The JumpML Blog!
Welcome to our first post! This is the beginning of a long and undoubtedly fun journey. Our goal is to help you build useful and intelligent things.
In order for a product or a thing to be useful, it needs to be easy to use and do it’s job well. Products that use machine learning, perhaps need to be the best-in-class, lest it be relegated to the useless and frustrating category. In order to be the best, one needs to adopt the best algorithms out there. It is essential to leverage the ideas in the latest state-of-the-art research presented at conferences and on arXiv. To leverage these ideas, we probably need to understand them to some extent. This will be a challenging task, as there is so much great research coming out every week. One only hopes this can be managed with some structure to the process and by developing a critical lens. At the end of the day, it just needs to work.
In our quest to build useful things, we must know how things work and their parameters and limitations. In order to organize this knowledge, we find it useful to think of articles into the following (evolving) categories
- How To: A sequence of steps on how to accomplish a given task or a very short program demonstrating some concepts. Example: How To plot the spectrogram of an audio file.
- Roundup: categorizes information from various sources and summarizes key ideas and differences. Tools, libraries, datasets, challenges, evaluation methods. Example: Speaker verification: a Roundup.
- Build: complete A-Z steps for the implementation of a useful thing (hardware, software, or both). Example: Build a keyword spotter on a Pi.
- Design: use case, architecture, structure, theory, concepts and ideas behind the implementation of a useful thing. Example: Design of a smart voice device.
- Concepts: math, modeling, theory on how something works. Example: Concepts behind MVDR beamforming.
- Review: a review of a paper, product, conference, competition, markets, hardware, open-source tools, libraries, etc. Example: BERT Review.
- Perspectives: aggregate key ideas, overarching philosophies to guide how to look at things with a big-picture context. Perspectives on machine-learned intelligence.
- Notes: these notes mostly serve the purpose to remind me of things that I tend to forget. Example: PyTorch Review.