The Shifting Landscape of ML Conferences

A palpable sense of nostalgia and concern is emerging within the machine learning community regarding the state of academic conferences. Once vibrant hubs for specialized research, conferences like BMVC (British Machine Vision Conference), ACCV (Asian Conference on Computer Vision), FG (International Conference on Automatic Face and Gesture Recognition), ICIP (International Conference on Image Processing), and ICASSP (International Conference on Acoustics, Speech, and Signal Processing) are perceived by some as having lost their former prominence. The sentiment, shared across online forums and discussions, suggests a move away from focused, community-driven events towards a more consolidated, flagship-centric model.

This shift is particularly felt by researchers who previously relied on these specialized venues for cutting-edge work in their niche fields. FG, for instance, was once the premier destination for face analysis research, while ICASSP served a similar role for signal processing. The perceived decline of these specialized conferences raises questions about the fate of research that doesn't fit into the broader themes of the now-dominant mega-conferences. The sheer volume of submissions to flagship events like NeurIPS, ICML, and ICLR, coupled with limited acceptance rates and what some describe as inconsistent peer review, creates a bottleneck. This bottleneck may force valuable research into non-archival tracks, relegated to arXiv pre-prints without formal peer review, or simply never seeing the light of day.

The core of the concern lies in the fragmentation of specialized communities. The old ecosystem fostered deep connections and focused discussions within specific sub-disciplines. Now, researchers worry that the broader, more generalized conferences dilute these specialized interactions. The challenge for a researcher today is not just about getting a paper accepted, but about ensuring it reaches the right audience and sparks meaningful engagement within its intended community. The current system, with its overwhelming scale and focus on general ML progress, might be inadvertently hindering the growth and visibility of niche research areas.

The Consolidation Effect and Its Consequences

The trend towards consolidation is not unique to machine learning. Many academic fields grapple with the economics and logistics of large-scale conferences. However, in a rapidly evolving field like ML, the consequences of this consolidation can be more pronounced. As fewer conferences become the gatekeepers of high-impact publications, the pressure intensifies. This creates an environment where researchers might tailor their work to fit the perceived themes of these mega-conferences, potentially at the expense of pursuing truly novel, albeit niche, research directions.

The idea of a paper being a "non-archival submission" is particularly galling. While arXiv has become an indispensable tool for rapid dissemination, it lacks the rigorous peer-review process that traditional conferences provide. When a paper is presented as a non-archival submission at a major conference, it signals a compromise – a paper deemed good enough to present but not strong enough for formal archival publication. This can impact a researcher's citation count, career progression, and the perceived impact of their work. The question then becomes: is this a sign of a maturing field, or a symptom of an ecosystem struggling to scale effectively?

Furthermore, the diminishing role of specialized conferences affects the mentorship and training of early-career researchers. Smaller, focused workshops and conferences often provide a more intimate setting for graduate students and postdocs to present their work, receive detailed feedback, and build a network within their specific subfield. The loss of these opportunities could hinder the development of future specialists. It's akin to a city losing its specialized artisan districts; while a large central market might offer more variety, it doesn't cultivate the same depth of craft or community as the old, dedicated neighborhoods.

The concern isn't merely about missing the