On Thursday, 23 June 2016 around 200 participants came to Oberkochen, Germany for the ZEISS Symposium "Optics in the Digital World." Even the weather cooperated, making the sudden switch from "early spring" to "full-on summer" mode. Yet the temperatures of more than 30°C did not distract the participants.
Here are a few statistics: with 47% of the participants coming to industry and 53% from academia, we achieved the intended balanced mixture of representatives consisting of those who conduct research and those who apply it. There were three workshops to choose from: 39% of participants opted for "Computational Imaging", making it the most popular. 31% decided to attend the "AR/VR and Large Data" workshop and 30% "Computer Vision and Machine Learning."
Following the introductory talk from Michael Kaschke, the President and CEO of ZEISS, on the inherent connection between optics and digitization already manifest in optical lithography, Laura Waller explained the principle of Fourier ptychography in her keynote. She demonstrated that improved algorithmics make an impressive increase in performance possible. In his keynote speech, Ingmar Posner from the University of Oxford provided interesting examples on the use of machine learning in self-driving cars. These methods have proven to be highly efficient during testing, which makes them especially well-suited for large data scenarios, e.g. in the segmentation of microscopy images. David Bohn from Microsoft gave the final keynote presentation and discussed the HoloLens. He predicted that new display technology and advances in mobile hardware will foster a revival boom of AR / VR concepts.
After the Symposium, the ZEISS Research Award was presented to Jörg Wrachtrup from the University of Stuttgart and Fedor Jelezko from the University of Ulm for their outstanding work on quantum technology with optically addressable spins in diamond.
Robert L. Byer brought the day to a close with his fascinating insights into the discovery of gravitational waves by the LIGO team.
As previously announced, the results from the workshops will be compiled in white papers and are published on the Symposium homepage. If you're curious to find out more, read on to find a short summary of the challenges ahead and excerpts on recommendations for action.
Computational Imaging (CI): While CI is an established field of research with its own academic community, its advanced forms have only enjoyed moderate commercial success. It is still an open question how CI can contribute to consumer/industrial optics and what the expected benefits might be. In other areas, like medical imaging or microscopy, the benefits of CI are much clearer (for example in applications like phase contrast imaging, 3D reconstruction or multispectral imaging), but there is still a gap between industrial needs and the current state of research.
A further challenge is the holistic design of CI systems which incorporates the whole imaging chain, starting from illumination, incorporating the optical system(s), modeling the sensor and ending with the post-processing and analysis of the data. The current tools are not yet adequate for addressing an end-to-end design of CI systems. Another issue is that the chosen optimization algorithm typically affects the result of the measurement. A simple suggestion for addressing this problem is to record larger and larger datasets for evaluation and testing. Not surprisingly, a central limitation for the short-term future of CI is bandwidth limitations in data transfer.
The action proposals from the workshop are to set up: i) standard benchmark data sets and open implementations for individual problems, ii) standardized data exchange formats and best practices, and iii) educational and community resources. All three topics only exist in a scattered and basic form.
Augmented and Virtual Reality: The traditional conflict between the physical and visual ergonomics of AR/VR devices is still a challenge. On the one hand, the HMD has to be lightweight and compact to be worn conveniently for hours. In addition, the design should be attractive enough to be worn in public (keyword: ‘social acceptance’). On the other hand, most applications require a natural field of view (i.e. for human eyes this would mean more than 150 degrees), a huge eye-box (for a robust projection of the image into the eye even under everyday user conditions) and high optical performance (for a brilliant and crisp 3D-image). Even though there are solutions for a restricted set of these specifications, up until now there is no design that meets all these contradictory requirements at once. In addition, AR devices are even more challenging because the virtual digital content has to be aligned and displayed in 3D over objects in the real world.
Large Data: The amount of digital data generated doubles every two years and is expected to reach 40 zetabytes in 2020. Better assistance and workflows for user interaction are needed to handle data and to boost productivity. The increasing size and the diversity of the datasets makes intelligent screening and evaluation algorithms like Deep Learning indispensable. Integrating machine learning concepts for the detection and classification of structures, geometries and morphologies paves the way for the move from conventional optics to smart systems. The exchange with different software platforms and applications has to be facilitated. Open source approaches and the standardization of interfaces are one consequence of the need for cooperation.
A straight-forward approach is to foster innovation and research programs in the direction of AR / VR and large data. Especially on a European level, existing programs such as the Horizon 2020 (and its sub-programs) or the Marie Sklodowska Curie research fellowship program should have a section on research in AR / VR and large data. Companies with a large industrial impact such as ZEISS could play their part in shifting the focus of such programs towards these key technologies. Similar initiatives can be addressed on a national level. The second task is to provide platforms with publicly available datasets and programs that encourage and facilitate the development in the field of AR / VR and large data.
Computer Vision and Machine Learning: The ability to engineer machine learning into product implementations is one of the key challenges for the adaptation of machine learning methods to industry because machine learning solutions require a diverse set of skills - ranging from math to efficient implementation. The readiness for cooperative research in this field is huge: many companies already publish their underlying basic machine learning methods and research, algorithms are common knowledge. The major platform to deploy machine learning algorithms in the future will be on embedded devices which must be able – consequently – to run these algorithms, requiring energy-efficient devices and algorithms.
The post Results and Impressions from the ZEISS Symposium “Optics in the Digital World” appeared first on Optics Beyond.