What is research? What is a project? How do I begin?
Good research is thorough and helps advance the field. You do not need to discover something groundbreaking for it to be publishable. You can build on existing frameworks by tying in new information, run small-scale experiments as proof-of-concepts, propose solutions to problems posed by other papers, and more.
These are by no means exhaustive, but rather serve as some fun areas of emerging research related to general intelligence in AI that might be a good place to start. However, feel free to explore whatever areas strike your team as fascinating!
Multi-agent reinforcement learning
- Communication and language between agents, socialization
Machine learning on Intel (or other) neuromorphic chips (Loihi)
Hebbian learning and spiking neural networks
Neural cellular automata
Measuring integrated information across neural networks
Numenta spiking dendritic networks
Field overview of brain-based intelligent systems
Explore cortical columns and the “Thousand Brains” theory
Explaining how brain structure is related to function in different areas
- Relate this to ML, how can we digitalize these systems?
- Biologically inspired forms of memory in AI and reinforcement learning
- Temporal Difference learning
You can also find concrete problems by
- Looking at the “further work” section of papers (near the bottom or in the Discussion) and seeing what has been done
- Reaching out to researchers at your university
- Going to conferences or watching research presentations at your university
Check out UW I2’s past projects here!https://interactive-intelligence.github.io/research
“Deinforcement Learning” Paper
- Using the neuroscience of pain to investigate how to learn faster from painful stimuli in the framework of reinforcement learning
“Emergent Language: Independent AI Development of a Language-Like Syntax”
- Explore how neural networks’ understanding may change when forced to communicate in discrete symbols to solve visual tasks