"Choose an 'important question' - that is, one that addresses a fundamental issue in the field; these questions might or might not be 'trendy'. Note that trendy areas are inevitably (and often inappropriately) competitive, and that future trends are not always predictable." - Nancy Rothwell, Choices in neuroscience careers, Nature Reviews, 2008.
Deep Learning is becoming insanely hot nowadays due to its unreasonable effectiveness and relatively low entry barrier. But "several old, somewhat discredited machine learning techniques might be still valuable in the solution of many problems" -- Ryan Michael Rifkin. Everything old might be new again. Current machine (deep) learning research is lacking diversity, so I'm trying to inject something fresh by bringing back some old topics.
The purpose of publishing papers is to communicate with other researchers. "While brilliant and progressive research continues apace here and there, the amount of redundant, inconsequential, and outright poor research has swelled in recent decades, filling countless pages in journals and monographs ... The amount of material one must read to conduct a reasonable review of a topic keeps growing. "--We Must Stop the Avalanche of Low-Quality Research. Or, maybe we need more publications, not fewer - but more robust, instant, open, social mechanisms for sorting. I am now training my academic and technical maturity in order to contribute to real knowledge. However, I am forced to be involved in this arms race to some extent. Hope I could be a "healthy" independent researcher one day.
Nature Editorial: "If you want reproducible science, the software needs to be open source". I agree with the Reproducible Research and will share the source code related to my publications when available. I'm managing projects according this guideline.