the recurrent neural networks by coupling them to an external memory resource and use attentional processes for the memory access. DNCs are fully differentiable neural nets that are learnt end-to-end using gradient descent as in RNNS.

The goals of this study are to investigate the potential application of DNC in telecom use cases. In particular, the problem of the round-trip-time (RTT) prediction could be studied. The DNC will be implemented in PyTorch and its performance will be compared against standard RNNs (e.g., LSTM) on this use case.

The work will be based on data from a 5G in-house testbed. The project will be performed in Kista and is expected to run through the Spring 2022. You will work with an experienced research team from Ericsson.

• Master’s student with most courses completed and good grades
• Good knowledge of neural networks including recurrent neural nets
• Strong programming skills in Python
• Good hands-on experience in PyTorch
• Knowledge of computer networks and cloud
• Excellent communication skills
• Excellent written and spoken English, ability to work as part of an international team

Here at Ericsson, our culture is built on over a century of courageous decisions. With us, you will no longer be dreaming of what the future holds – you will be redefining it. You won’t develop for the status quo, but will build what replaces it. Joining us is a way to move your career in any direction you want; with hundreds of career opportunities in locations all over the world, in a place where co-creation and collaboration are embedded into the walls. You will find yourself in a speak-up environment where empathy and humanness serve as cornerstones for how we work, and where work-life balance is a priority. Welcome to an inclusive, global company where your opportunity to make an impact is endless.

To prepare yourself for next steps, please explore here: https://www.ericsson.com/en/careers/job-opportunities/hiring-process

Hiring manager: In this role you will report to Master Researcher

Recruiter: Karolina Jałkowska ([email protected])

Location: Stockholm, Sweden

Kindly note that we do not accept applications sent via e-mail

Do you believe that an organization fostering an environment of cooperation and collaboration to execute with speed creates better business value? Do you value a culture of humanness, where fact based decisions are important and our people are encouraged to speak up? Do you believe that diverse, inclusive teams drive performance and innovation? At Ericsson, we do.

We provide equal employment opportunities without regard to race, color, gender, sexual orientation, transgender status, gender identity and/or expression, marital status, pregnancy, parental status, religion, political opinion, nationality, ethnic background, social origin, social status, indigenous status, disability, age, union membership or employee representation and any other characteristic protected by local law or Ericsson’s Code of Business Ethics.

Primary country and city: Sweden (SE) || || Stockholm || [[mfield2]]

Req ID: 596622