Research
Nature Inspired Computing and Engineering (NICE)
The Nature Inspired Computing and Engineering (NICE) Group is led by Prof. Dr.-Ing. Yaochu Jin, who is an Alexander von Humboldt Professor for Artificial Intelligence funded by the German Federal Ministry of Education and Research. The research of the NICE group is centred around the computational modelling of three main adaptation mechanisms in biology, namely evolution, development, and learning, which together have shaped the most intelligent systems in nature, the human brain.
Evolutionary Developmental Approach to Artificial Intelligence
While most current artificial intelligent (AI) systems heavily rely on big data to accomplish a single task, the human brain is capable of learning multiple tasks on small data. Disentangling the working mechanisms and principles behind the self-organization of biological nervous systems may provide insights into effectively simulating brain functions. We adopt the evolutionary developmental approach to AI, which uses evolutionary algorithms to generate developmental rules that eventually shape the architecture of a complex neural network. We will adopt developmental and generative representations to evolve complex and modular neural network models. In addition, we will consider both activity-independent and activity-dependent plasticity, and the coupling between neural development and morphological development for agents embedded in a complex and energy constrained environment in which they must autonomously accomplish many complex tasks. Spiking neural networks driven by neural plasticity and memory capacity will be designed for accomplishing advanced learning algorithms and hardware implementation will be investigated to demonstrate the effectiveness of neuromorphic hardware in developing brain-like AI techniques.
Trustworthy Multi-Objective Learning and Optimization
While machine learning has witnessed great success in solving many scientific, technical, and societal problems, concerns over security and privacy of the data used in machine learning have seriously increased. Here, we focus on two important aspects in trustworthiness, namely privacy preservation and security protection. To preserve the privacy of data, federated learning, differential privacy and homomorphic encryption methods will be integrated to achieve high-performance, privacy-preserving, communication efficient and lightweight deep learning via evolutionary multi-objective learning. Another important aspect of trustworthiness is to make sure that the learning model is insensitive to adversarial attacks on the model as well as on the data. It is particularly important to understand how uncertainty in the input data as well as in the model propagates, and how this affects the decision-making process, and in which situations a neural network model may completely fail.
We will also seamlessly integrate data-driven evolutionary optimization with federated learning to develop federated data-driven optimization algorithms for privacy-preserving and fairness-aware optimization and decision-making. To this end, new techniques for evolutionary multi- and many-objective optimization, robust optimization and dynamic optimization will be further studied. Bayesian optimization and various machine learning techniques, including semi-supervised learning, transfer learning and self-supervised learning will be made use of to assist federated data-driven evolutionary optimization.
Real-world Applications of Learning and Optimization
The NICE group is keen to apply learning and optimization algorithms to real-world problem-solving, such as industrial optimization, digital manufacturing, swarm robotics and intelligent control in close collaboration with industry. We are particularly interested in interdisciplinary research collaborations with the medical doctors and biological researchers to tackle various healthcare problems, ranging from medical image processing, disease diagnosis support, and personalized medicine. Healthcare problems will be used as an important use case for trustworthy learning and optimization algorithms.