Project Description: New applications of machine learning are rapidly emerging in advanced manufacturing and are part of the movement towards more intelligent factories. Precision industries, and in particular semiconductor and microdevice assembly, are already using many new tools from modern computer vision for improved process control and anomaly analysis. While the latest deep learning models are vastly improving on traditional techniques for visual recognition tasks, their benefits come at a cost: the effort and time required to label and annotate large amounts of data. This human supervision is even more tedious and slow in the case of precise object detection applications, where scaling the model training process is of high value. This project aims at exploring new methods for rapid labeling of machine learning datasets. One interesting direction will be the use of textual and basic graphical primitives to annotate images. Such labeling “by description” has been recently proposed and applied to a limited number of use-cases. In this project we aim at demonstrating the feasibility of textual and quick/imprecise visual labeling for semiconductor inspection data. We can build on the recent work on weak/distant supervision, generative models, visual-semantic embedding, and few-shots learning, to develop a new type of image labeler that is much easier to use and scale. Other multimodal approaches to data annotation will be also explored to exploit useful data correlations that will enable easy labeling of various objects. Additionally, machine learning frameworks, based on deep learning and beyond that can handle input noisy labels, will be a key focus of this work. The intern is expected to:
- Perform a background literature study focusing on state-of-the-art techniques in weak and distant supervision
- Select a limited number of methods for evaluation and further development.
- Propose a design for a new rapid labeling solution
- Use ASM PT relevant datasets to benchmark and test the proposed design
- Compile the results and analysis in a final and mid-term reports.
Frome, Andrea, Greg S. Corrado, Jon Shlens, Samy Bengio, Jeff Dean, and Tomas Mikolov. “Devise: A deep visual-semantic embedding model.” In Advances in neural information processing systems, pp. 2121-2129. 2013.
Ratner, Alexander, et al. “Snorkel: Rapid training data creation with weak supervision.” Proceedings of the VLDB Endowment 11.3 (2017): 269-282.
Company Description: ASM Pacific is a world-leader in the field of semiconductor equipment. At ASM ALSI in Beuningen, the Netherlands, we develop an advanced multi-beam laser dicing tool, but we also have a technology development team (ASM Center of Competency) that supports the global ASM company with high-tech innovations in various domains, like artificial intelligence, advanced motion control, industrial vision, machine learning and data science. In both the laser dicing business unit and the Center of competency ASM Pacific Technology provides challenging internship positions for university level students. Both short-term (3 months) as well as full-lengths (9 months) MSc internships are supported. We currently have several master students working on their internships in the fields of advanced motion control and artificial intelligence and are always looking for new projects + students.
A dynamic high-tech industrial environment in which you learn to apply academic knowledge to real-life problems
Exposure to the larger ASM organization (>16000 person), opportunities to interact/present to colleagues in Hong Kong and Singapore.
Solid supervision by senior professionals that have a strong academic and industrial background
A market-conform internship compensation and travel expense compensation
Contribute to state-of-the art industrial developments