Efficient AI

Energy efficient AI for the future - on-device inference and training

Whether for automating processes or analyzing large volumes of data: Intelligent and self-learning systems are becoming increasingly important in business processes. Until now, these intelligent systems have always had to be connected to a cloud, as this provides the necessary computing power for AI models. With Edge AI, short for Edge Artificial Intelligence, the next generation of intelligent systems is now entering the home: intelligence is being transferred directly to the end devices.

Benefits of Edge AI

Energy Efficiency and Resource Conservation

Edge AI requires up to a thousand times less power for ML applications than a standard GPU, as models are executed locally instead of sending data back and forth. This means that devices with Edge AI can be operated for years without batteries, depending on the application. The energy savings also reduce the size of the batteries required and the consumption of valuable resources.

Realtime

 

Since the model is run locally, in order to perform inference, the raw data does not need to be sent to the cloud first and then sent back when processed. This reduces output latency as well as communication bandwidth requirements, which in turn enables rapid response.

Independence, Privacy and Security

The user is independent of cloud service providers, which means that data does not have to be shared externally and privacy remains protected. In addition, there is no dependency on a communication connection, making it easy to retrofit AI.

Our Service Offer

Research and Development

We offer you partial or complete R&D services.

 

  • Embedded AI: development tool for developing embedded AI solutions to reduce costs and improve the quality of your application.
  • Optimized AI model for your hardware: a tailor-made solution adapted to your hardware with optimized performance through AI.
  • Mentoring : We accompany you in your R&D projects from data acquisition to the development of an AI model.

Consulting

We advise and support you with your individual concerns relating to your AI solutions.

 

 

  • Potential analyses: We carry out a quick potential analysis of your personal concerns.
  • Directional decisions: We provide you with groundbreaking support by creating an automated report to help you make initial directional decisions on your personal project.
  • Hardware recommendations: We advise you on suitable hardware and give you recommendations for use in your AI solution.
  • Personal support: Our interdisciplinary team and the network at IIS will support you with your personal project.

Licensing

We may have already developed the right AI model and this can be integrated directly into your use case or licensed by you.

 

We optimize your application through standardized processes and extensive automation of time-intensive work specifically adapted to your use case.

Through training and automatic reduction of complex AI models by removing redundancies, we generate optimal AI models in terms of accuracy and efficiency.

We support you with rapid integration!

Contact us!

We realize the efficient processing of your R&D projects, as well as the training of junior staff with this new competence profile.

Contact us
for an individual consultation at

machine-learning-lv@iis.fraunhofer.de

Become a Certified Data Scientist Specialized in Edge AI

TinyML enables machine learning on microcontrollers and opens doors to a wide range of application areas. The course is designed for specialists with data analysis experience and hardware developers who want to improve their knowledge of machine learning.

You will learn how to develop ML applications on microcontrollers, recognize challenges at an early stage and develop appropriate strategies. Use our comprehensive know-how for your success.

ML Seminar - How to implement AI projects successfully

In the ML seminar, you will learn the basics of machine learning using supervised and unsupervised learning methods. In addition, you will learn more about proven methods for fast results and decisions as well as examples of the use of machine learning in practical implementation.

Use Cases

 

Vision

From agriculture and biodiversity to people counting. AI analysis directly behind the camera sensor enables numerous new applications without sending a lot of raw data to the cloud. Privacy remains protected at all times.

 

Condition Monitoring

Using machine learning in embedded systems to monitor the status of systems and machines in order to be able to react at an early stage or increase efficiency.

Retail

We've all been there: long queues at the checkout again.

With privacy-protecting AI on edge devices, seamless shopping is already possible.

This leaves more time for the finer things.

 

Speech & Audio

Machine learning in embedded sensor modules for cognitive speech and audio analysis. The audio commands are recognized without a cloud connection.

 

Tools

Machine learning in embedded sensor modules for cognitive hand tools to recognize assembly processes and ensure quality.

Publications

2024

Mutschler, C., Münzenmayer, C., Uhlmann, N., & Martin, A (2024): 

Unlocking Artificial Intelligence: From Theory to Applications

In: Springer

 

Witt, N., Deutel, M., Schubert, J., Sobel, C., & Woller, P. (2024)

Energy-Efficient AI on the Edge

In: Unlocking Artificial Intelligence: From Theory to Applications (pp. 359-380)

 

Deutel, M., Hannig, F., Mutschler, C., & Teich, J (2024):

Fused-Layer CNNs for Memory-Efficient Inference on Microcontrollers

In: Workshop on Machine Learning and Compression, NeurIPS 2024

 

Deutel, M., Hannig, F., Mutschler, C., & Teich, J. (2024):

On-Device Training of Fully Quantized Deep Neural Networks on Cortex-M Microcontrollers

In: IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 2024

 

Deutel, M., Mutschler, C., & Teich, J. (2024):

microYOLO: Towards Single-Shot Object Detection on Microcontrollers

In: arXiv preprint arXiv:2408.15865.

 

Herzog, B., Schubert, J., Rheinfels, T., Nickel, C., & Hönig, T. (2024):

GreenPipe: Energy-Efficient Data-Processing Pipelines for Resource-Constrained Systems.

In: Proceedings of the 21st International Conference on Embedded Wireless Systems and Networks (EWSN'24), ACM

2021

Blauberger, P., Marzilger, R., & Lames, M. (2021):

Validation of player and ball tracking with a local positioning system

In: Sensors, 21(4), 1465

 

Potortì, F., Torres-Sospedra, J., Quezada-Gaibor, D., Jiménez, A. R., Seco, F., Pérez-Navarro, A., ... & Oh, H. L. (2021):

Off-line evaluation of indoor positioning systems in different scenarios: The experiences from IPIN 2020 competition

In: IEEE Sensors Journal, 22(6), 5011-5054

 

Löffler, C., Nickel, C., Sobel, C., Dzibela, D., Braat, J., Gruhler, B., ... & Mutschler, C. (2021):

Automated quality assurance for hand-held tools via embedded classification and AutoML

In: Machine Learning and Knowledge Discovery in Databases. Applied Data Science and Demo Track: European Conference, ECML PKDD 2020, Ghent, Belgium, September 14–18, 2020, Proceedings, Part V (pp. 532-535)

2020

Feigl, T., Kram, S., Woller, P., Siddiqui, R. H., Philippsen, M., & Mutschler, C. (2020):

RNN-aided human velocity estimation from a single IMU

In: Sensors, 20(13), 3656

 

Feigl, T., Gruner, L., Mutschler, C., & Roth, D. (2020):

Real-time gait reconstruction for virtual reality using a single sensor

In: 2020 IEEE International Symposium on Mixed and Augmented Reality Adjunct (ISMAR-Adjunct) (pp. 84-89)

 

Redžepagić, A., Löffler, C., Feigl, T., & Mutschler, C. (2020):

A sense of quality for augmented reality assisted process guidance

In: 2020 IEEE International Symposium on Mixed and Augmented Reality Adjunct (ISMAR-Adjunct) (pp. 129-134)

2019

Niitsoo, A., Edelhäußer, T., Eberlein, E., Hadaschik, N., & Mutschler, C. (2019):

A deep learning approach to position estimation from channel impulse responses

In: Sensors, 19(5), 1064

 

Feigl, T., Kram, S., Woller, P., Siddiqui, R. H., Philippsen, M., & Mutschler, C. (2019):

A bidirectional LSTM for estimating dynamic human velocities from a single IMU

In: Computer Vision Foundation (CVF) (Eds.), Joint Workshop on Long-Term Visual Localization, Visual Odometry and Geometric and Learning-based SLAM (pp. 42-43)

2018

Feigl, T., Mutschler, C., & Philippsen, M. (2018):

Supervised learning for yaw orientation estimation

In: 2018 international conference on indoor positioning and indoor navigation (IPIN) (pp. 206-212)

 

Feigl, T., Mutschler, C., & Philippsen, M. (2018):

Human compensation strategies for orientation drifts

In: 2018 IEEE Conference on Virtual Reality and 3D User Interfaces (VR) (pp. 409-414)

 

Feigl, T., Mutschler, C., & Philippsen, M. (2018):

Head-to-body-pose classification in no-pose VR tracking systems

In: 2018 IEEE Conference on Virtual Reality and 3D User Interfaces (VR) (pp. 1-2)