Intern - Computationally Optimized Causal Model Learning for the Edge

Company: Siemens Mobility
Location: Princeton
Posted on: May 23, 2023

Job Description:

We are currently seeking an intern to join our Software Systems and Processes research group located in Princeton, NJ. The internship will focus on creating a framework for enhancing the performance of (transferrable) deep neural networks (DNNs) by learning the physical properties of the underlying system while intelligently using both edge and cloud resources.

Challenge:

The goal is to integrate domain knowledge with deep machine learning models by learning the underlying causal model from data first and then using this knowledge for subsequent prediction tasks. This framework can have several advantages, including efficiency, improved interpretability and reliability, and transferability and generalization.

The primary challenge is to make the training pipeline computationally efficient by enabling distributed training. The secondary challenge is to overcome model staleness, by enabling continual learning, where model is updated based on the recent changes in the input data distributions. Due to performance, privacy, regulatory considerations, there is a growing trend to perform model-update in a federated setting where the model updates locally using on-premises edge devices and then synchronizes with a global model.

The increased computational power of the edge devices (Integrated GPU boards from Nvidia Jetson family) has made model-update possible. However, co-locating a resource intensive of model training will require intelligent scheduling of resources (GPU cores) to minimize the stress on existing time-sensitive tasks. Moreover, the communication cost for performing federated learning can be significant for resource constrained edge devices especially when frequent model updates are warranted. Furthermore, communication mediums (ethernet, Wifi, 5G) and the underlying protocols (zmq, gPRC) affect the training time which can further impact accuracy and cost for performing model update.

The intern will develop algorithms to increase the performance of the existing dynamics integrated DNNs, optimize communication for federated learning, enhance resource orchestration and apply the solution in an industrial setting.

Responsibilities:

• Contribute to the task of developing a causal reinforcement learning framework.
• Setup of edge distributed learning testbed with a focus on hardware and software/networking infrastructure.
• Instrument the application to measure the cost of communication and computation across different nodes in the testbed.
• Design a communication layer-aware scheduler for optimal task placement over the network by intelligently reserving resources (GPU cores etc.) and selecting communication technologies and protocols.
• Evaluating the model's performance under more challenging conditions, such as when the training and testing scenarios are significantly different.
• Collaborate and exchange ideas with experts in the fields of distributing computing and industrial communication.
• Document your findings as a publishable scientific paper
  Skills:
  
  • A current M.S./ Ph.D. student in Computer Science or Electrical Engineering
  • Proficient programming skills in C/C++, Python, Go or any other system programming language
  • Strong experience with machine learning frameworks such as Tensorflow, FATE, PyTorch etc.
  • Research specialization in the reinforcement learning and/or causal reasoning.
  • Familiar with techniques in probabilistic programming and probabilistic graphical models.
  • Familiar with operating on graph structured data.
  • Knowledge of communication protocols such as zmq, gPRC
  • Knowledge of performance and network monitoring tools such as Collectd, Ganglia, Nagios etc.
  • Knowledge of tracing and profiling tools such as Linux Perf, OProfile
  • Knowledge of orchestration platforms such as kubernetes.
  • Experience with general purpose computing on graphics processing units (GPGPU)
  • Experience with Nvidia's parallel computation libraries
  • Interest in performance analysis and modeling.
  • Preferably, background in systems research demonstrated by peer-reviewed publications
  • Strong analytical and problem-solving skills
  • Ability to conduct research under guidance
  • Outstanding written and verbal communication skills in English with an excellent interpersonal skills and can-do attitude
  • Drive and motivation for career development and open to take responsibilities for challenging tasks
  • Team player who can also be independent, prioritize work and thrives in a fast-paced dynamic environment
    About Us
    Siemens is a global technology powerhouse that has stood for engineering excellence, innovation, quality, reliability and internationally for more than 172 years. To tap business opportunities in both new and established markets, the Company is organized in operating and strategic companies: Smart Infrastructure, Digital Industries, Siemens Mobility, Siemens Gamesa, Siemens Healthineers & Siemens Energy. (see also http://www.siemens.com/businesses/us/en/ ).
    Right now, we're powering a whole country, finding ways to prevent malaria, making cities smarter, developing renewable energy, designing 3D robots, transporting people at 250 mph and helping NASA explore Mars. Our experience has shown that key technology components coming from areas like AI, analytics, IOT, Cyber-Security, additive manufacturing and materials science are extremely relevant for such innovations.
    
    Organization: Technology
    Company: Siemens Corporation
    Experience Level: Student (Not Yet Graduated)
    Full / Part time: Full-time
    
    Equal Employment Opportunity Statement
    Siemens is an Equal Opportunity and Affirmative Action Employer encouraging diversity in the workplace. All qualified applicants will receive consideration for employment without regard to their race, color, creed, religion, national origin, citizenship status, ancestry, sex, age, physical or mental disability unrelated to ability, marital status, family responsibilities, pregnancy, genetic information, sexual orientation, gender expression, gender identity, transgender, sex stereotyping, order of protection status, protected veteran or military status, or an unfavorable discharge from military service, and other categories protected by federal, state or local law.
    
    EEO is the Law
    Applicants and employees are protected under Federal law from discrimination. To learn more, Click here .
    
    Pay Transparency Non-Discrimination Provision
    Siemens follows Executive Order 11246, including the Pay Transparency Nondiscrimination Provision. To learn more, Click here .
    
    California Privacy Notice
    California residents have the right to receive additional notices about their personal information. To learn more, click here .

Keywords: Siemens Mobility, Trenton , Intern - Computationally Optimized Causal Model Learning for the Edge, Other , Princeton, New Jersey