PhD - Embedding Methods for Time Series Modeling

  • Robert-Bosch-Campus 1, 71272 Renningen, Germany
  • Full-time
  • Legal Entity: Robert Bosch GmbH

Company Description

Do you want beneficial technologies being shaped by your ideas? Whether in the areas of mobility solutions, consumer goods, industrial technology or energy and building technology – with us, you will have the chance to improve quality of life all across the globe. Welcome to Bosch.

The Robert Bosch GmbH is looking forward to your application!

Job Description

The goal of this PhD project is to develop embedding methods that project time series of varying length into an embedding space. In the embedding space, each time series is associated with a fixed-length vector representation. Its dimensionality is fixed and often much lower than the dimensionality of the original data. A good embedding method will map two similar time series to two vectors that are close to each other in the embedding space. Having such an embedding is useful because the vector representations can be used as input to other machine learning methods. These methods can then exploit the geometry of the embedding space to share statistical information between the embeddings and thereby better generalize between the time series. Key to the success of an embedding method is that the distances in the embedding space do indeed reflect the relevant similarities between the time series. There are two main information sources that should influence what the relevant similarities are.

First, the systematic differences between the time series are reflected in the data and a good embedding method can leverage black-box approaches such as neural sequence models to extract them. On the other hand, it is important to inform the design of the embedding method through domain expertise. For this reason, we propose to develop a hybrid approach that explicitly structures part of the embedding model according to underlying physical principles and prior knowledge while the remaining part of the model enjoys the full flexibility of neural approaches. The development of useful embedding methods requires careful evaluation of competing approaches. Especially a hybrid embedding method has many design choices that need to be justified in terms of usefulness of the resulting embedding. Therefore an important part of this PhD project will also consist of designing a variety of tasks that measure the quality of the time series embeddings and, in turn, inform their design and training.

During your PhD you will be part of the active research team at the Bosch Center for Artificial Intelligence (BCAI, www.bosch-ai.com).

Your tasks will include:

  • Invention of novel time series embedding methods.
  • Prototype implementations and benchmarking on real-world data sets.
  • Literature screening and building close contact with the academic community.
  • Active participation in the academic activities of the BCAI research team.
  • Publications in top-tier conferences (ICML, NIPS, ICLR, AISTATS, etc.) and journals (JMLR, PAMI, etc.).

Qualifications

  • Personality: highly motivated, excellent communications and theoretical skills
  • Working Practice: collaborative mindset and independent problem solving
  • Experience and Knowledge: programming experience in python, preferably with tensorflow or another machine learning library, prior experience implementing a machine learning algorithm and running it on real data is a plus
  • Languages: excellent in English (written and spoken)
  • Education: excellent degree (Master) in machine learning, mathematics, computer science, statistics, physics, or a related field

Additional Information

Please submit all relevant documents (incl. curriculum vitae, certificates) and a cover letter emphasizing why you want to do a PhD in machine learning and what you are hoping to learn during your PhD at the Bosch Center for Artificial Intelligence.

The final PhD topic is subject to your university. Duration: 3 years
                                                                                    
Need support during your application?
Kevin Heiner (Human Resources)
+49 711 811 12223

Need further information about the job?
Maja Rita Rudolph (Functional Department)
+49 711 811 91510

Videos To Watch

Privacy Policy