Deprecated: Unparenthesized `a ? b : c ? d : e` is deprecated. Use either `(a ? b : c) ? d : e` or `a ? b : (c ? d : e)` in /home/sciehar4/ on line 446
Canadian Science Policy Centre | Panel 309 - Policy considerations around the convergence of high performance computing and artificial intelligence

Panel 309 - Policy considerations around the convergence of high performance computing and artificial intelligence

Conference Day: 
Day 2 - November 8th 2018
Takeaways and recommendations: 

Policy considerations on the convergence of high performance computing (HPC) and artificial intelligence (AI)

Organized by: Compute Ontario

Speakers: Chris Loken, PhD, Chief Technology Officer, Compute Ontario; Suzanne Talon, PhD, Chief Executive Officer, Calcul Québec; Alison Paprica, PhD PMP, Vice President, Health Strategy and Partnerships, Vector Institute; Assistant Professor, Institute for Health Policy, Management and Evaluation (IHPME); Alain Veilleux, Chief Technology Officer, Calcul Québec

Moderator: Nizar Ladak, President & Chief Executive Officer, Compute Ontario

Takeaways and recommendations

  • The 2018 federal budget included a one-time investment of $572.5 million over five years (with $52 million on an ongoing basis) to implement a Digital Research Infrastructure Strategy that will deliver more open and equitable access to advanced computing and big data resources to researchers across Canada. This provides an opportunity to review policies related to issues like big data, skilled labour, technology investment, public-private collaborations, multi-level governance, funding, evaluation and impact.

  • HPC methods are continually evolving and AI/ machine learning is the latest area of convergence.

  • The convergence of HPC and AI has already begun, and as more research disciplines (e.g., drug discovery, weather forecasting, land use, psychology, construction) begin using AI to guide their simulations, analyze data and tackle more complex problems, demand for HPC will continue to increase.

  • Converging AI and HPC can create economies of scale, avoiding duplication of hardware and competition for skilled researchers to support systems. It also minimizes the need to move large amounts of data between HPC and AI sites. It makes sense to have an integrated system where researchers can do both HPC and AI.

  • It is important to distinguish between the people that do research into AI and those who use AI in their research, and to recognize that key scientific research achievements result from collaboration between the two research areas

  • AI evolves faster than computers so that can be a challenge as the two fields converge.

  • Some researchers need the ability to develop tools specifically designed for their needs (i.e., their own portal for data sharing). Being able to work closely with HPC teams can help to facilitate this.

  • Canada needs to go big in platforms that have data at their heart, especially in areas where we already have large datasets (i.e., health, cities, financial).

  • Rather than partition funding for compute power, connectivity, storage etc., invest in data platforms that bring all the required elements of digital infrastructure together, including large datasets and people who make the whole thing run are recommended. 

The need for both people and computing resources

  • Canada needs to emphasize the training of graduate students, post-doc fellows and principal investigators to fully exploit the power and capabilities of HPC.

  • Competition is increasing for talent. Offering challenging work is often a bigger incentive for recruitment than salary levels.

  • “To out-compete, we must out-compute” (US Council on Competitiveness)

  • Two recent Compute Ontario reports highlight the strengths and gaps in provincial HQP and computing-related resources. Recommendations include:

    • Increased access to accelerators and other ARC resources designed to support emerging applications involving big data and AI

    • Consider a new initiative dedicated to AI and high performance data analysis

    • Broaden industry access to HQP and systems

    • Predictable and sustained funding

Canada’s data advantages

  • The data contained in provincial single-payer health systems can provide insights into wellness and disease that of which other countries can only dream. Canada also has people who are highly skilled at using that data (e.g., clinician scientists).

  • Funding from the Government of Ontario is making it possible to move population-wide longitudinal health data holdings into a secure environment with the compute power required for modern machine learning research. The new HAIDAP platform is a partnership between HPC4Health, IC/ES and the Vector Institute.

  • HAIDAP is operational in Ontario, but could be scaled nationally.

  • Siloed data assets have more value when they can be combined for analyses with other assets (note that data assets do not necessarily have to move or be combined in a central repository to accomplish this).

Collaborating with industry

  • Current efforts to create a national Digital Research Infrastructure Strategy have focused primarily on HPC, big data and network capacity. But it is important to also focus on complementary activities such as scaling AI and promoting academic-industry collaborations, which requires additional funding (i.e., for skilled personnel).

  • HPC infrastructure in Canada has been primarily funded by the Canada Foundation for Innovation, which means its main users have been academic researchers. Broaden the definition of researcher to include industry and look for more opportunities to collaborate (SOSCIP is a good example of industry-academic collaboration.)

  • If we want industry to be successful in using HPC and AI they need support from organizations like Calcul Québec and Compute Ontario. As such, it is important to fund that type of support.