This notebook begins with a perspective, rather than a result. I want to understand intelligence that can be interpreted and designed by looking at the architectures and learning algorithms that give rise to it. I call this perspective Architecture Intelligence.
There are no paper readings or project results to report yet. This opening note sets out the questions I want to carry into that work.
What I mean by architecture
Here, architecture includes more than the arrangement of layers or connections in a neural network. I also want to consider learning algorithms and the way computation is organized. My interest is in how these choices relate to the capabilities a system develops, and how that relationship might help us understand or design it.
This is a working lens for inquiry, not a claim that one concept already explains intelligence.
Representations are one part of the question
Neuronal representations and their geometry, including manifolds, are an important part of this perspective. But I do not want to stop at describing a representation.
I want to ask what produced it, how the system uses it, and what changes when the architecture or learning rule changes. A useful reading note should distinguish what a paper observes from the mechanism it proposes, and from my own interpretation.
Reading AI alongside neuroscience
The reading will focus on AI, while drawing questions from neuroscience and cognitive science. My background makes the connections interesting, but a connection should be examined rather than assumed.
When reading a paper, I want to keep track of:
- The question it asks, and the assumptions behind that question.
- The architectural choices and learning procedures it studies.
- The evidence for its explanation, and where that evidence stops.
- What I understand, what I doubt, and what I still need to learn.
- A small experiment that could make a promising question more concrete.
These are prompts for my reading, not a checklist every paper must satisfy.
From a note to a project
Not every paper needs to lead to a project. Some notes may clarify a concept; some may reveal a misunderstanding. When an idea does seem worth testing, I want the project to retain a link to the question that started it.
The record should include the experiment, implementation decisions, failed attempts, and changes in my understanding—not just a polished final result. A project log can point back to a paper note or an earlier idea, so the reasoning remains visible.
A notebook that can change
This blog is first a record for myself. I also hope it will help future collaborators understand how I think and work. Its value should come from the actual reading and practice that accumulates here.
The perspective itself is open to revision. If a paper or an experiment changes how I think about architecture, learning, or representation, that change belongs in the notebook too.