In September 2018, EPFL President Martin Vetterli announced the creation of the Open Science Fund to support the best ideas from everyone on campus with a total of CHF 3 Mio over the period 2019-2021.
The two successive calls for proposal attracted nearly 75 propositions submitted in 2018 and 2019. Eighteen projects were selected by the members of the open science strategic committee, joined for the occasion by representatives of the various EPFL school, and will receive support to develop ideas fostering open and reproducible research on campus, and beyond. You can find a short description of the laureate ideas below.
Transportation engineers struggle to produce accurate models of traffic that can contribute in efficient and sustainable mobility management. In a one-of-a-kind experiment, Nikolaos Geroliminis and his colleagues from the Urban Transport Systems Laboratory (LUTS) have used a swarm of drones in a congested urban area to produce a dataset containing trajectories of half a million vehicles, an order of magnitude larger than benchmarks available till now. They will extract and organize this information in order to make this large and fully annotated dataset openly available to all. Researchers from around the globe will be able to use this trove of information to develop and test their own models.
The Energyscope developed at EPFL is a web platform available to anyone interested in assessing the various options for the necessary energy transition to reach the targets of the COP21. The group of Industrial Process and Energy Systems Engineering develops an open access database in order to share and document the data used for generating energy transition scenarios models that are currently mainly based on proprietary data. The platform will contain notebooks that can be used to document, generate, consolidate and trace the data underlying the transition models.
The campus-wide initiative [email protected] and the Swiss Data Science Center (SDSC) join forces to develop a strategy to facilitate Interdisciplinary collaborations in imaging at EPFL, and beyond. Using the functionalities of the open-source platform RENKU, data, metadata and code originating from the imaging community at EPFL will be made available openly to the global imaging community as a repertoire of reproducible, reusable, and well-documented image-processing workflows.
Modern light-microscopy techniques generate tremendous amounts of images and movies that need careful attention to be useful. Like any other digital data, they benefit from following the FAIR principles. The Bioimaging and Optics platform (BIOP) is deploying the open source user-friendly image database software OMERO provided by the Open Microscopy Environment. On top of this infrastructure, it develops a connector that allows researchers at EPFL and beyond for adopting rapid and reproducible image processing with neural networks.
Three EPFL laboratories led by Aleksandra Radenovic (LBEN), Suliana Manley (LEB) and Daniel Sage (LIB) are joining forces to develop a comprehensive platform for researchers in the field of super-resolution localisation microscopy (SMLM). This community-led initiative (SMLM HUB) will provide benchmarking resources, including reference datasets, scientific evaluation of the image reconstruction process, automatic assessment tools, interactive data exploration, and deep learning models. It will promote transparency and reproducibility of complex procedures both for the end-users and for engineers developing new methods.
Experimental research in psychology, neuroscience, and medicine, is in the midst of a “replication crisis”. Researchers from the Laboratory of Cognitive Neuroscience develop the Open Virtual Psychology platform to provide the research community with tools for developing behavioural experiments in immersive virtual reality (VR). Owing to the affordability and availability of state-of-the-art VR equipment, these technologies are progressively becoming more common in the fields of behavioral, social, cognitive sciences, and in basic and clinical neurosciences. The proposed platform will allow packaging, sharing and re-execution of experiments.
With single-cell transcriptomic analyses becoming popular, there are new computational and analytical challenges presented by the data produced. The Laboratory of Systems Biology and Genetics teams up with Gene Expression Core Facility to further develop the web-based Automated Single-cell Analysis Portal (ASAP), adding major new features, including a crowd-based annotation of cells in projects hosted at EPFL and at the Human Cell Atlas (HCA). The database will be later used for the interactive and automated annotation of cells, transforming ASAP into a hub where researchers can openly mine datasets across cell types and species.
The photovoltaics research community has a well-maintained standard for measuring and reporting efficiency of their device, but the equivalent for the solar fuels community is long overdue. In order to visualise, compare and evaluate the performance of new setups, Sophia Haussener and her colleague at the Laboratory of Renewable Energy Science and Engineering (LRESE) build an open access database of experimental descriptions with clear reporting standards. An initial comprehensive systematic review is conducted to populate the resource with key descriptors and classifications (e.g. efficiency, configuration and stability) available in a machine-readable format. It will ultimately become community-maintained.
X-ray, neutron or electron diffraction methods are routinely used to determined the 3D atomic structure of materials, a critical step in understanding and exploiting their properties. Currently, public databases only contain the structural parameters that are the result of interpretative data reduction – so-called refinement. The original diffraction data are rarely openly available. Henrik Ronnow and colleagues from the Laboratory for Quantum Magnetism develop an open science platform where researchers gets rewarded when depositing experimental diffraction data by having access to recent advances in machine learning and ab initio refinement methods.
The integration of Artificial Intelligence (AI) in autonomous vehicles in a real-world context remains a grand challenge: cars and trucks need not only to perform transportation tasks, but also to carry them out in an environment shared with vulnerable pedestrians. The topic of forecasting human behaviour is therefore fundamental. However, the dozen of reports published on the topic report 87% performance discrepancy when evaluating the same forecasting methods on the same data. Alexandre Alahi and colleagues from the laboratory of Visual Intelligence for Transportation(VITA) will make a benchmark publicly available and develop and maintain an open source library of forecasting methods.
The web plays a major role in the adoption of open science best practice. But how can research outputs that cannot be shared as digital artefacts be made openly available? Starting with a cutting-edge structured illumination microscopy (SIM) developed in the Laboratory for Bio- and Nano-Instrumentation, this project will build and share a framework – including IP guidelines – to facilitate the broad dissemination of state-of-the-art instruments in the research community.
This dataset is invaluable for researchers in acoustics, signal processing and musicology, amongst other disciplines. Unfortunately, it cannot be made available due to copyright restrictions. This collaboration between the MetaMedia Center and the Digital and Cognitive Musicology Laboratory will build a secure computation platform allowing anyone to perform open and reproducible research on the archive without touching it.
Transmission electron microscopes (TEM) are widely used research tools that currently generate data and metadata in proprietary and undocumented formats, make it difficult to share results openly. With this project, Cécile Hébert from the Electron Spectrometry and Microscopy Laboratory will create a suitable open standard and metadata scheme for TEM with the help of researchers and manufacturers. The goal is also to develop open tools for reading, converting and manipulating TEM data.
Choosing a laboratory information management system (LIMS) or an electronic laboratory notebook (ELN) can be difficult. The many options available all have their pros and cons, and are not all suited for all research disciplines. Building on an existing open source tool, this project led by the POWERLab will develop a web solution adapted to the needs of the physical sciences, allowing users to adopt important open science best practices.
The Integrated Systems Laboratory at EPFL develops a collection of open software libraries and benchmarks to keep up the driving force in improving optimization algorithms in the field of logic synthesis. This open science infrastructure can be used to tackle complex research problems. Making research software openly available has improved the reproducibility of experimental results and the comparability to the state of the art. Overall, the developed libraries and benchmarks have received positive feedback from academia and industry and have enabled several vivid research collaborations.
This joint project between NCCR MARVEL and Centre Européen de Calcul Atomique et Moléculaire (CECAM) will build an open and collaborative online hub to host existing simulation and data-analysis tools, effectively creating an open-science environment offering software tools as easy-to-use services, with minimal-to-zero setup time.
With one of the largest academic facility for micro- and nano-fabrication in the world, EPFL is well positioned to set some standards in the way information about processes is shared between researchers. This project initiated by the Laboratory of Photonics and Quantum Measurements will design, build and promote a knowledgebase to facilitate the exchange of best practice between users, covering both success and failures.
In an age of fake news and mistrust towards experts, weeding out misreporting of scientific studies is important. The Distributed Information Systems Laboratory will develop SciLens, a platform that automatically generates indicators using weakly supervised learning, effectively helping non-experts to evaluate more accurately the quality of a science news article.