ML4QO: Robust and Efficient Query Optimization and Processing Using Machine Learning

Query optimizers are integral components of database management systems (DBMSs), tasked with identifying execution plans deemed optimal for given queries. However, they often rely on inaccurate parameter estimates and make assumptions that may not align with real-world scenarios. Additionally, their utilization of heuristics in navigating the search space for query execution plans can result in suboptimal join orders. Consequently, at runtime, when these estimates, assumptions, or heuristics prove inadequate, suboptimal execution plans may be selected, undermining the robustness of the database system.

Recognizing these challenges, there has been a growing focus within academia and industry on developing query optimizers capable of learning from data, workload patterns, and past execution errors.

Inspired by this imperative, M2oDA, in collaboration with IBM Db2, has established a dedicated research stream aimed at exploring the application of machine learning techniques to overcome the limitations of current query optimization methodologies. This research initiative delves into various aspects, including but not limited to, ML-driven cardinality estimation, cost modeling, query and plan representation learning, search space enumeration, and robust plan evaluation, with the goal of envisioning effective and practical solutions.

Tagged: machinelearning, ml4qo, ml

Project stars



Remanufacturing - A manufacturing paradigm shift for deep decarbonization in a sustainable economy

A NSERC Alliance project

Every stage of manufacturing, from raw materials extraction, refinement, and processing to part manufacturing contributes immensely to greenhouse gas (GHG) emissions. A shift toward remanufacturing, where worn products or component parts are returned to like-new condition, will dramatically reduce anthropogenic GHG emissions caused by material production and distribution, help industries achieve stringent environmental mandates (net zero GHG by 2050), and contribute directly to a circular economy.

Concept map of the proposed researchThe proposed project will connect expertise across mechanical and environmental engineering, computer science, supply chain and logistics, and climate change governance and sustainability to drive advances in the technologies needed for part assessment and remanufacturing, and generate data needed to build understanding and predictive models that capture the benefits of a remanufacturing-based decarbonization strategy and how it will impact policy.

Specifically, the proposed research (Fig. 1) includes life assessment methodology for identifying damaged zones of parts at the end of their life, remanufacturing technology (cold spray additive manufacturing) to replace damaged parts, reverse logistics to design the optimum operation for accessing parts at the end of their life cycle and transferring them to local remanufacturers, assessment of GHG savings for one exemplar component, and forecast models that predict savings across different scenarios (e.g., for the automotive industry, we would forecast savings for exemplar components in all new cars manufactured by 2035, and 2050). A Transfer Learning algorithm will be employed to extend proposed methodology to other components. Policy barriers at federal, provincial, and municipal levels will be analyzed and mitigation strategies for sustainable transition to remanufacturing will be developed.

The proposed strategy willsignificantly reduce GHG, provide exceptional economic opportunities for SME, and generate required data for decision makers to propose policies based on remanufacturing.

Tagged: remanufacturing, decarbonization, industry

Project stars



Workload-driven query planning and optimization using machine learning

An IBM CAS Canada project

Infusing AI in Db2 is a key strategy within IBM. The query optimizer, in particular, is an ideal candidate for such an infusion, which can follow a similar path of AI infusion in machine translation of languages. The architecture of the optimizer model started off with rules and heuristics and was followed by a reliance on data statistics to estimate and plan the access and join strategies. As with machine translation that started off with a rule based approach, then a statistical approach end eventually an Machine Learning (ML) approach, the next generation optimizer using ML promises to deliver superior capabilities at a much less cost of development. The key goals are to: (a) Simplify performance tuning and even achieve complete automation. (b) Achieve reliable and robust query performance by constantly learning and improving the model. (c) Simplify Optimizer development by training the model in the specific user environment. (d) Infuse ML gradually and seamlessly.

Tagged: machinelearning, queryoptimization, ibm

Project stars



Holistic Innovation in Additive Manufacturing 2.0 (HI-AM 2.0): Capitalizing on Prior Achievements and Exploring New Frontiers in Directed Energy Deposition Processes

The Pan-Canadian NSERC Strategic Network, "Holistic Innovation in Additive Manufacturing (HI-AM)," was established in 2017 to coordinate and accelerate national research efforts in additive manufacturing (AM). Despite the challenges posed by the pandemic, HI-AM 1.0 achieved exceptional outcomes. For example, the initial goal of training 85 highly qualified personnel (HQP) was surpassed when 131 HQP were successfully trained. The network's principal investigators and HQP generated 405 publications, filed ten patents and invention disclosures, and registered two standards development working groups with ASTM/ISO. One of the standards drafts is currently undergoing an international ballot.

Building on the success of HI-AM 1.0, the team will leverage the established network through multiple independent NSERC Alliance proposals. The current proposal focuses on Directed Energy Deposition (DED) Processes, one of the seven classes of AM. It aims to address two primary research thrusts: 1) Model Driven Workflow for In-Situ Monitoring and Quality Assurance in Robotics DED Platforms used for Large-scale AM, and 2) DED Process Development for Advanced Materials, Novel Products, Large Scale Manufacturing, and Parts Refurbishment/Repair. The multidisciplinary HI-AM 2.0-DED team, comprising seven universities and eleven companies, has made significant advancements in intelligent control technologies, innovative DED processes, and new materials. These achievements provide a foundation for overcoming obstacles in commercial AM applications within the framework of HI-AM 2.0 and position Canada as an important player in the global AM supply chain. The team's industry partners span sectors such as aerospace, automotive, defense, energy, natural resources, and tooling, underscoring the potential impact of their research.

This project aims to sustain the momentum of AM initiatives in Canada, given the critical timing amidst global geopolitical and supply chain challenges. The proposed Alliance seeks to establish partnerships, develop intellectual property, and train HQP essential for Canada's competitiveness in the emerging field of AM.

Tagged: hi-am, additivemanufacturing

Project stars