Participation

Who can participate

The challenge is open to students and working professionals alike, worldwide. No affiliation with an academic institution is required.

Employees of STMicroelectronics, and anyone who has had access to the unreleased test set or to its collection process, are not eligible to participate.

Team composition

• Team size. There is no upper limit on team members, but every member is expected to contribute actively and to be listed as an author on the submitted paper.

• Multiple teams. An individual may belong to only one team. Supervisors and advisors may be associated with more than one team, provided they respect the sharing rules outlined in the next block.

• Multiple teams from one laboratory. Permitted, but each team must operate independently. Two teams from the same lab sharing code or intermediate results are a single team, and will be treated as such.

• Team mergers. If two registered teams wish to combine, contact the organisers by email before the submission deadline. Mergers after submission are not possible.

Changes to a team. Notify the organisers by email if members join or leave after registration.

Code and data sharing

• Private sharing outside your team is prohibited. Do not share code, trained weights, curated datasets, or intermediate results with other teams before the results are announced.

• Public resources are fair game. Open-source libraries, published architectures, and publicly available pretrained models may all be used, provided they are properly cited in your paper. Using a published method well is not a weakness; failing to credit it is.

• BOXRR-23 terms apply. Comply with the licence and access conditions published by UC Berkeley, do not attempt to re-identify users, and do not redistribute raw or derived data outside your team.

• After the challenge, you are encouraged — but not required — to release your code publicly alongside your paper.

Paper requirements

Either 4-6 pages in IEEE format or 8-10 pages IJABC format scientific paper describing your work.

Expected contents:

• Abstract — a concise summary of your approach and results.

• Method — your data pipeline, architecture, training procedure, and any compression or optimisation techniques applied.

• Results and discussion — your own evaluation figures, an honest account of strengths and limitations, and a comparison against at least a constant-velocity baseline.

• Conclusion — findings, implications, and future work.

• References — including the BOXRR-23 dataset citation, which is mandatory.

Report your numbers. Your paper must state your model's total trainable parameter count and its arithmetic cost per streaming step. These figures are cross-checked against the delivered model, and they feed two of the four evaluation criteria.

Presentation. At least one team member must present the paper at the conference. Accepted papers will be published with the ABC 2027 proceedings.

For conference registration details, see the main ABC 2027 website.