Moxie Marlinspike, creator of Signal, recently wrote a blog post entitled My first impressions of web3 where he critiques web3’s growing centralization tendencies. His stint into the blockchain world is surely a very short one, and he mostly focuses on his recent NFT related experiments, but one month for an engineer of his caliber is worth a year’s time for us mere mortals. His cogent, powerful, and thought-provoking piece has certainly impacted my thinking greatly, and I cannot but recommend taking the time to read it.

Agustín Carstens, head of the Bank for International Settlements (BIS), said in a 2018 interview: “My message to young people: stop trying to create money.” It remains unclear to me if he had specific projects in mind at the time, or whether he even understands the distinctions between BTC, ETH, USDC, DAI, COMP, UNI, and the rest of the DeFi tokens zoo. Nonetheless, one can quite easily guess why he was raising alarms after browsing CoinGecko for a short while:

Statistical hypothesis testing can be very confusing at first. From its inception, Fisher disagreed with Neyman and Pearson on its use and role in science. Neyman and Pearson argued that they were bringing mathematical rigor to Fisher’s approach, but Fisher considered their framework incompatible with any real day-to-day scientific practice. Unfortunately, no method was ever declared a clear winner, and often hypothesis testing is taught as a hybrid of the two approaches that resulted from confusion by writers of statistical textbooks (as predicted by Fisher) beginning in the 1940s. Furthermore, in the current era where data trumps and nonparametric methods like deep learning are king, many people only get a brief introduction to formal statistics in a data science or machine learning course. It doesn’t help that the professors often don’t need or use hypothesis testing in their daily research any more than their students will in their future employment, and so the subject is quietly brushed under the carpet and quickly forgotten.

For a continuous-time dynamical system $\dot{x} = F(x)$ on $\mathbb{R}^n$ with flow map $F_t$, the discrete-time Koopman operators are defined as $K_t g = g \circ F_t$, where $g: \mathbb{R}^n \rightarrow \mathbb{R}$. The continuous-time Koopman operator is then defined as $ \lim_{t \rightarrow 0} \frac{K_t g}{t}$.

I’m moving back to jekyll (although reluctantly) from google sites. Although google sites is extremely simple and fun to use, its lack of support for latex and mathjax makes it cumbersome when it comes to technical writing.

Emacs に慣れた人は僕のように結局EメールもEmacsで読みたくなるかもしれま せんからこのブログ記事を書こうと思った。