Research Internship Offer: Extending Wireless Sensing Generalisability through Novel Machine Le[...]

Epizeuxis; Kevin (MT); On-site; USD 33,480 - 44,640; Part time

Job summary

Epizeuxis is offering an internship focused on wireless sensing to improve generalization of sensing systems across variable environments. You will leverage public datasets to train models and evaluate on real hardware, exploring transfer learning and multi-task learning techniques.

The role emphasizes eagerness to learn and collaboration with the team. Strong ML/DL background and solid programming in C++, Python, or Java are required.

Qualifications

• Hands-on experience with ML/DL
• Solid programming skills in C++/Python/Java
• Knowledge of network protocols functioning is a plus

Responsibilities

• Develop a novel approach to improve generalization of wireless sensing systems across environments.
• Train models using publicly available datasets and evaluate on real hardware.
• Explore Transfer Learning and Multi-task Learning to handle environment diversity.
• Produce results and documentation; contribute to potential PhD thesis considerations.

Skills

Machine Learning

Deep Learning

Multi-task Learning

Transfer Learning

Programming (C++, Python, Java)

Network Protocols

Education

Pursuing PhD in relevant field

Tools

Python

C++

TensorFlow

PyTorch

Job description

Wireless networks are more and more pervasive and ubiquitous in our society where they enable a large variety of applications for our daily lives, while still facing challenges that require them to constantly evolve. At the same time, if they are primarily designed for communication, their spectrum of services is increasingly expanding to include related uses to which they are particularly well suited. Ongoing research around wireless sensing has shown that these networks can be used, among other things, to track the position of users [1], monitor their health (heart rate, breathing rate, sleep quality, etc.) [2], authenticate them in a biometric way (recognize their walking, breathing, etc.) [3], or to recognize what they are doing (gesture and activity recognition [4]). The list is far from being exhaustive. As an emerging multidisciplinary research field, wireless sensing takes advantage of the physical properties of electromagnetic waves when they encounter or travel through obstacles (reflection, absorption, diffraction, etc.) and builds on top of knowledge from different scientific fields – including Networking, Signal Processing and mostly recent advances in Machine Learning/Artificial Intelligence – to enable a wide variety of applications and therefore speed up the entrance in a more connected and smarter world.

However, although being in full expansion, most existing wireless sensing approaches still suffer a lack of generalization capability, i.e., sensing systems generally tend to perform pretty bad in environments different from the one they were trained in. This is because signals used for training the system inherently captures specific characteristics of the environment they are collected in.

The aim of this internship is to develop a novel approach to tackle this problem and make those systems more robust to environment changes. The envisioned approach is to reuse publicly available datasets to train a model capturing a large variety of different environments, while solving the multiple challenges raised by their diversity with the help of Transfer Learning, Multi-task Learning or other suitable approach. Evaluation will be done with existing datasets as well as custom ones collected on real hardware that we already possess.

The most important skill for this internship is to be eager to learn while trying new solutions.

On top of that, the following skills would be strongly appreciated.

Expected Candidate Skills

• Hands-on experience and strong skills in Machine and Deep Learning. Knowledge of modern learning schemes such as Multi-task Learning, Autoencoders and Transfer Learning would be appreciated.
• Strong programming skills in any common language such as C++, Python, Java, etc.
• Knowledge of network protocols functioning would be appreciated.

• This internship can lead to a PhD thesis. Funding already available.

References

[1] J. Wang, X. Zhang, et al., "Device-free wireless localization and activity recognition: A deep learning approach," IEEE Transactions on Vehicular Technology, vol. 66, no. 7, pp. 6258–6267, 2017. doi: 10.1109/TVT.2016.2635161.

[2] S. Yue, H. He, et al. "Extracting multi-person respiration from entangled rf signals," Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, vol. 2, no. 2, pp. 1–22, Jul. 2018. doi: 10.1145/3214289.

[3] F. Lin, C. Song, et al., "Cardiac scan: A non-contact and continuous heart-based user authentication system," in International Conference on Mobile Computing and Networking (MobiCom), ACM, 2017, pp. 315–328. doi: 10.1145/3117811.3117839.

[4] E. Kim, S. Helal, et al., "Human activity recognition and pattern discovery," IEEE Pervasive Computing, vol. 9, no. 1, pp. 48–53, 2010. doi: 10.1109/MPRV.2010.7.

Period & Practicalities

• Starting date: Early Spring 2025. Duration: 5-6 months.