[2015.1.1] Happy new year.
As per plan, I am intending to dedicate the whole year to machine learning studies and polishing my programming skills, so do not expect any updates in the interim. But hopefully once my studies are done and 2016 rolls around I will be ready to output a few more arcs. For me the time will pass quickly.
Look forward to it and see you in a year.
[2015.6.23] A short update describing what I’ve been up to.
[2016.1.1] Happy new year. Here is a short update. Still doing grunt work, but I hope to transition to more interesting things in 2016. Currently, I am working on a ML tutorial and a reverse AD GPU library. I’ll link it here once I am done with the tutorials by the end of the month at most. Apart from that, I have no intention of continuing Simulacrum until I attain financial security. That will no doubt be a long war.
I’ll keep you posted every six months as usual.
[2016.1.3] Here it is. The tutorials are higher level than those basic “make a neural net using Numpy Python” that get posted all the time on the ML sub. They assume the reader knows what backprop is, but not how to code it effectively.
WordPress was not the best place to put them (the best would be to dedicate a folder and a readme.md on Github for each post,) and the tutorials are a bit like massive code dumps – more like showcases of my coding skills than actual tutorials, but I think they are good enough to count as a contribution even if for a niche audience.
Originally, I wanted to do even more with the tutorials, but will hold myself back and finally do what I wanted to in the first place now that I have the tools in place.
[2016.6.23] Here are my recollections of the past six months. In summary: Given the steady pace at which the field is progressing, I do not consider ML algorithms to be a bottleneck, but rather the problem is in the tooling. From here on out I intend to master more rewarding things such as interfaces for RL agents and will find a way to contribute to GPU language development.
If I manage to accomplish something by 2018, I’ll resume writing of Simulacrum at that time. Until then, there will be an update every six months. Cyao.
[2017.1.1] Another year flew by. In the latest update I ruminate on the meaning of programming languages and compilers. Programming is a risky activity. This year I wanted to do RL, but ended up mostly pulling apart and putting back together the Spiral library. I must have done it over 10 times already if we count all the versions.
I won’t say programming is hard; even just deciding what good programming is, is hard. I vow that the post in six months time won’t be on programming languages, but on reinforcement learning.
[2017.6.22] So much for that vow. The latest update is all about languages as I’ve been working on one since February. Until I am done with it I won’t be able to make the ML library properly so that is my focus now. I still have a while to go.
By the way, I’ve just turned on Markdown for the site for the sake of the code samples in the latest update, but can’t be bothered to check everything by eye for whether that messed anything up. If you spot some formatting strangeness, please do tell me in the comments and I will fix it.
[2018.1.1] It’s 2017 in review and the Spiral language that I worked on for so long is out. I am currently working on documenting it. There won’t be more Simulacrum this year, not until I meet financial security. I’ll fill you in on how Spiral does in action in the following update which will be in six months just as usual.
Lain’s Dream arc: