Молюсѧ да не возомнитъ мѧ кто безꙋмна самъ себѧ хвалющѧ: Правдоу воистиноу глаголю и дарованна дꙋха храбрости ѿ бога данна ми не таю: Kъ томоу и конѧ зѣло быстра и добра имѣхъ.
— Kurb., Hist. Mosc., 2.
Thank you for dropping by, and welcome to my 100D space! Yes, a one-hundred-dimensional space. Yes, I’m here. Right here. Yo? Still don’t see me? C’mon, you are not farther than 10 units from me (waves his hand)! Just look a little bit to your left, and a little bit down, and a little bit… er… well, I have no words for the remaining 97 cardinal angles. When we are doing our ML things that we are doing, it’s too easy to forget how mind-bogglingly empty the higher-dimensional spaces are, and how our three-dimensional intuition fails and betrays us here. In fact, you won’t be able to hear me in a physical 100D space from half a foot away: the sound pressure would fall off with the distance \(r\) as \(r^{-99}\). But I’m glad you can read this 2D projection of my thoughts. And fascinated by this, too.
Anyway, I’ve slowly migrated from the unpronounceable unpronounceable✼I have a very strong opinion about this. I tried it myself. Kirill ‘kkm’ Katsnelson (which you can find in old sources here and there) to a more palatable✼I do not mind if you have your own opinion about this. Cy ‘kkm’ K’Nelson, but I respond to, and prefer to go by just kkm. Lowercase, please. This is also my e-mail moniker in this domain, in case you want to contact me. I’m a software engineer, an AI practitioner, linguist, physicist (a very former physicist, but still thinking like one: the Fisher information matrix in NGD is really the metric tensor of the Riemannian curved space in which we are looking for the steepest direction down. Ouch, be careful, there’s a black hole on this slope, do not step into it!), an EE, DSP junkie (wavelets, anyone?), tinkerer, and a voracious learner-by-doing. And now… am I now also a blogger??? Oops. I am.
I’ve been fascinated with the human mind, and that thing which minds use to communicate to each other that they call “language”, from my tender school years. How is it even possible that you read and understand the words that I say or write? I am not a bit closer to the answer than my 12-year-old self 1. It was no wonder that natural language understanding, both spoken and written, have been my main interests for many years, and will likely be for years to come.
I’m contributing to, and helping maintain Kaldi, co-maintain a Windows port of OpenFST, keep a Windows port of MITLM, and cannot help but boast about being a winner of the 2019 Google Open Source Peer Bonus for integrating gRPC and Protocol Buffers compilation into .NET and Visual Studio ecosystem. As a proud guild member, I respect the peer award as the highest possible recognition, and more so since my days as a Google SWE in the early 2010s. I also happen to be responsible for most of the Windows-specific and a few other moving parts of XEmacs 2, but as it stands now, this only reveals my age experience.
I started this blog to share my experience running Kaldi on Google Compute Engine with what became a suite of command-line tools, called BurrMill; these step-by-step guides are published under the category with that name. But I do not think I’ll stop at that. I’m learning new stuff every day, from every person I meet, and from every book that finds me. Come by and say hello, and I’m certain you’ll teach me something neither you nor myself could have expected!
More of kkm, if you really need more
I’m not on LinkedIn or Facebook, but here are a few pointers.
Acknowledgments
Warm hugs to Kira, who volunteered to proofread and copyedit my hasty and sloppy writing. Thanks to Aakash for his amazing (and free!) Shortcoder plugin, which allowed me to effortlessly add all the missing typesetting features of WordPress I needed.
____________________
1 Although I had to reject most of the conjectures I had back then.
2 Not to be confused with Emacs, and the ‘X’ is not about X Windows.