Mila is primarily a machine learning research center focused on deep learning and reinforcement learning.
At Mila, we continually push the boundaries of cutting-edge research. Mila is best known for its scientific breakthroughs in deep learning, having contributed many of the first papers in the field, such as:
- introducing word embeddings in neural nets
- denoising autoencoders
- deep networks using ReLU instead of tanh or sigmoid functions self-attention and the consequential revolution in machine translation and natural language processing (NLP)
- generative adversarial networks (GANs)
- and more.
Mila also published one of the most successful textbooks on deep learning (MIT Press 2016) and co-founded the International Conference on Learning Representations (ICLR).
Mila has been a hothouse for the development of deep reinforcement learning and the theoretical foundations of deep learning (mechanisms, optimization methods and analysis, generalization, in and out-of-distribution, causality, generative and probabilistic methods), building a critical mass of expertise at the intersection of theoretical neuroscience and deep learning (a.k.a neuro-AI), most recently by focusing on high-level conscious processing to help deep learning move from its current system 1 competence (intuitive knowledge) to a joint competence of system 1 and system 2 levels (verbalizable knowledge, reasoning, systematic generalization).
Mila has also developed expertise in many application areas of machine learning (with a focus on AI for social good), from traditional areas such as computer vision, NLP and robotics to applications in healthcare (e.g., medical imaging, drug discovery), the environment (co-founding the Climate Change AI organization), AI4Science (to help modeling and discovery in physics, chemistry, biology, etc.) and social sciences (fighting sexual exploitation, contributing to the legal and philosophical aspects of Responsible AI, notably through the Montreal Declaration for the Responsible Development of AI).
Visit the Core Expertise page to learn more.