While I’ve thoroughly loved my return to academia, unorthodox though it may be, I’ve been feeling more and more that the core principles of my research just don’t have a home in modern academia anymore.
In this post, I want to briefly tell you what those core principles are.
Principles
Small Data I think it is an intrinsic virtue to collect as little data as possible, and to maximize the inference for whatever data you have available. This is bass-ackwards to modern big data approaches, with can give some pseudo-guarantees only on massive datasets.
Reverse Engineering I’m fascinated by real world, messy things. I have zero interest in understanding how they behave in contrived environments, and I know that knowing those conditionally contrived behaviors tell you almost nothing about the in situ behavior. The best way to study something is to understand it in its natural environment - with all those messy confounds and collisions varying how they already do and likely always will.
Math, not measurements To accomplish both of the above, we need to stress the importance of math over measurements. This was a section of my dissertation epilogue, but it has also come up again in the context of my current lab’s work in lesion network mapping.
You have to do the math first, period. Understand what it is you’re wanting to do, what you think you’re doing, and what you’re actually doing - they all differ with some angle. The math, via a system diagram, should be the first step in any investigation; especially one in clinical settings with so much unavoidable messiness. You have to circumscribe that messiness, not put up a pleasant fiction that it will go away for some magical thinking reason [^stats].
Control Theoretic Frameworks My favorite way of accomplishing all of the above: control theory. A control system is a very useful canonical structure, one that bakes in minimal assumptions while being maximally expressive in the space of problems that real-world people tend to care about (like medicine).
Something about “control” centers us, humans, or at least it begs the question of “wait, who is doing the controlling?”. Contrast this with AI, where we evoke this mythical “other thing” that is somehow absorbing all the autonomy and volition (and liability) of the problem at hand. For reasons that are still unclear to me, it also skews heavily pro-human, pro-community, and downright anti-cynicism. The best people exist in control theory, period, and I want to work with them.
So much more to say, but this’ll suffice for a quick post - considering I’ve been drafting this for 6 years…
-V