Prizes

There is a $50,000 prize pool. Monetary prizes will be awarded to the top three teams in the leaderboard for each track according to the values below. Final ordering in the leaderboard will be determined on a held-out test set using the primary metric for each track, with secondary metrics used for tie-breaking as specified in the Tracks page. Each team that wins the first-place prize in a track will also be invited to co-author a publication summarizing the competition results and will be invited to give a short talk at the competition workshop at NeurIPS 2022 (registration provided). Special awards will also be given for top submissions satisfying certain criteria.

Eligibility

To be eligible for prizes, winning teams are required to share their methods, code, and models with the organizers as well as the names and associations of each team member. We cannot give awards to teams on US terrorist lists or those subject to sanctions.

Primary Round

Trojan Detection Track

  • 🥇 1st place: $10,000
  • 🥈 2nd place: $5,000
  • 🥉 3rd place: $2,000

Trojan Analysis Track: Target Label Prediction

  • 🥇 1st place: $5,000
  • 🥈 2nd place: $2,000
  • 🥉 3rd place: $1,000

Trojan Analysis Track: Trigger Synthesis

  • 🥇 1st place: $5,000
  • 🥈 2nd place: $2,000
  • 🥉 3rd place: $1,000

Evasive Trojans Track

  • 🥇 1st place: $5,000
  • 🥈 2nd place: $2,000
  • 🥉 3rd place: $1,000

Final Round

  • 🥇 1st place: $3,000
  • 🥈 2nd place: $1,000
  • 🥉 3rd place: $500

Awards

💾 Most data-efficient: $500

For the best method using fewer than 50 training networks*

*Awarded for the Trojan Detection Track and the two Trojan Analysis subtracks in the primary round. Requires sharing code and evidence of using less than 50 training networks with the organizers.

💸 Most resourceful: $500

For the best-performing team that was awarded compute credits to participate in the competition*

*Awarded for all tracks in the primary round. Only for teams that apply for and receive $100 in compute credits. To be eligible for this award, the submission must outperform the baseline submissions.