Principal Research Developer

About Mila and BRDG Mila is a world-renowned research institute in artificial intelligence founded by Yoshua Bengio. Recognized globally for its significant contributions to the field of deep learning, Mila has particularly excelled in language modeling, machine translation, object recognition, and generative models. The BRDG team (Benchmarking, Research, and Development for AI Computing Clusters) is dedicated to optimizing computing resources at Mila through technological innovation. This mission unfolds across three main axes: evaluating and benchmarking cutting-edge scientific computing technologies, developing new methodologies to improve the efficiency of research codes, and establishing strategic collaborations with accelerator manufacturers to influence the development of future technologies and facilitate their adoption at Mila. We are seeking a Principal Research Developer to join the BRDG team and transform computational efficiency in AI research at Mila. In this strategic role, you will evaluate new accelerator architectures, work closely with research teams to identify critical components requiring optimization, and develop low-level optimization solutions that maximize the performance of our state-of-the-art computing infrastructure. Your technical expertise will directly accelerate the most critical scientific discoveries of our time by drastically improving computing resource efficiency, paving the way more quickly for revolutionary breakthroughs in numerous fields, including medicine, materials science, and climate change mitigation. As Canada develops its national strategy to compete globally in high-performance computing for AI, your work in optimization and co-development with industry will significantly enhance Mila's strategic role, establishing the institute as a cornerstone of Canada's technological competitiveness and democratizing access to cutting-edge AI technologies for institutions worldwide. Key Responsibilities 1. Advanced Optimization and Performance Evaluation Benchmarking new architectures: Evaluate and analyze performance profiles of new computing accelerators to identify their optimal characteristics and limitations. Bottleneck identification: Conduct in-depth analysis of complex research codes (large-scale model training, inference pipelines, distributed algorithms) to detect computational and memory inefficiencies. Low-level optimizations: Implement advanced optimizations including rewriting custom kernels, migrating simulations to GPUs, and leveraging specialized accelerator features (tensor cores, shared memory, bandwidth optimizations). 2. Cross-functional Collaborations Multi-domain expertise: Collaborate with research groups specializing in diverse areas (natural language processing, computer vision, reinforcement learning, graph neural networks) to understand their specific computational needs and recommend suitable accelerators. Optimization during research: Work closely with researchers to identify stable and critical computational components to optimize within a constantly evolving research environment. Reproduction and optimization: Accurately reproduce research work and develop optimized versions specifically tailored for next-generation computing accelerators. 3. Software and Tool Development Open-source libraries: Design and implement minimal open-source libraries to share optimized algorithm versions with the Mila community and external research ecosystem. Contribution to benchmarks: Develop and integrate optimized workflows into Mila's benchmark suite to standardize performance best practices. 4. Leadership and Communication Representation to partners: Present technical expertise and research projects during strategic discussions with potential industry partners. Scientific outreach: Enhance the team's visibility by presenting optimization work at major machine learning and high-performance computing conferences (NeurIPS, ICML, SC, etc.). Mentorship and team development: Supervise junior research developers and actively contribute to structuring and growing the BRDG optimization team. Requirements Required Skills and Qualifications • Accelerator optimization expertise: In-depth mastery of modern GPU architectures (CUDA, ROCm) and demonstrated experience in low-level optimization, including rewriting custom kernels and leveraging specialized features (tensor cores, shared memory) • Significant industry experience: Minimum of 5–7 years of experience optimizing high-performance systems and developing complex computational solutions, with a proven track record of collaboration in demanding technical environments • Versatility in machine learning: Solid understanding of machine learning architectures and algorithms across multiple domains (computer vision, natural language processing, reinforcement learning), with the ability to quickly adapt to new research paradigms • Advanced programming: Expert-level proficiency in Python and C++, with extensive experience in ML frameworks (PyTorch, Jax), performance profiling tools, and open-source library development • Research collaboration: Demonstrated experience working in dynamic and uncertain research environments, with the ability to identify critical components for optimization despite constant project evolution • Technical leadership: Autonomy in making complex technical decisions and the ability to guide multidisciplinary teams toward optimization goals • English proficiency: Excellent oral and written communication skills in English for effective collaboration with international researchers and scientific publication writing Additional Assets • Scientific publications and outreach: Contributions to research in ML systems, high-performance computing, or computational optimization, with publications in recognized conferences (NeurIPS, ICML, SC) • Advanced hardware expertise: In-depth knowledge of emerging accelerator architectures and experience evaluating pre-commercialization hardware performance • Open-source contributions: History of developing and maintaining open-source libraries used by the AI research community • Mentorship experience: Demonstrated experience in technical supervision and training development teams, particularly in applied research contexts • Presentation and communication: Experience presenting technical work to diverse audiences (industry partners, academic conferences, internal teams) • Specialization areas: In-depth expertise in one or more specific research domains (large-scale language models, computer vision, graph neural networks, reinforcement learning) • French proficiency: Knowledge of French to facilitate collaboration with colleagues and integration into the Quebec work environment Benefits Good reasons to work at Mila The opportunity to contribute to a unique mission with significant impact; The opportunity to contribute to launching a strategic team at the heart of modern AI computing challenges. Access to state-of-the-art computing infrastructure, including a wide variety of GPUs and accelerators. A comprehensive group insurance plan (health, dental, disability, life, travel insurance, and supplementary coverage); An employee and family assistance program; Access to a telemedicine service; A vacation policy offering a base of 20 vacation days from the start of employment; A retirement savings plan with a minimum 4% employer contribution; A generous flexible benefits envelope allowing you to customize your benefits according to what contributes to your well-being. You can select and combine options that suit your needs from lifestyle credits, enhanced insurance, additional vacation days, and an enriched retirement contribution; Flexible hours, summer hours, and remote work possibilities; A workplace located in the heart of Little Italy, in the trendy Mile-Ex neighborhood, close to public transit; A team of domain experts, passionate and inspiring people; A collaborative and inclusive work environment.