The question feels familiar to anyone who has watched the research world tighten its orbit around a few big names. When you see a community like the one at BMVC or ICASSP start to thin out, it is natural to wonder if the work is still out there, just scattered. The user's sense of loss is not misplaced nostalgia. It is an observation about how the incentives for sharing research have shifted, and how the pressure to land a spot in a flagship event can push valuable work into quieter corners, like an arXiv preprint that never gets the discussion it deserves. This is a practical concern, not just an emotional one. It affects what gets seen, what gets cited, and ultimately what kind of feedback a researcher receives when they are still shaping their ideas.
That concentration has a real cost. When a field like face analysis or signal processing loses its dedicated home, the focused conversations that used to happen there fragment. A researcher might now find their paper sitting in a non-archival limbo, or worse, they might decide the review lottery is not worth the effort and just post it online. The result is that the community loses the connective tissue that made those smaller venues feel like a home base for a specific question. We have seen this pattern in other areas of applied machine learning, where the pressure to optimize for a single metric or a single leaderboard can push out the exploratory work. It is worth asking whether the current system is serving the people doing the work, or just the institutions that count the acceptance rates. If you are feeling the squeeze in your own workflow, you are not alone. The same friction that makes you hesitate before submitting to a major conference is likely why you are also thinking about how to clean data before it skews your model or how to deploy models on real-world edge devices. The problems are connected.
The practical takeaway here is not to mourn the old days, but to be more deliberate about where you share your work and how you evaluate it. Instead of waiting for a single big acceptance, consider what a smaller venue offers: faster feedback, a more engaged audience, and a community that actually reads your paper. If you are a researcher, that might mean treating arXiv as a primary publication point rather than a fallback. If you are a reader, it means seeking out those non-archival papers and citing them, because that is what keeps the ecosystem alive. The question we should be asking is not whether the old conference model was better, but whether we can build a new one that values the work itself over the venue it lands in. That might start with something as simple as judging a paper on its own merits, not the logo at the top of the page. And if you are curious about how to navigate these choices, exploring the role of functions and evaluation in machine learning, as discussed in Forrester function analysis, might give you a different lens on how to measure progress beyond a single score. The real test is whether we can keep the conversation going when the conference calendar is no longer the only stage.