alt text 

Guo Yu (喻果)   Ph.D.
Associate Professor
Institute of Intelligent Manufacturing
Nanjing Tech University, Nanjing, 211816, China

Email:gysearch@163.com
[Google Scholar Profile] [Github Profile]
[ORCiD Profile] [DBLP Profile]
[中文网页]


Bio

Education:

Work Experience:

Research Interests

  • AI for Chemical Engineering

  • Intrinsic Safety and Security in Chemical Engineering

  • Fairness-Aware Multi-Objective Optimization and Machine Learning

  • Evolutionary Deep Learning

  • Multi-/Many-Objective Evolutionary Optimization

  • Data-Driven Evolutionary Optimization

  • Multi-Criteria Decision Making

Research Topics

  • AI driven Chemical Engineering: mainly develop efficient AI models or optimization algorithm to assist in the safty production of the process and discrete industry. The following directions are summarized:

    • 1. AI driven Chemical Engineering: firstly, we can use first principle model to assist in building trust-worthy AI model for safty production of the process industry and discrete industry; Secondly, we can use AI models to refine the first principle model; Lastly, we may use hybrid models consisting of both the first principle model and AI models to get more robust model for secure or safety production either in process industry or discrete industry.

    • 2. Intrinsic safety and Security in Chemical Engineering: how to essentially ensure the safety and security of the production in Chemical Engineering.

    • 3. Various applications: various applications of safty production of the process and discrete industry, etc.

  • Deep learning and fairness-aware optimization: Aim to find robust deep neural architectures and perform optimization with respect to designed fairness. The following directions are involved:

    • 1. Fairness-aware machine learning: Aim to build a model under the consideration of individual fairness, group fairness, etc.

    • 2. Fairness-aware optimization: Aim to optimize objectives with respect to the designed fairness, in order to protect some groups of people in certain situations.

    • 3. Federated learning or optimization: Aim to resolve the grand challenges in big data driven online decision-making in real applications, including the extremely high complexity of the data, strongly limited data interaction and communication, strict data security and privacy, and difficult global real-time optimization and decision-making.

    • 4. Adversarial learning: Aim to find robust deep neural architectures under different attacks.

    • 5. Evolutionary deep neural architecture search: mainly search robust deep neural architectures via evolutionary algorithms with respect to a number of objectives.

  • Single-objective optimization and its applications: mainly develop efficient optimization algorithm to solve large-scale optimization problems with differnet characteristics, including contraint nonlinear programming (NLP) problems, mixed-integer nonlinear programming (MINLP) problems, and expensive optimization problems. The following directions are summarized:

    • 1. Large-scale optimization: Aim to develop high-efficient optimization algorithms to resolve high-dimensional NLP, MINLP, or other constraint engineering optimization problems.

    • 2. Expensive single-objective optimization: Aim to construct cheap meta-model or surrogate model to replace the expensive objective function, aiming to improve the efficiency of the optimization.

    • 3. Various applications: Large-scale Crude Oil Scheduling Problems, Welded Beam Design, Tension/Compression Spring Design, Pressure Vessel Design, Power Dispatch, etc.

  • Multi-/many-objective optimization and its applications: mainly design efficient multi-objective evolutionary algorithms (MOEAs) to solve many-objective optimization problems, dynamic multi-objective problems, constraint multi-objective problems, expensive multi-objective problems, multi-task optimization problems, variable-length mixed-variable multi-objective optimization problems. The following directions are concluded:

    • 1. Preference-driven multi-objective optimization: Aim to facilitate the decision makers in decision making by learning their explicit or implicit preferences in sovling the real applications, such as the crude oil scheduling and hybrid electrical vehicles controller design.

    • 2. Multi-objective or many-objective optimization: Aim to develop novel frameworks of MOEAs to achieve a good representative solution set to the Pareto optimal front (PF) or desired regions of interest (SOIs) of the multi-objective problems on different cases that the number of objectives is high, the optimization environment is dynamicly changing, and the problem with multiple types of decision variables (e.g., the continuous and discrete variables) with a variable-length structure (i.e., the number of the decision variables is unknown beforehand).

    • 3. Expensive multi-objective optimization: In most practical problems, the objective function estimation is expensive. The main research aims to design efficient data-driven evolutionary algorithms to solve the actual costly optimization problems, such as such as the crude refining process and combustion model optimization.

    • 4. Evolutionary multi-task optimization: Aim to introduce transfer learning or other machine learning techniques to assist in simultaneously handling multiple tasks.