Hi - Liberated Techno-Humanism

i build cool things to help a good future happen a little faster i consider myself a techno-humanist based in boston (previously in vienna, berlin, and london) > contact me: mail [at] timfarkas.com ![[RE_UP_04_peak_health_lush_forest_nature_classical_hellenic_retr_ae9e22d0-ddb5-44bd-b2e0-bc864015cb33.png]] #### *My current project*: **Automatic proofreading of brain maps**. Connectomics is the field of neuroscience which tries to map all individual neurons of entire brains, and their connections. <figure style="margin: 0;"> <img src="https://cdn.prod.website-files.com/69a0ca6ef8ecfacbc1d10ea2/69db9c2115fa642cf8ff17b0_minnie-black_optimized.jpg" alt="3D brain map example" loading="lazy" decoding="async" style="max-width: 100%; height: auto; display: block;" /> <figcaption style="font-size: 0.9rem; color: #666; margin-top: 0.5rem;"> 3D rendering of mapped neurons in a small region of a mouse brain. Data source: Allen Institute </figcaption> </figure> Connectomics is interesting to me for three reasons: - I think it is one of our best shots at understanding the mind. - It could transform understanding and therapy of psychiatric/neurological diseases such as schizophrenia, depression, or Alzheimer's. - It could be the foundation for running high-fidelity brain simulations, allowing us to 'upload' minds, which might help humans better co-exist with AI. Excitingly, connectomics has recently advanced a lot, most notably with the release of a full map of all 140K neurons of the entire fruit fly brain in 2024! <figure style="margin: 0;"> <img src="all_neurons_Flywire_©MRC_LMB_UniversityofCambridge-crop-scaled.jpg" alt="3D brain map example" loading="lazy" decoding="async" style="max-width: 100%; height: auto; display: block;" /> <figcaption style="font-size: 0.9rem; color: #666; margin-top: 0.5rem;"> 3D rendering of the full fruit fly brain connectome: 140K neurons, 55M synapses, in a brain that's about the size of a grain of sand (~0.1 mm³). Data source: <cite style="font-style: normal;">FlyWire.ai</cite>; Rendering by Philipp Schlegel, University of Cambridge/MRC LMB </figcaption> </figure> These maps are derived from slices of brain tissue, that are imaged in microscopes, and then digitally stitched together. Tracing algorithms automatically trace the branches of neurons across these volumes to identify each neuron's location, shape, and neighbors. But, due to the sheer number, size, and complexity of these neurons, these tracing algorithms — while already extremely precise — still make lots of errors at scale. Which currently need to be identified and fixed by hand! This took 33 person-years for the 140k neurons of the fly brain, and at this rate it would take thousands of years for the 80B neurons of the human brain! <figure style="margin: 0;"> <img src="Pasted image 20260602185640.png" alt="3D brain map example" loading="lazy" decoding="async" style="max-width: 100%; height: auto; display: block;" /> <figcaption style="font-size: 0.9rem; color: #666; margin-top: 0.5rem;"> Errors (False Splits and False Merges) that occur during automatic tracing, which currently need to be spotted and fixed by hand (Merge & Split Corrections). </figcaption> </figure> This is the problem that I'm working on fixing right now at the Boyden Lab: **We're training computer vision models to make these proofreading corrections automatically.** Automating proofreading would be a great step toward mouse- and human-size connectomes, as proofreading costs are the dominant cost factor right now. I'd love to talk about this work if you're working on something similar or interested in learning more! Reach out at: mail [at] timfarkas.com.