Vision

ICBINB’s vision is to create and foster a community that goes beyond benchmark climbing and new algorithms with bold numbers. In the long-term, we want to influence and be an example for other venues on the promotion of good science in machine learning; science that is collaborative and advances the understanding, the elegance, and the diversity of the field rather than just rewarding science that proves to be better than previous work on a given benchmark.

The desired outcome is to create a ML sub-community with a much more diverse set of values, highlighting the importance of unexpected negative results, questioning well-established default practices, re-evaluating and understanding existing methods; defining entirely new problems, or challenging formulations of existing problems; rewarding simplicity over unnecessary complexity.

How to get there?

ICBINB organizes activities that provide a space for beautiful ideas that advance the understanding of the ML field even when those ideas did not result in a new leaderboard on the latest benchmark. Furthermore, ICBINB takes actions to create a supportive community that can help junior scientists, or simply help ideas that are stuck, to emerge and flourish.

ICBINB organizes workshops to allow researchers to showcase their works whose characteristics fit under the values expressed by the initiative, as they would likely be undervalued in existing venues. ICBINB also organizes seminar series, where speakers will normally discuss projects that fit into the description of undervalued works, as per ICBINB’s values.

In the future, ICBINB will expand and diversify its activities, to cover more aspects of the long-term vision and further spread values to the broader ML community. This includes: local events at the organizers’ host institutions; mentoring activities, such as an “adopt  a paper” initiative; a special issue in an existing journal; an entirely brand-new journal, focused on works currently undervalued as per ICBINB’s values; a repository for dissemination and tracking of valuable negative results in machine learning, a podcast and/or blog, for researchers and ICBINB’s organizers to share their stories of works which do not fit the “bold numbers” paradigm.