W2.1 Lightweight Models for Real-time Applications

Teaching Methodology: 

will follow problem-based learning (PBL) paradigm. By transforming students' initial research interests into "trigger problems" that drive the entire learning process, this workshop aims to cultivate students into active, collaborative, and self-directed learners. The LM-RtA follows up the 4 principles of PBL, which are constructive, collaborative, contextual, and self-directed. . Those four pillars can ensure that learning is not only the transmission of knowledge but also active construction of meaning. In the workshop, we will utilize “Maastricht 7-Jump” as the core activity cycle.

Research focus: 

object tracking, robust localization, picture fusion, high-speed scenario applications, and integrated navigation systems. The designed models can be used and applied in a variety of sectors like dynamic environmental monitoring, autonomous cars, and robots. The considered techniques used in LM-RtA will include: (1) CNNs: is an ideal model to process the image data from different sensors to extract and learn features from the received data, thus making the designed algorithm integrated with image-base sensors. Extended/advanced versions such as fast R-CNN and Mask R-CNN have better capability to identify the objects in the images. The lightweight CNN-based model such as SuffleNet is an efficient model to work on the limited computer powers, which is possible to be adapted in Task 3.3; (2) Recurrent Neural Network (RNN) / Long Short-Term Memory (LSTM)/ Gated Recurrent Unit (GRU): are effective models to process the time-series data, e.g., motion or environmental data, and detect the moving objects.

Supervisors:

Prof. Jerry Chun-Wei Lin (SUT)

Prof. Dariusz Mrozek (SUT)

Prof. Rafał Cupek (SUT)

Core activities:

1. Qualification of candidates:  Due to the limited number of places at PBL course, candidates will be asked to solve qualification tests and to present proposed methods of solving. Only candidates who successfully pass the entry tests will be admitted to further points.

2. Clarify terms and concepts: All students should have the common understanding of the topics of each PBL in the project.

3. Define the research problems: Each presenter should precisely mention the core research problems of the studied/presented papers for discussion.

4. Analyze the problem in real cases: Activate the existing knowledge, what do we know about the research problems in real applications/cases.

5. Structure and hypothesize: Participants should be divided into several groups for systematically brainstormed ideas and propose the potential solutions/ solution paths in real case scenarios.

6. Formulate the learning objectives: Each group should be based on their expertise combined with the presented topics to precisely list what the specific knowledge or skills they “need to learn and understand” to solve the potential and new research problems, in which this is the basic self-directed learning principles of their studying.

7. Independent Study: Each group work individually together with its members (participants) to conduct the independent (external group) but collaborative (internal members) research activities (e.g., literature, watch tutorial videos, finding the data, attending the relevant lectures…etc.).

8. Discuss and Synthesize: The group should share their learning materials and outcomes as the report, and present the results in the final discussion.

9. Final grade: - will be awarded based on the report and final discussion conducted within the PBL

ECTS: 5

Literature:

[1] Chip, H. (2022). Designing machine learning systems: An iterative process for production-ready applications; ISBN-13978-1098107963; PublisherO'Reilly Media; Publication date June 21, 2022

[2] https://efficientdlbook.com/

[3] Situnayake, D., & Plunkett, J. (2023). AI at the Edge. " O'Reilly Media, Inc.".

[4] Iodice, G. M. (2022). TinyML Cookbook: Combine artificial intelligence and ultra-low-power embedded devices to make the world smarter. Packt Publishing Ltd.