At Ericsson, you can be a game changer! Because working here isn’t just a deal. It’s a big deal. This means that you get to leverage our 140+ years of experience and the expertise of more than 95,000 diverse colleagues worldwide. As part of our team, you will help solve some of society´s most complicated challenges, enabling you to be ‘the person that did that.’ We’ve never had a greater opportunity to drive change; setting the bar for technology to be inclusive and accessible; empowering an intelligent, sustainable, and connected world.

Are you in?

Master thesis: Tuning deep neural networks through optimization of tropical polynomials

Job Description

Date: Nov 8, 2019

Do you want to do your thesis with us in Kista?

 

Would you like the chance to do your thesis together with one of the world’s leading technology companies? We are on a quest to enable communication for everyone and everything and we believe that we do that by being as innovative as possible. If you want to help us to be even more innovative – then you are on the right track!

 

We want you to become a part of a truly global company working across borders in 180 countries, offering a diverse, performance-driven culture & an innovative & engaging environment where employees enhance their potential every day. Come and learn and grow with us at Ericsson.

 

Background

Machine learning methods are today used in essentially all corners of society, through non-critical systems such as dialog agents in shopping centers to highly-critical systems such as medical diagnosis and intrusion detection. Despite it every increasing number of applications it is still very much an “art” to train and deploy such methods.

Recent works have shown that several deep neural networks architectures can be represented as Jordan decompositions of semiring polynomials in tropical algebra. This together with results on semi-definite performance estimation problems, that show that the worst case error of common training algorithms admit a closed-form under mild assumptions, suggest that it can be feasible to find optimal parameter values (e.g. learning rate) of the training algorithms in several common cases.

 

Project goal

Based on the connection between convex optimization and machine learning this thesis will,

Characterize the optimal parameters values under the hypothesis that the dataset is generated from a fragment of the exponential family of distributions.

Demonstrate its applicability by constructing and evaluating a training framework in Matlab or Python on synthesized datasets.

 

Thesis work description/Qualifications

This thesis work is suitable for one to two students with background in mathematics/theoretical computer science (optimization, abstract algebra, machine learning, automata theory).

The students will study relevant literature, conduct theoretical work and develop training framework.

Extent: 30 hp

Location: The thesis will be conducted in Ericsson’s office in Kista, Sweden.

 

If you feel interested, please send in your application as soon as possible. The start date can be adjusted to both your and the business needs – the intention is to start in January.

 

While applying please attach your updated CV, current grades and cover letter written in English into one document (under CV field in the application tool)  and clearly define your technical knowledge.

 

The above presented thesis is not written in stone. Have an idea on how we can make it even better? We would love to hear them.

 

If you have any questions, please reach out to Sylwia Kwiecień at sylwia.kwiecien@ericsson.com

 

To be a driver of global transformation is in our DNA and we use innovation to empower people, business and society. Through our technological leadership, global presence and the empowerment we provide to our employees, we offer enormous opportunities to change people's lives and the world we live in.

 

We welcome the opportunity to meet you!

 

Ericsson provides equal employment opportunities (EEO) to all employees and applicants for employment without regard to race, color, religion, sex, sexual orientation, marital status, pregnancy, parental status, national origin, ethnic background, age, disability, political opinion, social status, veteran status, union membership or genetics.

Ericsson complies with applicable country, state and all local laws governing nondiscrimination in employment in every location across the world in which the company has facilities. In addition, Ericsson supports the UN Guiding Principles for Business and Human Rights and the United Nations Global Compact.

This policy applies to all terms and conditions of employment, including recruiting, hiring, placement, promotion, termination, layoff, recall, transfer, leaves of absence, compensation, training and development.

Ericsson expressly prohibits any form of workplace harassment based on race, color, religion, sex, sexual orientation, marital status, pregnancy, parental status, national origin, ethnic background, age, disability, political opinion, social status, veteran status, union membership or genetic information.

 

Primary country and city: Sweden (SE) || || Stockholm || Stud&YP

Req ID: 305461