I wrote these questions in 2022, and I spent much of the next year trying to answer them. If you're curious about particular questions and want to know my answers, feel free to email me.

1. What is a good 1st order electrical/sound model of the brain?

   • Electric Fields of the Brain proposes a 4 concentric sphere model, with fixed dielectric constant and conductivity for each layer.

2. How important is the diploe layer in scattering of ultrasound? What is a good model for the diploe layer?

3. What is the difference between optics and sound in terms of wave propagation in the brain? Why is ultrasound attenuation lower compared to electromagnetic waves for a fixed frequency? Why are ultrasound power limits higher?

4. Why is the signal from EEG no good?

5. Why can magnetic fields penetrate the brain? 

   • Possibly because the brain (and most materials) have negligible magnetic susceptibility.

6. What level of precision do humans have in their control of neurons? Can humans control a single neuron? To what extent can you control arbitrary subsets of neurons?

7. What happens if you heat up or cool down the brain?

   • One effect — though not sure how significant — is you’d change the probabilities of ion channels opening/closing.

   • Apparently, you can’t heat the brain by more than 2º C (how do we know this?)

8. Why do axons not look like dendrites?

9. What is an electrode made of? Can you make them arbitrarily small?

10. What are the sources of noise in neural recordings?

    • Johnson noise is a lower bound on electrical noise. Ion channels open/close stochastically according to a Boltzmann distribution.

11. What is Johnson noise?

12. Why is wireless BCI hard? Why do transmitters require so much power?

13. What is electromagnetic radiation?

14. Why does voltage decay exponentially near a neuron? Is this related to Debye shielding?

15. How does scattering work?

16. How do all the different types of microscopy work? Light beads, light sheet microscopy, confocal, 2-photon, etc. 

17. How well can we simulate a neuron if we know all its inputs? What would we need to measure to perfectly simulate a neuron? (up to noise in ion channels / other sources of “true” randomness)

18. What are all the markers of neural activity?

    • Some: voltage change, blood flow change, change in concentration of ions, slight (~10 nm) movements.

    • Another marker is gene expression changes. Immediate early genes (IEGs) get activated when neurons fire to help with things like memory formation.

19. Why are local field potentials a low pass filter?

20. What is CMOS?

21. How does nano-lithography work?

22. Why are we measuring calcium in two-photon microscopy? Why not say potassium or sodium?

    • Possibly simply because we have indicators for it

    • Also, the calcium increase during spiking is much higher than for potassium and sodium

23. Why are calcium indicators brighter than voltage indicators?

    • There is more volume for calcium indicators in the cell because they live in the cytoplasm, whereas voltage indicators live on the membrane.

      • Can you supplement the voltage indicators with amplifiers that live in the cytoplasm?

24. What are the right inductive biases for neural data? If the inductive biases for neural data are very different from text/image data, then can we expect AI progress to translate to neural data?

    • One inductive bias is viewing populations of neurons as a dynamical system. Translation invariance in the visual cortex is another.

25. Can you have cooling systems for electrodes? (might it reduce Johnson noise? Not to say Johnson noise is the main limitation of electrodes now)

26. Why is the low-pass filter temporal response a problem (e.g. hemodynamic response)? Isn’t it invertible?

    • Maybe with noise it’s harder. For example, if you reduce high frequencies by 100x and add a tiny bit of white noise, the high frequencies become just noise.

27. Are there any attempts to get way larger neural datasets? Maybe the datasets don’t need to be from explicit tasks. For example, you can get a monkey to explore its natural environment and collect both neural and camera data.

28. Can you build datasets of human neural activity that maintain the privacy of the subjects?

29. How does guidestar-based wavefront shaping work?

30. Are spikes all you need?

31. Why do so many neurotech people seem to come from electrical engineering backgrounds?

32. Why is so much BCI research focused on the motor cortex?

    • Possibly because there is fast objective supervision (e.g. I can measure joint movement)

33. People say that power consumption is the bottleneck for electrodes. But it's really heating that matters, not power consumption, right?

    • Maybe we assume that some significant fraction of power consumption goes to heating.

34. Do we have good models for going from neural activity to blood flow?

35. Is there a limit to the number of electrodes you can put in the brain?

    • Heating — electrodes themselves shouldn’t produce any Joule heating since there is no current ($I = ∆V/R = 0$). But amplifiers, analog to digital converters, and RF transmitters produce heat (why?)

36. How dangerous is gene therapy? What are ways of making proteins emit light?

37. What is impedance matching?

38. What is a transistor?

39. What is an LED?

40. What is stress and strain?

41. What is a MOSFET?

42. What are triplets?

43. What sets the resolution limit in MRI?