AI and especially its subdomain machine learning have been significantly impacting software and systems development in recent years. Innovation both in the computing infrastructure (such as GPUs accelerating the calculations) and innovative concepts in ML (such as large language models / LLMs) allow for addressing new problems that where formerly close to impossible to address with “classical” algorithmic approaches. Detecting pedestrians, extracting knowledge out of a body of text or even generating images based on textual descriptions are problems that are impossible to solve by designing an algorithm and implementing a program in C, C++, Java, etc.
The design and implementation of machine learning based systems requires expertise in a new technical domain as well as mastery of the related tools. Aspects such as data acquisition, data management and engineering, network design/training/validation, optimization and deployment as well as life cycle management and operations (MLOps) pose new challenges and are best introduced with an experienced partner.
At itemis, we support our customer’s projects in various aspects of machine learning:
Leverage the power of large language models, graph machine learning and natural language processing to improve your software/systems engineering process and methods, e.g. by analyzing your requirements (redundancy, content, classification), your traceability (missing links, analysis) or creating models from natural language.
Creating a ML-based product involves a lot of decisions: Which network architecture and framework to use. Which infrastructure to leverage for data management, training and validation, simulation of the environment as well as deployment of the final product.
Training a neural network on a high-end GPU is only part of the story and moving the inference from the cloud to the edge offers various benefits as it is closer to the data source resulting in reduced latency, enhanced privacy and increased robustness. If the AI is to be run on constrained devices (such as in cyberphysical systems) additional steps are required in order to optimize the network for a specific target platform (Quantization, Hardware Specific Compilation...). This requires knowledge of specific tools and frameworks that are specifically designed for this job. Embedded/Edge AI is one of the specialities of itemis.
As Machine Learning continues to pervade different sectors, its associated environmental cost continues to increase. At the dawn of responsible AI, there is a growing need to reduce the carbon footprint of AI workflows to align with sustainability goals. Reducing the energy consumption of ML workflows results in efficient yet highly accurate models and can be implemented into different areas of the workflow (data preprocessing, training, inference...). The reduction and monitoring of the energy consumption and its resulting CO2 emissions throughout the ML Lifecycle is one of the specialities of itemis that is rarely found as an offering.
As in software development, an efficient product management lifecycle with a high degree of automation is crucial for the development and maintenance of ML-based products. A plethora of tools and systems is available with a high variation of functionality and interoperability – an experienced partner helps navigating the required choices.
“Data is the new oil” is a well known adage. The exchange of both data and services operating on this data in a secure and trustworthy manner is the goal of several initiatives of dataspaces. itemis is working actively as member of a GAIA-X lighthouse project in contributing to dataspaces as well as offering services within the regulatory frameworks of these dataspaces.
From individual experts that help out in solving the most challenging problems on-the-job to well-practiced teams that serve as an extended workbench, we offer the support that best suits your needs.
Machine Learning is a rapidly developing domain with an unbelievable high rate of innovation. It is vital to monitor today’s latest developments and papers in order not to create a solution that is obsolete soon. “Transfomers” and “Large Language Models” are well known technical terms, but have you heard about Bayesian Neural Networks, Liquid Neural Networks or the Forward Forward Algorithm? itemis as a high ratio of nationally and EU funded public research projects where we work in consortia with large companies such as Bosch, Infineon and Mercedes as well as universities and small specialized companies to investigate latest technologies, develop and expand them and evaluate and prepare them for commercial deployment. We also have a high number of master theses that we supervise in cooperation with universities in order to both learn about the state-of-the-art as well as push it (and, of course, attract the best talent).