As AI continues to develop in great strides, CRIM is there to regulate it and is developing methods to ensure AI remains safe and productive.
CRIM is an applied research centre founded in 1985 by a group of companies and universities who wanted a neutral organisation to accelerate the transfer of new technologies to the marketplace. A not-for-profit organisation, CRIM has played a major role over the past 35 years in achieving many of the milestones that have marked the development of Quebec’s technology sector, particularly in the field of artificial intelligence. CRIM’s main raison d’être is to develop cutting-edge technologies and transfer them to companies, turning them into commercial successes and contributing to the creation of high-tech jobs.
Since its inception, CRIM has been at the heart of the global transformation of industry by AI. The promises of AI are numerous, but it is difficult to evaluate and operationalise, includes a large share of unknowns and risks, and is difficult to implement. The methodologies around new methods are not yet established, and expertise and experience in these areas are scarce.
As a bridge between scientific and academic progress and the industry, CRIM presents a critical mass of AI experts, research scientists, data scientists, software developers, and architects, providing a multidisciplinary expertise to process all types of data: audio, speech, images, video, text, sensor signals, geospatial data, time series, and tabular data. CRIM is also a leader in the development and adaptation of generative AI, in all these modalities, for the creation of reliable solutions in all ranges of problems and fields.
A catalyst for innovation, CRIM has a long track record of delivering concrete solutions using the latest technologies in AI, dedicated first to the needs of the industry and institutions. It offers its customers and partners an advanced know-how in experimental AI development, with a view to delivering concrete solutions. This includes a vast set of methodologies consolidated over hundreds of AI projects, including data-centric approaches, metric development, building evaluation benches, developing and fine-tuning machine learning models in several modalities, MLOps, DevOPs and DataOps, generative methods, trusted AI, etc. CRIM stands out in particular for its ability to apply recent scientific advances in highly constrained application contexts.
Trustworthy AI: Towards a reliable and responsible AI
The world of artificial intelligence is exploding, propelled by the development of deep learning and generative AI, offering possibilities never seen before. Despite this, few AI projects find their way into actual products. The gap between the promises of AI and the practice remains immense. The issues are numerous: the solutions present significant biases, they are opaque and difficult to interpret or explain, they lack consistency and robustness, they are vulnerable to various forms of attack, and generative models are plagued by hallucinations. In general, they are difficult to constrain to standards or specifications. The measurement of the models’ confidence is also a challenge. All these factors constitute a major obstacle to the responsible use of these methods. In critical application domains, such as healthcare, transport, finance, etc., they would simply be unacceptable if used as such, as they would represent a threat to security.
In this context, especially for applications in critical or regulated domains, these approaches need to be framed by a set of methodologies, which we call Trustworthy AI, to deal concretely with all these issues: bias, uncertainty, robustness, compliance, explainability and so on.
As a leader in applied AI for industry, CRIM is investing heavily in the development of trust-based AI. Its integration into the realisation approach is carried out at all levels, in a ‘trust by design’ manner: from the design of the projects themselves, then in all stages of realisation, in the choice of methods and models, in the processing and governance of data, and in the test benches. Everything is guided by CRIM’s goal to systematically develop and transfer Trustworthy AI, in order to increase the real impact of reliable and responsible AI solutions.
Trustworthy AI is a field yet to be built. CRIM is working with industry, in particular through consortia dedicated to this issue, to understand the challenges and use cases, in order to develop and propose solutions.
Combating visual and multimodal misinformation
CRIM aims to get involved in the major issues facing society which could benefit from AI, by contributing with its multidisciplinary technological know-how and mastery of AI development.
Today, misinformation and disinformation constitute a growing threat to society, institutions and individuals alike. The threat is exacerbated by social media, which accelerates the spread of misinformation, and also by the recent rise of generative AI, which enables documents to be produced or modified in a highly realistic way, in all types of media: images, text, audio and video. Technologies are evolving at a breathtaking pace, and at the same time a host of accessible, easy-to-use tools are falling into the hands of the general public. The malicious use of these technologies is likely to have a major impact in all areas and fields of society: in elections and democratic processes, in public health, security, defence, the economy, and so on. These issues are still poorly understood, the mechanisms of disinformation are poorly mastered, and above all, the methods for detecting it are not very advanced, and lag far behind the tools and methods for producing it.
CRIM has joined forces with the University of Ottawa’s Information Integrity Lab to study this issue from a number of social, legal and technological angles. The aim of this collaboration is firstly to consolidate and disseminate knowledge on the issue, and secondly to develop methods and tools to help individuals and institutions detect misinformation. The monitoring work has resulted in a substantial report, the main points of which are currently being disseminated.
CRIM already has extensive experience in related topics: Deep fakes detection, voice and multimodal biometrics, user recognition in audio, etc. Several interdisciplinary projects are underway to apply this expertise to the detection of disinformation. CRIM is actively seeking partnerships to extend the scope of this major project.
Climate data
Climate change and its impact are a major concern for scientists in many fields. CRIM is contributing to this critical issue by leveraging its expertise in geomatics, Earth observation and satellite imagery processing, as well as its know-how in AI and data architecture. On the one hand, CRIM has realised numerous projects and research activities on these subjects (including a first-place finish in a recent competition on sub-seasonal climate prediction), but it has mainly invested in them through its work on climate data platforms and standards.
CRIM has played an active role in a number of Open Geospatial Consortium (OGC) initiatives, and has made several contributions to extensions to the SpatioTemporal Asset Catalog (STAC) used to record geospatial metadata. Through its contributions, CRIM has strategically positioned itself within the steering committees of these projects, developing standards, specifications and reference implementations, enabling the communication and traceability of data sources used in open science efforts on climate.
This opportunity has enabled CRIM to position itself as a Canadian and international leader in climate informatics and geospatial platforms. These projects have led to concrete benefits not only for the scientific community (Marble project developed in collaboration with Ouranos and the University of Toronto), but also for the general public via the ClimateData.ca portal. This portal, developed and maintained by CRIM in the context of a network of scientific and technological partners built by the Canadian Climate Services Centre (CCC), aims to offer quality climate information, and enables all users to visualise, understand and analyse the data available and produced by the very latest scientific advances. This tool is invaluable to scientists and decision-makers alike, helping them to better understand climate risks and their impact.
“Françoys Labonté, CEO of CRIM said: During its nearly 40 years of existence, CRIM has contributed its state-of-the-art expertise to numerous projects involving international collaborations with European partners. We are thrilled by Canada’s recent agreement to participate in Horizon Europe as an associate country under Pillar 2. This exciting development paves the way for a wealth of new collaborative opportunities, and CRIM is eager to explore and engage in these promising ventures”
Please note, this article will also appear in the 19th edition of our quarterly publication.
