Zeke

@auonsson Overfitting is your enemy. Follow best practices is your friend. Try Python/Scikit-learn. The key is data- and feature engineering (add features - eg history). Be very careful with train/test/validation data. It’s a classification problem.

Aaah, I made it too easy for the model I realise. So, no, it did not work. Now, back to hard. Pls send cycles.

What type of modelling to do? Trying to learn about machine learning by making a model that can identify GPS-jammed ships. I have a large amount of data (GB's). I dont know if this is jam free. I have 1301 rows of confirmed jammed rows. Supervised? Unsupervised? Think again?

1/ Yesterday was a prime example of why crypto Twitter (and often crypto media) should not be trusted when it comes to crypto policy. Let's debunk claims that the EU is banning anonymous crypto transactions or self-custodial wallets. Here is what’s actually in the EU Anti Money Laundering Regulation (AMLR) and what in means for crypto in the EU 🧵

@gakonst Some projects with more info to look at: openmined.org, flower.dev, scaleoutsystems.com

@gakonst Some keywords and concepts to consider. Cross device and cross silo, model type and size, mode distribution and inference (centralized or at the nodes?), model convergence requirements, frequent model updates or not etc

what are the best resources on why collaborative ml training (whether involving trusting or distrusting parties) will work or not work? is the problem bandwidth for shared mem across gpus? adversaries who step your training the wrong direction? are all problems solved?

@weikengchen @gakonst Oh it’s already happening (we’re doing image detection in a big project across hospitals and countries), but it will take time until you have an ml only system in use.

@gakonst I think the lack of use cases is the key problem. For example, I don’t see hospitals collaborating to train a model.

We’re Hiring! 👇

I’m going to start following more what @OlympusDAO is doing. Very interesting developments lately

First 17 minutes outlines the exact problem space that OHM hopes to be a solution to. The 1:1 overlap is eery. youtube.com/watch?v=KNa-fe…

A little something I've been working on recently: The data & metrics team at @OlympusDAO wanted to work in BigQuery with data from our (many) @GraphProtocol subgraphs. I tried finding OOTB solutions for syncing data from GraphQL -> BigQuery, to no avail.

Olympus’ Range Bound Stability mechanism is finally here. Welcome to the new era of on-chain monetary policy⚡ 🔗: app.olympusdao.finance/#/range

Ohmies have just posted what I consider the most consequential proposal: OIP-123 - RBS Launch and initial parameters Contents: - Brief history of OHM - What is RBS? - Why does it matter? - How does it work? - What could go wrong? - Insights from Simulations ( @0xRusowsky 🫡🤝)

I'm claiming testnet tokens for @zksync v2, an EVM-compatible ZK Rollup! My Address: 0x1915bb60b4770216c14a208dc0901d5a7eb37594 Learn more: v2.zksync.io

Great summary of value creation with @OlympusDAO Olympus $OHM monetary policy goal is to increase stability. With demonstrated relative stability, protocols adopt OHM liquidity products. This is win-win-win (OHM, partner protocol and community).

@0xJem Full respect for the pros behind it 👀

@zeke850 Nice graph you’ve got there 😎

New OHM base rate is now set. Small but significant change for OHM liquid backing - it's now increasing.

This goldmine is now online, tying together money theory and history with Olympus. "The argument goes that demand will be greatest for currencies which are capable of keeping a stable value and maintaining liquidity" - Hayek Awesome work @Asfi3333 🔥

@Asfi3333 @EvanDeKim @OlympusDAO @Playgrounds0x @DuneAnalytics They changed the acronym, as it’s not only miners that can extract value in this value chain - and Ethereum merged.

@EvanDeKim @OlympusDAO @Playgrounds0x @DuneAnalytics Thank you Evan! I always thought MEV was Miner Extractable Value. You use the word Maximal, are they different or same?

How do atomic arbitrage MEV bots behave differently than humans using @OlympusDAO POL? On-chain swap data is collected using @Playgrounds0x and Flashbots data from @DuneAnalytics Read the full report here: mirror.xyz/evandekim.eth/… Some highlights👇

@MakadiaHarsh Bonus tip: use site:reddit.com or just append “reddit” to bypass all the spam, click bait and robo content

Checkout my newly started newsletter "Tech That Matters" Bi-weekly 4 Tech Tools, Content Creation, Web Development Tips. getrevue.co/profile/makadi…