Impact Report 2023-2024

With the alarming decline in insect populations, which poses a serious threat to the balance of ecosystems and agriculture, the importance of research into species discovery, monitoring and analysis is greater than ever. Led by Professor David Rolnick's lab and Mila's applied projects team, in collaboration with the Montreal Insectarium, Antenna is an online platform that enables entomologists, ecologists and computer scientists to create reliable and dynamic image datasets using advanced machine learning techniques for real-world applications. Through comprehensive insect monitoring, Antenna provides experts with the data they need to inform policies on land use, climate change and biodiversity conservation.

In the context of game world creation, what constitutes a healthy dynamic between purely learning-based and purely human methods of content creation raises important questions that need to be addressed. As AI continues to advance in the video game industry, Mila and Ubisoft have teamed up to explore the responsible use of AI in scaling game worlds. Launched in September 2023, this five-year initiative led by Professor Derek Nowrouzezahrai aims to guide technological innovation in the video game industry and help train a new generation of professionals skilled in evaluating and integrating creative AI solutions. 

The advent of surprisingly competent generative AI has shifted the perspectives of many leading researchers, including Mila’s founder and scientific director Yoshua Bengio, on the future of AI. These experts now believe that AI models may approach human-level capabilities and beyond much sooner than they had expected and that reaching this milestone could involve major risks. This observation has prompted Professor Bengio to focus much of his research on improving AI safety, particularly by addressing alignment and control problems. In June, he launched Safe AI for Humanity (SAIFH), an independent research endeavour with a dedicated team searching for ways to develop safe-by-design AI. 

David Rolnick, Core Academic Member of Mila, published a paper arguing that most AI algorithms are not designed with specific problems in mind and often fail to solve them. Instead of focusing solely on developing increasingly powerful general AI models, Rolnick advocates for addressing the particular requirements and constraints of real-world problems. He suggests that application-driven and methods-driven approaches to AI research should complement each other. By focusing on the needs of specific problems, researchers can create cross-cutting innovations that are applicable to a wide range of challenges.

Last November, at the UK’s AI Safety Summit, our founder and scientific director, Yoshua Bengio, was commissioned to chair work on the International Scientific Report on the Safety of Advanced AI as part of an international mandate following a resolution by representatives of 30 countries, the EU and the UN. Drawing on the work of the Intergovernmental Panel on Climate Change (IPCC), this initiative aims to provide policy-makers and global audiences with a shared, science-based, up-to-date understanding of the safety of advanced AI to inform national and international policy making. The interim report was published in May during the AI Seoul Summit. The final report will be presented at the AI Action Summit in France in February 2025.

In February, Mila welcomed key stakeholders from government, academia, civil society and industry for its first international conference on AI and human rights. Co-organized by Catherine Régis, Full Professor at the Université de Montréal and Associate Academic Member of Mila, and Benjamin Prud'homme, Mila's Vice President of Policy, Safety and Global Affairs, the three-day event highlighted the urgent need to integrate human rights in AI governance mechanisms. The conference opened with a highly anticipated conversation moderated by our president and CEO, Valérie Pisano, featuring Amandeep Singh Gill, the UN Secretary-General's Envoy for Technology, Alondra Nelson, Harold F. Linder Professor at the Institute for Advanced Study, and Yoshua Bengio.

In September, Mila, in collaboration with the Office of the UN Secretary-General's Envoy for Technology and the OECD GPAI, co-organized the high-impact dialogue “Towards a Common Understanding of AI Capabilities, Opportunities and Risks: Forging the Path for a Positive Future for All,” held at UN headquarters in New York. Given the urgent need for cross-institutional collaboration and knowledge sharing to address the complex challenges posed by AI, the discussion addressed key issues such as balancing innovation and ethics, equitably distributing the benefits of AI, and managing the associated risks.

In spring 2024, our founder and scientific director, Yoshua Bengio, was included in the TIME100 annual list of the world's 100 most influential people, alongside the likes of Nvidia CEO Jensen Huang, Russian public figure Yulia Navalnaya, and American football quarterback Patrick Mahomes. 

In September, he also appeared for the second year running on TIME’s 2024 list of the most influential people in AI. 

Mila researchers, in collaboration with BrainBox, used reinforcement learning techniques to improve the performance of heating, ventilation and air-conditioning (HVAC) systems, which are responsible for an estimated 15% of global CO₂ emissions. The applied project optimized energy consumption by predicting indoor temperature and HVAC system energy use through physics-informed machine learning models, creating a predictive algorithm to adapt indoor temperatures while conserving energy during peak demand times, without compromising occupant comfort. By enhancing prediction accuracy and data efficiency for sustainability, the collaborative project addresses major challenges in building management and energy conservation.

In April, Canadian Prime Minister Justin Trudeau announced a historic investment of over $2 billion in AI, including a strategic national computing infrastructure and the establishment of a dedicated AI safety research institute. Mila is proud to contribute to Canada’s global AI presence and strengthen its world-leading AI research ecosystem, all while advocating for safe and responsible AI development.

Mila researchers collaborated with Amgen to create an accessible, open-source protein language model that offers competitive performance without the need for exorbitant financial or computational resources. By democratizing AI in life sciences, this cutting-edge model empowers smaller labs to participate in drug and protein discovery and drive global advances in therapeutic research.

In partnership with Hydro-Québec, Mila’s Applied Machine Learning Research Team adapted computer vision algorithms for anomaly detection in robotic power line inspections. The researchers produced a robust dataset of annotated anomalies across various cable types that enables more extensive and in-depth algorithm analysis for inspecting cable conditions across the network, notably in few-shot and zero-shot learning scenarios. The project is expected to enable faster, more effective anomaly detection, ultimately enhancing Hydro-Québec's ability to improve network security and operational efficiency. This research, which produced a paper accepted at NeurIPS 2024, fills a critical gap by offering realistic benchmarks for AI models, with the potential to generalize to complex, real-world scenarios and improve AI adoption in industrial applications.

Mila’s partnerships are major drivers of applied research and innovation. In the past two years alone, Mila researchers have published 47 papers in collaboration with partner Intel, many of which have been accepted and showcased at the top global academic conferences such as NeurIPS, ICLR and ICML. The research spans a wide variety of fields, including material discovery, climate change, drug discovery and omics. 

Mila concluded a collaborative project with D-BOX, a world leader in haptic and immersive experiences, to integrate AI in the automation of haptic feedback in movie theatre seats, enhancing realism for a more immersive cinematic experience. The project leveraged deep learning to detect instances where motion, texture and vibration could be integrated in audio and video content to synchronize seat movements. These automated tools helped D-BOX improve efficiency and scale its design processes, demonstrating AI's potential to transform creative industries and offer audiences an elevated entertainment experience.

Launched at Mila in May and delivered in partnership with Indspire, Indigenous Pathfinders in AI is a pioneering summer program designed to ignite Indigenous talent to learn, develop and lead the evolution of AI. Rooted in community worldviews, it bridges Indigenous perspectives with AI technologies. For its inaugural session, the program brought together 11 First Nations, Métis, and Inuit participants from across Canada. Throughout the summer, they engaged in a blend of classroom learning and hands-on experiences, developing AI projects aimed at benefiting Indigenous communities in Canada. To mark the program’s conclusion, they presented three AI prototypes addressing critical issues such as Indigenous food sovereignty, access to education and health.

In December, Dhanya Sridhar, Assistant Professor at the Université de Montréal and Core Academic Member of Mila, represented our institute at the Women in Machine Learning Workshop (WiML) during NeurIPS 2023. This annual event offers women and women-identifying individuals a platform to engage in panel discussions and share their work with established female researchers from academia and industry. Professor Sridhar hosted a roundtable and, alongside her PhD student Mizu Nishikawa-Toomey, shared insights on the experiences of underrepresented groups at Mila while providing guidance to aspiring younger researchers in academia.

In March, Mila celebrated its growing alumni community by hosting its first annual alumni event. This memorable evening offered an opportunity for reunion, reflection, and celebration of Mila's 30-year journey, from its beginnings at the Université de Montréal to the present day. The 150 or more attendees enjoyed inspiring remarks from professors Yoshua Bengio, Nicolas Leroux, Joelle Pineau, Hugo Larochelle and Golnoosh Farnadi, followed by a delightful evening of food, drinks and music.

Inaugurated in 2019, Mila’s 90,000-square-foot office in the Mile Ex neighbourhood needed an upgrade to better accommodate our expanding research community and the evolving needs of the post-pandemic work environment. To this end, our teams collaborated with a design partner on a comprehensive revitalization effort. The initiative included many improvements, such as creating more collaborative spaces, enhancing our common areas, greening our environment, and launching initiatives to creatively showcase the stories of Mila and its members over the years.