About Me
I mainly present comprehensive introductions and unconventional views on AI and machine learning. I’m aiming at a high signal to noise ratio to present these topics. These less conventional views emerge from working on applications involving numerics, statistics and machine learning, with some sort of (strong) underlying physical meaning, for more than 10 years.
Common issues seem to grow with more and more people working on problems who lack the understanding of physical meaning of what they are working on. It is perfectly alright to do mistakes but we shouldn’t repeat our previous mistakes and hopefully not other peoples previous mistakes. Therefore, aftermath of failed models/concepts (empirical, statistical and numerical) applying first principles is key.
For practical and especially legal reasons, I became extremely interested in automated machine learning (AutoML) to have some justification of (hyper)parameter selection. I mean this not only in a strict or narrow view AutoML but in a more general sense that covers not only the model selection and training but also enforcing mathematical justification for using certain (subsets of) datasets (distributions) over others. Moreover AutoML in a loose sense is a great way to overcome many of these common issues we can observe in practice. Another significant application of a subset of AutoML I’m interested in is neural architecture search. Not for making models bigger but mainly with respect to tinyML and deployment in real-time systems that come with hard time and resource constraints.
Unlike many others, I actively try to reduce software complexity as much as possible and I personally find it ridiculous to deploy slow software that wastes compute resources on overhead that is not needed. Even the training side of machine learning pipelines are often significantly slower than they need to whereas inference is almost always significantly slower than possible.
Many things you can find on my website are heavily influenced by:
- working on applications that require somewhat soft or hard real-time guarantees
- doing disaster recovery
- applying first principles
About This Website
I consider my website as a subset of notes I took that can be public and given the number of monthly visitors and the feedback I receive it seems to be of use for others as well.
The contents is basically three-fold:
- blog articles that cover something recent or some initial thoughts and may be merged into notes or lists after a while without notice
- lists containing collected literature references, software libraries etc.
- notes covering various things I came across (either in work or by curiosity).
NB!: Restructuring started. I started this website in a blog format. That format proved to be too inefficient and not maintainable enough as many posts might be better if merged, updated and structured differently. I’m in the process of cleaning it up, merge posts, update them and also remove posts that are outdated. The most frequently linked sites are maintained and forwarded to the new locations, others are not. Expect that some links are broken and it may take a couple more weeks/months until they re-appear. I expect the migration to be finished at the end of Q3/early Q4 in 2024. This notice will be removed once the migration is finished.
Contact Me
If you like, then you can contact me.
Feedback on this blog, papers that could interest me, etc. | [info [at] simonwenkel.com] |
Business inquiries | [info [at] cryopred.com] |