Research

Role of Machine Learning in Embedded Systems

By: Bhawna Garg1, Dr. Gurinder Kaur Sodhi2

1 Department of Electronics and Communication, Desh Bhagat University (Punjab), PHD(Scholar)

2 Department of Electronics and Communication, Desh Bhagat University (Punjab), Assistant Professor

Abstract

The field of embedded systems continues to develop rapidly as a result of the widespread usage of ML techniques. Machine learning allows systems to work not only with structured but also with unstructured data, use past experience in order to make smart decisions and predictions. Implementation of ML in embedded systems leads to improvement of processes by increasing their automation and effectiveness. Embedded machine learning finds its application in such areas as computer vision, speech recognition, healthcare, robotics and intelligent automation. In addition, implementation of machine learning techniques increases the level of security and confidentiality of data due to the absence of necessity to send information to distant cloud servers. As the Internet of Things develops, embedded systems become more dependent on large number of sensors that collect large amount of data. Data collection and transmission to cloud servers cannot be always effective; hence, the role of local processing of data with help of machine learning algorithms becomes relevant. The purpose of this survey is to provide an overview of machine learning algorithms used in embedded systems and to study recent trends in the field.

Index Terms: Machine Learning, Embedded System, Natural Language Processing, Internet of Things, Robotics.

1. Introduction

Embedded system technologies [1], [2] are currently facing revolutionary changes by adopting machine learning (ML) technology into embedded systems [3], [4]. Due to the recent breakthroughs in artificial intelligence and machine learning technologies, the application of ML models and technologies can now be realized using embedded systems that were constrained by limited computational resources. The application of machine learning technologies to embedded systems is known as Embedded Machine Learning (E-ML).

Machine learning algorithms and models are nowadays widely used in different parts of our lives [6], [7]. In fact, people constantly use ML models every day without even realizing it by interacting with systems like recommendation algorithms in social networks, weather prediction tools, virtual assistants, customer support systems, and personalized web pages:

1.1 Support Vector Machine (SVMs)

Support Vector Machine (SVM) is one of the most popular machine learning techniques, commonly used in natural language processing (NLP) to enhance classification efficiency. This is a supervised learning approach that operates as a binary classification algorithm through constructing a separating hyperplane for differentiating various data points based on the class they belong to. The difference between logistic regression and SVM is that the latter does not provide class probability; instead, it strives to maximize the margin between the two classes. This technique is especially useful where the connection between predictors and targets is not linear since it can employ kernel methods. SVMs have been commonly used in various classification problems including image segmentation, text classification, and pattern recognition.

1.2 Convolutional Neural Network (CNNs)

These networks share their parameters and are used for image classification and recognition. They are widely used in face recognition, object detection and many more applications. CNN [8][9] takes an image as an input after that classify it and then process it in a category.

1.3. K-Nearest Neighbour (K-NN)

KNN, which stands for K-Nearest Neighbors, is one of the easiest to use and popular supervised learning methods in machine learning. In this approach, new observations are classified on the basis of their proximity to other observations already present in the database. As the KNN algorithm makes no assumptions about the probability distribution of the data, it is regarded as a non-parametric classifier. The technique is simple to apply, understandable, and useful for several classification and regression problems, especially if enough amount of training data is at hand. Moreover, the KNN algorithm is easy to train because no separate model is built here. Nevertheless, its computational complexity rises with the increase in the number of predictors and the size of the database.

1.4 Naïve Bayes

Naïve Bayes is a machine learning algorithm under supervised learning using the concept of Bayes’ theorem and is used to perform classification problems. This algorithm is characterized by the assumption that all attributes are independent of each other, thus simplifying the learning process and making it computationally faster. The Naïve Bayes algorithm performs well in text classification problems when the data is very large and high dimensional, such as spam filtering, opinion mining, and document categorization. Its efficiency is due to its ease of use, minimal computational power required, and quick training time.

2. Machine Learning Models

The use of machine learning allows for more sophisticated processes and enables us to achieve more accurate results. The approach can comprehend both speech and text the way humans do. There is no need for continuous modification or introduction of new rules; the system will be able to interpret and process the information using its acquired knowledge [10-13]. It involves supervised and unsupervised learning processes.

2.1 Supervised learning:

Supervised learning refers to a method of training algorithms using labeled data sets. Under this type of learning algorithm, a model creates a map of input and output data using the mapping functions. Supervised learning requires the presence of a supervisor in training the model, just as students require teachers when learning a particular subject.

2.2 Unsupervised Learning:

Another method of learning used in machine learning is unsupervised learning where patterns are extracted from unlabelled input data. Here structure and pattern are inferred from the input data without requiring any supervision.

The processes that need to be carried out for the purpose of implementing an embedded machine learning system includes training of machine learning algorithms and model execution [15][16]. Machine learning models training can be done using computing clusters [17][18], whereas model execution occurs within embedded devices. Certain models of embedded machine learning use small hardware components such as microcontrollers to run ML models and are referred to as TinyML.

2.3 TinyML:

TinyML is one area of research within EML that is involved with studying low-power components such as microcontrollers. These allow for low-latency, low power consumption and lower bandwidth model inference. Normal computers’ CPUs have an energy requirement ranging from 65 Watts to 85 watts, whereas microcontrollers have power consumption in the range of milliwatts to microwatts.

2.4 Advantages of TinyML

(i) Low Latency: This framework operates on the edge side; therefore, there is no need to send to the server. This results in low latency of output.

(ii) Low Power Consumption: Microcontrollers consume a small amount of power. It allows the devices to operate without the plug-in for a long duration of time.

(iii) Low Bandwidth: The usage of internet bandwidth will be less because there will be no need to send the data back to the server.

(iv) Privacy: There is no storage of data anywhere as this model operates on the edge side.

3. Application of Embedded Machine Learning

3.1 Intelligent Sensor Systems

Intelligent Sensor Systems, also known as smart sensors, combine sensing, processing, and communicating functionalities into a single system. The inclusion of machine learning into embedded sensor systems has become a common practice nowadays [19] – [21], providing real-time processing, analysis, and making decisions based on data collected. These functionalities can also be employed by small mobile and wireless sensor networks [22] – [24].

3.2 Heterogeneous Computing Systems

Heterogeneous Computing System utilizes several processors that perform different kinds of computing. In this way, some parts of computational load are offloaded from the CPU to other processors, significantly improving performance, reducing consumption of resources, and increasing the efficiency of the computations. It becomes especially useful when executing computationally intensive tasks such as those related to machine learning.

3.3 Embedded FPGAS

The increasing popularity of field-programmable gate arrays in both embedded and computing systems is attributed to their advantages of cost-effectiveness, high speed, adaptability, and energy efficiency. FPGAs can be designed in such a way that they help boost the speeds at which machine learning algorithms can run while saving energy costs as well. Parallel processing is also supported by these devices.

4. Conclusion

The paper under consideration emphasizes the great potential that lies in the integration of embedded systems and machine learning solutions. It is possible to implement machine learning within the embedded environment, thereby ensuring more efficient data processing, real-time decisions making, and improvements in system performance in order to develop cutting-edge technologies. Machine learning is very useful for the creation of energy-efficient machines or those working on limited power supply, which can be applied in the areas of microcontroller systems and edge computing. In addition, various machine learning algorithms can be implemented in embedded environments.

References

  1. Singh, S., Singh, R., & Bhatia, M. (2012). Performance evaluation of hybrid reconfigurable computing architecture over symmetrical FPGAs. International Journal of Embedded Systems and Applications, 2(3), 107-116.
  2. Singh, S. K., Singh, R. K., & Bhatia, M. P. S. (2012, December). Design flow of reconfigurable embedded system architecture using LUTs/PLAs. In 2012 2nd IEEE International Conference on Parallel, Distributed and Grid Computing (pp. 385-390). IEEE.
  3. Singh, S. K., Singh, R. K., & BHATIA, M. S. (2010). System level architectural synthesis & compilation technique in reconfigurable computing system. In ESA 2010: proceedings of the 2010 international conference on embedded systems & applications (Las Vegas NV, July 12-15, 2010) (pp. 109-115).
  4. Singh, S. K., Bhatia, M. P. S., & Jindal, R. (2009). Architectural modeling for hardware and software in reconfigurable embedded system. International Journal of Recent Trends in Engineering, 1(1), 575.
  5. Mengi, G., Singh, S. K., Kumar, S., Mahto, D., & Sharma, A. (2023, February). Automated Machine Learning (AutoML): The Future of Computational Intelligence. In International Conference on Cyber Security, Privacy and Networking (ICSPN 2022) (pp. 309-317). Cham: Springer International Publishing.
  6. Singh, I., Singh, S. K., Singh, R., & Kumar, S. (2022, May). Efficient loop unrolling factor prediction algorithm using machine learning models. In 2022 3rd International Conference for Emerging Technology (INCET) (pp. 1-8). IEEE.
  7. Peñalvo, F. J. G., Maan, T., Singh, S. K., Kumar, S., Arya, V., Chui, K. T., & Singh, G. P. (2022). Sustainable Stock Market Prediction Framework Using Machine Learning Models. International Journal of Software Science and Computational Intelligence (IJSSCI), 14(1), 1-15.
  8. Singh, I., Singh, S. K., Kumar, S., & Aggarwal, K. (2022, July). Dropout-VGG based convolutional neural network for traffic sign categorization. In Congress on Intelligent Systems: Proceedings of CIS 2021, Volume 1 (pp. 247-261). Singapore: Springer Nature Singapore.
  9. Kaur, P., Singh, S. K., Singh, I., & Kumar, S. (2021, December). Exploring Convolutional Neural Network in Computer Vision-based Image Classification. In International Conference on Smart Systems and Advanced Computing (Syscom-2021).
  10. Aggarwal, K., Singh, S. K., Chopra, M., Kumar, S., & Colace, F. (2022). Deep learning in robotics for strengthening industry 4.0.: opportunities, challenges and future directions. Robotics and AI for Cybersecurity and Critical Infrastructure in Smart Cities, 1-19.
  11. Panch, T. et al. Artificial intelligence: opportunities and risks for public health. Lancet Digit Health 1, e13–e14 (2019).
  12. Dignum, V., (2018). Responsible artificial intelligence: Designing AI for human values. ITU Journal ICT Discoveries, Vol 1, No.1, pp. 1-135.
  13. Legg, S. & Hutter, M., (2007). A collection of definitions of intelligence. Frontiers in Artificial Intelligence and Applications, Vol.157 pp. 17-24.
  14. Gupta, S., Agrawal, S., Singh, S.K., Kumar, S. (2023). A Novel Transfer Learning-Based Model for Ultrasound Breast Cancer Image Classification. In: Smys, S., Tavares, J.M.R.S., Shi, F. (eds) Computational Vision and Bio-Inspired Computing. Advances in Intelligent Systems and Computing, vol 1439. Springer, Singapore. https://doi.org/10.1007/978-981-19-9819-5_37.
  15. Aggarwal, K., Singh, S. K., Chopra, M., Kumar, S., & Colace, F. (2022). Deep learning in robotics for strengthening industry 4.0.: opportunities, challenges and future directions. Robotics and AI for Cybersecurity and Critical Infrastructure in Smart Cities, 1-19.
  16. Bogale, T. E. & Wang, X. B. & Le, L. B., (2018). Machine intelligence techniques for nextgeneration context-aware wireless networks. ITU Journal ICT Discoveries, Vol 1, No.1, pp. 109-135.
  17. Kumar, S., Singh, S. K., Aggarwal, N., & Aggarwal, K. (2021). Evaluation of automatic parallelization algorithms to minimize speculative parallelism overheads: An experiment. Journal of Discrete Mathematical Sciences and Cryptography, 24(5), 1517-1528.
  18. Kumar, S., Singh, S. K., Aggarwal, N., Gupta, B. B., Alhalabi, W., & Band, S. S. (2022). An efficient hardware supported and parallelization architecture for intelligent systems to overcome speculative overheads. International Journal of Intelligent Systems, 37(12), 11764-11790.
  19. Peñalvo, F. J. G., Sharma, A., Chhabra, A., Singh, S. K., Kumar, S., Arya, V., & Gaurav, A. (2022). Mobile cloud computing and sustainable development: Opportunities, challenges, and future directions. International Journal of Cloud Applications and Computing (IJCAC), 12(1), 1-20.
  20. Singh, R., Singh, S. K., Kumar, S., & Gill, S. S. (2022). SDN-Aided Edge Computing-Enabled AI for IoT and Smart Cities. In SDN-Supported Edge-Cloud Interplay for Next Generation Internet of Things (pp. 41-70). Chapman and Hall/CRC.
  21. Saini, T., Kumar, S., Vats, T., & Singh, M. (2020). Edge Computing in Cloud Computing Environment: Opportunities and Challenges.
  22. Sharma, A., Singh, S. K., Kumar, S., Chhabra, A., & Gupta, S. (2023, February). Security of Android Banking Mobile Apps: Challenges and Opportunities. In International Conference on Cyber Security, Privacy and Networking (ICSPN 2022) (pp. 406-416). Cham: Springer International Publishing.
  23. Otani, T. & Toube, H. & Kimura, T. & Furntani, M., (2018). Application of AI to mobile network operation. ITU Journal ICT Discoveries, Vol 1, No.1, pp. 59-135.
  24. Jain, A. K., & Gupta, B. B. (2018). Towards detection of phishing websites on client-side using machine learning based approach. Telecommunication Systems, 68(4), 687-700.
  25. Gupta, B. B., Gaurav, A., Marín, E. C., & Alhalabi, W. (2022). Novel graph-based machine learning technique to secure smart vehicles in intelligent transportation systems. IEEE transactions on intelligent transportation systems, 24(8), 8483-8491.
  26. BHACHU E.S., (2026) The Role of Memory in Large Language Models: Context Windows, Limitations, and Emerging Extensions, Insights2Techinfo, pp.1
  27. Elechi, P., Ekolama, S. M., Okowa, E., & Kukuchuku, S. (2025). A review of emerging technologies in wireless communication systems. Innovation and Emerging Technologies, 12, 2550005.

Cite As

Garg B. , Sodhi G.K. (2026) Role of Machine Learning in Embedded Systems, Insights2Techinfo, pp.1

91600cookie-checkRole of Machine Learning in Embedded Systems
Post Views: 4
Share this:

Leave a Reply

Your email address will not be published.