This is not a comprehensive list of topics, or a study guide. These are examples; the actual exam questions will be very similar to these. Use the lecture notes (i.e. presentation slides) as the study guide. To get the most benefit out of these examples, try to answer them first yourself. Then go to the slides/readings to make sure you got it right. Then think of other similar questions yourself! Note that there are also review questions at the end of each chapter in the reading assignments which should be useful as well.
Below (as well as in the exam) there are a number of detailed questions about the second half, followed by general questions that compare and contrast the various approaches we've talked about during the entire class (but no detailed questions about the first half). The answers are short, i.e. a couple of sentences or paragraphs each. These are a bit more extensive and challenging than the questions in the exam, but should be good practice!
INTERACTIVE NEUROEVOLUTION
Compare and contrast interactive neuroevolution with traditional neuroevolution. Discuss how incorporating human guidance in the evolutionary process can both enhance and constrain the discovery of solutions. Provide examples from the NERO game or Picbreeder to support your answer.
OPEN-ENDED NEUROEVOLUTION
Discuss the five key elements of open-ended evolution derived from biological principles. For each element, provide an example of how it might be implemented in a computational neuroevolution experiment.
NEURAL ARCHITECTURE SEARCH
Surrogate models and supernets have been introduced to reduce the computational expense of NAS. Critically evaluate these methods in terms of their advantages and potential limitations. How might these techniques impact the discovery of novel architectures?
METALEARNING
How does the Pangaea system optimize activation functions for neural networks? Highlight the significance of adapting activation functions over time and space in a network and discuss the duality between activation function learning and weight learning.
NEUROMORPHIC SYSTEMS
Describe the primary objectives and challenges of neuromorphic computing. How does neuroevolution optimize the design of spiking neural networks, and why is it particularly suited to neuromorphic hardware constraints?
NEUROEVOLUTION + REINFORCEMENT LEARNING
Discuss the key trade-offs between reinforcement learning (RL) and neuroevolution (NE). In your answer, focus on their effectiveness in handling sparse rewards, sample efficiency, and exploration strategies.
NEUROEVOLUTION + GENERATIVE AI
Describe the process of evolutionary model merging in data flow and parameter spaces. How do these methods enable the creation of composite models with emergent capabilities, such as solving domain-specific challenges like Japanese mathematical reasoning?
NEUROEVOLUTION IN BIOLOGY
Analyze the role of neuromodulation in enabling adaptation to changing environments, such as the T-maze navigation task. How does neuroevolution demonstrate the emergence and utility of neuromodulation in these scenarios?
GENERAL QUESTIONS
Discuss how indirect encoding techniques like CPPNs and HyperNEAT contribute to modularity in neural networks. How do these techniques relate to the principle of minimizing wiring length observed in biological systems? Propose a future a
pplication where such integration could be impactful.
Analyze the role of diversity in neuroevolution with examples from novelty search and quality diversity methods. How does diversity contribute to solving deceptive fitness landscapes, and what parallels exist in biological evolution, such as the evolution of mobbing behavior in hyenas?
Compare the evolution of communication codes in simulations of mate selection and hunting with the optimization of prompts in generative AI. How do the principles of encoding, exploration, and adaptability manifest in these two contexts, and what further research could integrate them?