Kenta Iwasaki

I have left my role as CTO at @PerlinNetwork. I've tinkered with distributed systems, cryptography, Go/Rust, and kickstarted 3 large open-source projects during my time there. Moving on and looking for the next big thing to work on now - if you've got anything in mind, HMU 🙏.

Superpower or Function? @superpower does 3x more for half the price. Function = 100+ labs then nothing Superpower = 100+ labs then so much more: + 24/7 medical team + Upload past lab tests + Personalized protocol + Health autopilot offering + Hormones, GLP-1s and peptides + Gut testing, toxin testing and more + 20% off future tests, supplements, RX You deserve to make informed decisions.

Taking down this Twitter and moving to @lithdew over time! Please do follow me if you'd like to keep updated :).

Thinking about open-sourcing a few of my hackathon tools for making production-ready backends w/ analytics, monitoring, logging, error handling, etc. in 10 mins. #golang #starlark

@mynameisbouds @PerlinNetwork Unfortunately, funding was pulled away for my work. I've got plans to continue developing Wavelet on my own resources outside of Perlin, though that is the most I can do. I still want to see the tech I envisioned Perlin to be around to see the light of day.

I have left my role as CTO at @PerlinNetwork. I've tinkered with distributed systems, cryptography, Go/Rust, and kickstarted 3 large open-source projects during my time there. Moving on and looking for the next big thing to work on now - if you've got anything in mind, HMU 🙏.

@theprincessxena xachi (?)

@waybeforenow let's collab bro

i put some old stuff on bandcamp today

i will put some old stuff on bandcamp soon

@CollinCusce @JoshuaTobkin @thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @avalabsofficial @PerlinNetwork 🤷‍♀️

@JoshuaTobkin @thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial @PerlinNetwork Last I worked on consensus I was exploring a few numerical opt. methods actually from work I used to do at Naver in AI - unfortunately don't have the time to and am currently working on perf opt. and enterprise stuff rn 😢.

@JoshuaTobkin @thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial @PerlinNetwork 10 nodes/sample w/ 50 samples at a minimum. # rounds quadratically increase w.r.t. # Byzantine; double spending isn't possible (although forking is). I must admit to not going more in-depth into preventing forks just yet: all existing forks are simply recorded for now.

@JoshuaTobkin @thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial @PerlinNetwork Oh, block time (latency/tx) is roughly 0.6 seconds. With induced latency via tc(1), it is roughly 4 seconds.

@JoshuaTobkin @thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial @PerlinNetwork Avg. 5,438 non-batched and 31,240 batched (40 tx/batch) over 240 nodes w/ 220ms latency via tc(1). Disabling signatures, 18,438 non-batched and 543,182 batched. Storage is syndtr/goleveldb, persistent AVL+ Merkle tree. 4GB RAM 2vCPUs DigitalOcean sgp1 region.

@JoshuaTobkin @thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial @PerlinNetwork The idea of statistical sampling to establish consensus is the main part we adopted. The differences we made on our end are 1) weighing votes per sample, and 2) keeping historical votes and deriving a confidence score based on these votes.

@xtwokei @thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial @PerlinNetwork Thanks for sharing, did not know about your recent design choices. Snowball doesn’t scale well, but you’ve adopted parts of it? Mind sharing which parts? Anything else you’d like to point out regarding differences?

@Jonatha20289121 I gotta make my project names more cryptic 😂

@xtwokei dkv: decentralized key-value Are you building a decentralized levelsDB?

Almost hit my targets for this weekend,

@CollinCusce @JoshuaTobkin @thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @avalabsofficial @PerlinNetwork :c

@thebtcpodcast @HashingItOutPod @AlexSkidanov @JoshuaTobkin @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial @PerlinNetwork Happy to give a perspective - I agree mostly with the points @AlexSkidanov raised though. Statistical sampling to achieve consensus is a very interesting direction however which we're spearheading at @PerlinNetwork.

@HashingItOutPod @AlexSkidanov @JoshuaTobkin @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial might be interesting with perspective from @PerlinNetwork too

@JoshuaTobkin @thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial @PerlinNetwork We stopped using Avalanche about a year ago now. There were a lot of bad choices regarding how they maintained a replicated log using a DAG. The only portion of Avalanche we've adapted is Snowball, which by itself does not scale within across a global network.

@thebtcpodcast @HashingItOutPod @AlexSkidanov @kevinsekniqi @NEARProtocol @el33th4xor @CollinCusce @avalabsofficial @PerlinNetwork I think their protocol is based on Avalanche so maybe better to go directly to the originators

Everybody, time to freak out! Nesterov does it again, getting O(1/k^4) rate with a second-order method! How refreshing and cool with all that ML craze, hail high-order methods! Main idea, use a third-order method and solve it using second-order information dial.uclouvain.be/pr/boreal/obje…