From 5ba44205c9dd5783ead556d94a8c02b2c1a4d358 Mon Sep 17 00:00:00 2001 From: Edwin Chen Date: Fri, 13 Feb 2026 06:00:17 +0000 Subject: [PATCH 1/4] FCSxIIMOC blog --- _posts/2026-02-12-FCS-IIMOC-collab.md | 139 ++++++++++++++++++++++++++ 1 file changed, 139 insertions(+) create mode 100644 _posts/2026-02-12-FCS-IIMOC-collab.md diff --git a/_posts/2026-02-12-FCS-IIMOC-collab.md b/_posts/2026-02-12-FCS-IIMOC-collab.md new file mode 100644 index 0000000..683b6de --- /dev/null +++ b/_posts/2026-02-12-FCS-IIMOC-collab.md @@ -0,0 +1,139 @@ +--- +layout: distill +title: "Frontier-CS × IIMOC Collaboration" +description: "Frontier-CS and the International Invitational Math Optimization Challenge (IIMOC) announce a formal collaboration to study open-ended algorithmic problem solving, solution diversity, and long-horizon improvement across humans and AI systems." + +date: 2026-02-12 +date_display: "Feb 12, 2026" +htmlwidgets: true +toc: true + +authors: + - name: "Edwin Chen" + url: "https://github.com/echen5503" + affiliations: + name: "Independent" + - name: "Alvin Cheung" + url: "https://people.eecs.berkeley.edu/~akcheung/" + affiliations: + name: "UC Berkeley" + url: "https://www.berkeley.edu/" + - name: Qiuyang Mang + url: "https://joyemang33.github.io" + affiliations: + name: UC Berkeley + - name: Hanchen Li + url: "https://hanchenli.github.io/" + affiliations: + name: UC Berkeley + - name: "Frontier-CS Team" + url: "https://frontier-cs.org" + - name: "IIMOC Team" + url: "https://iimoc.org" + +_styles: > + d-article img { + max-width: 100%; + height: auto; + display: block; + margin: 1.5rem auto; + border-radius: 6px; + } + d-article p img { + display: block; + margin-left: auto; + margin-right: auto; + } +--- +![](https://raw.githubusercontent.com/FrontierCS/Frontier-CS/main/assets/logo.png) + +We are pleased to announce a formal collaboration between **[Frontier-CS](https://github.com/FrontierCS/Frontier-CS)** and the **[International Invitational Math Optimization Challenge (IIMOC)](https://iimoc.org)**. + +Frontier-CS is an open research benchmark designed to evaluate high-difficulty algorithmic and optimization problems on a **continuous 0–100 scale**, rather than through binary correctness. This distinction is central: many meaningful optimization tasks do not admit a single canonical “correct” answer, but instead reward incremental refinement, structural insight, and strategic iteration. + +By contrast to classical contest-style problems that collapse to a single solution, Frontier-CS emphasizes problems where progress is measurable and multi-directional. Improvements can be partial, iterative, and comparative — allowing performance to be modeled as a trajectory rather than a point outcome. + +--- + +## Why This Collaboration Matters + +Open-ended benchmarks are strongest when evaluated across diverse environments. Competitive settings introduce dynamics that static benchmarks cannot capture: time constraints, adversarial iteration, exploration under uncertainty, and heterogeneous strategy formation. + +The International Invitational Math Optimization Challenge (IIMOC) provides a large-scale, human-driven optimization setting that naturally complements model-based evaluation. Integrating Frontier-CS problems into IIMOC allows us to observe how humans: + +* allocate effort under time pressure +* balance exploration versus exploitation +* iterate across many submissions +* converge (or fail to converge) toward high-performing strategies + +This pairing enables a rare alignment between controlled research benchmarking and real competitive dynamics. + +--- + +## The 2025 Pilot Run + +In 2025, a pilot integration of Frontier-CS problems into IIMOC attracted: + +* **209 teams** +* **9,549 total submissions** + +Rather than rapidly converging to a dominant solution, teams exhibited sustained improvement behavior over time. Multiple solution paradigms emerged, with distinct structural trade-offs and performance characteristics. + +Empirically, the pilot demonstrated that Frontier-CS problems: + +* support long-horizon improvement rather than short-term exploitation +* admit multiple competitive approaches +* resist collapse to a single dominant heuristic +* produce measurable improvement curves across repeated submissions + +This evidence strengthens Frontier-CS as a benchmark for studying *optimization dynamics*, rather than static correctness or one-shot performance. + +--- + +## Research Directions Enabled by the Collaboration + +Building on the pilot, Frontier-CS and IIMOC are formalizing this partnership as an evaluation and data-collection framework. + +The collaboration enables systematic study of: + +* **Solution multiplicity** in open-ended algorithmic search +* **Human improvement curves** under iterative submission models +* **Comparative improvement dynamics** between humans and AI systems +* **Strategy diversity and convergence behavior** over time + +Because performance is scored on a continuous scale, we can model: + +* fine-grained optimization trajectories +* plateau and breakthrough phenomena +* exploration–exploitation trade-offs +* variance across teams and across problem classes + +This transforms competitive programming from a binary success metric into a longitudinal study of adaptive search. + +--- + +## Toward a Long-Term Pipeline + +Frontier-CS will continue expanding its library of high-difficulty algorithmic tasks. Select problems may be incorporated into future IIMOC competitions, forming a sustained pipeline between: + +* an **open research benchmark infrastructure**, and +* a **competitive, time-bounded evaluation ecosystem** + +This structure supports: + +* controlled experimentation +* large-scale human behavioral data +* cross-comparison with frontier AI systems +* longitudinal benchmarking across benchmark releases + +By integrating open benchmarking with competitive optimization dynamics, Frontier-CS × IIMOC aims to advance the study of measurable progress in algorithmic discovery — across both human and machine systems. + +We look forward to deepening this collaboration and continuing to develop Frontier-CS as a benchmark that meaningfully captures evolving intelligence across domains. + +--- + +If you'd like, I can also: + +* Make this slightly more formal and ICML/NeurIPS-blog style +* Tighten it further into a sharper announcement tone +* Or add a short technical paragraph on scoring mechanics and evaluation reproducibility From fe22e3e9686fe2be318078bd215e1544d8f9d631 Mon Sep 17 00:00:00 2001 From: Edwin Chen Date: Fri, 13 Feb 2026 06:01:22 +0000 Subject: [PATCH 2/4] filename change to lower case --- ...6-02-12-FCS-IIMOC-collab.md => 2026-02-12-fcs-iimoc-collab.md} | 0 1 file changed, 0 insertions(+), 0 deletions(-) rename _posts/{2026-02-12-FCS-IIMOC-collab.md => 2026-02-12-fcs-iimoc-collab.md} (100%) diff --git a/_posts/2026-02-12-FCS-IIMOC-collab.md b/_posts/2026-02-12-fcs-iimoc-collab.md similarity index 100% rename from _posts/2026-02-12-FCS-IIMOC-collab.md rename to _posts/2026-02-12-fcs-iimoc-collab.md From da1afbdc72f5288734ab8355f0eb8511b1bf162d Mon Sep 17 00:00:00 2001 From: Edwin Chen Date: Fri, 13 Feb 2026 06:00:17 +0000 Subject: [PATCH 3/4] FCSxIIMOC blog --- _posts/2026-02-12-FCS-IIMOC-collab.md | 139 ++++++++++++++++++++++++++ 1 file changed, 139 insertions(+) create mode 100644 _posts/2026-02-12-FCS-IIMOC-collab.md diff --git a/_posts/2026-02-12-FCS-IIMOC-collab.md b/_posts/2026-02-12-FCS-IIMOC-collab.md new file mode 100644 index 0000000..683b6de --- /dev/null +++ b/_posts/2026-02-12-FCS-IIMOC-collab.md @@ -0,0 +1,139 @@ +--- +layout: distill +title: "Frontier-CS × IIMOC Collaboration" +description: "Frontier-CS and the International Invitational Math Optimization Challenge (IIMOC) announce a formal collaboration to study open-ended algorithmic problem solving, solution diversity, and long-horizon improvement across humans and AI systems." + +date: 2026-02-12 +date_display: "Feb 12, 2026" +htmlwidgets: true +toc: true + +authors: + - name: "Edwin Chen" + url: "https://github.com/echen5503" + affiliations: + name: "Independent" + - name: "Alvin Cheung" + url: "https://people.eecs.berkeley.edu/~akcheung/" + affiliations: + name: "UC Berkeley" + url: "https://www.berkeley.edu/" + - name: Qiuyang Mang + url: "https://joyemang33.github.io" + affiliations: + name: UC Berkeley + - name: Hanchen Li + url: "https://hanchenli.github.io/" + affiliations: + name: UC Berkeley + - name: "Frontier-CS Team" + url: "https://frontier-cs.org" + - name: "IIMOC Team" + url: "https://iimoc.org" + +_styles: > + d-article img { + max-width: 100%; + height: auto; + display: block; + margin: 1.5rem auto; + border-radius: 6px; + } + d-article p img { + display: block; + margin-left: auto; + margin-right: auto; + } +--- +![](https://raw.githubusercontent.com/FrontierCS/Frontier-CS/main/assets/logo.png) + +We are pleased to announce a formal collaboration between **[Frontier-CS](https://github.com/FrontierCS/Frontier-CS)** and the **[International Invitational Math Optimization Challenge (IIMOC)](https://iimoc.org)**. + +Frontier-CS is an open research benchmark designed to evaluate high-difficulty algorithmic and optimization problems on a **continuous 0–100 scale**, rather than through binary correctness. This distinction is central: many meaningful optimization tasks do not admit a single canonical “correct” answer, but instead reward incremental refinement, structural insight, and strategic iteration. + +By contrast to classical contest-style problems that collapse to a single solution, Frontier-CS emphasizes problems where progress is measurable and multi-directional. Improvements can be partial, iterative, and comparative — allowing performance to be modeled as a trajectory rather than a point outcome. + +--- + +## Why This Collaboration Matters + +Open-ended benchmarks are strongest when evaluated across diverse environments. Competitive settings introduce dynamics that static benchmarks cannot capture: time constraints, adversarial iteration, exploration under uncertainty, and heterogeneous strategy formation. + +The International Invitational Math Optimization Challenge (IIMOC) provides a large-scale, human-driven optimization setting that naturally complements model-based evaluation. Integrating Frontier-CS problems into IIMOC allows us to observe how humans: + +* allocate effort under time pressure +* balance exploration versus exploitation +* iterate across many submissions +* converge (or fail to converge) toward high-performing strategies + +This pairing enables a rare alignment between controlled research benchmarking and real competitive dynamics. + +--- + +## The 2025 Pilot Run + +In 2025, a pilot integration of Frontier-CS problems into IIMOC attracted: + +* **209 teams** +* **9,549 total submissions** + +Rather than rapidly converging to a dominant solution, teams exhibited sustained improvement behavior over time. Multiple solution paradigms emerged, with distinct structural trade-offs and performance characteristics. + +Empirically, the pilot demonstrated that Frontier-CS problems: + +* support long-horizon improvement rather than short-term exploitation +* admit multiple competitive approaches +* resist collapse to a single dominant heuristic +* produce measurable improvement curves across repeated submissions + +This evidence strengthens Frontier-CS as a benchmark for studying *optimization dynamics*, rather than static correctness or one-shot performance. + +--- + +## Research Directions Enabled by the Collaboration + +Building on the pilot, Frontier-CS and IIMOC are formalizing this partnership as an evaluation and data-collection framework. + +The collaboration enables systematic study of: + +* **Solution multiplicity** in open-ended algorithmic search +* **Human improvement curves** under iterative submission models +* **Comparative improvement dynamics** between humans and AI systems +* **Strategy diversity and convergence behavior** over time + +Because performance is scored on a continuous scale, we can model: + +* fine-grained optimization trajectories +* plateau and breakthrough phenomena +* exploration–exploitation trade-offs +* variance across teams and across problem classes + +This transforms competitive programming from a binary success metric into a longitudinal study of adaptive search. + +--- + +## Toward a Long-Term Pipeline + +Frontier-CS will continue expanding its library of high-difficulty algorithmic tasks. Select problems may be incorporated into future IIMOC competitions, forming a sustained pipeline between: + +* an **open research benchmark infrastructure**, and +* a **competitive, time-bounded evaluation ecosystem** + +This structure supports: + +* controlled experimentation +* large-scale human behavioral data +* cross-comparison with frontier AI systems +* longitudinal benchmarking across benchmark releases + +By integrating open benchmarking with competitive optimization dynamics, Frontier-CS × IIMOC aims to advance the study of measurable progress in algorithmic discovery — across both human and machine systems. + +We look forward to deepening this collaboration and continuing to develop Frontier-CS as a benchmark that meaningfully captures evolving intelligence across domains. + +--- + +If you'd like, I can also: + +* Make this slightly more formal and ICML/NeurIPS-blog style +* Tighten it further into a sharper announcement tone +* Or add a short technical paragraph on scoring mechanics and evaluation reproducibility From b73d850a521bf775246b359977276837a0f92a19 Mon Sep 17 00:00:00 2001 From: Edwin Chen Date: Fri, 13 Feb 2026 06:03:09 +0000 Subject: [PATCH 4/4] file name change --- _posts/2026-02-12-FCS-IIMOC-collab.md | 139 -------------------------- _posts/2026-02-12-fcs-iimoc-collab.md | 7 -- 2 files changed, 146 deletions(-) delete mode 100644 _posts/2026-02-12-FCS-IIMOC-collab.md diff --git a/_posts/2026-02-12-FCS-IIMOC-collab.md b/_posts/2026-02-12-FCS-IIMOC-collab.md deleted file mode 100644 index 683b6de..0000000 --- a/_posts/2026-02-12-FCS-IIMOC-collab.md +++ /dev/null @@ -1,139 +0,0 @@ ---- -layout: distill -title: "Frontier-CS × IIMOC Collaboration" -description: "Frontier-CS and the International Invitational Math Optimization Challenge (IIMOC) announce a formal collaboration to study open-ended algorithmic problem solving, solution diversity, and long-horizon improvement across humans and AI systems." - -date: 2026-02-12 -date_display: "Feb 12, 2026" -htmlwidgets: true -toc: true - -authors: - - name: "Edwin Chen" - url: "https://github.com/echen5503" - affiliations: - name: "Independent" - - name: "Alvin Cheung" - url: "https://people.eecs.berkeley.edu/~akcheung/" - affiliations: - name: "UC Berkeley" - url: "https://www.berkeley.edu/" - - name: Qiuyang Mang - url: "https://joyemang33.github.io" - affiliations: - name: UC Berkeley - - name: Hanchen Li - url: "https://hanchenli.github.io/" - affiliations: - name: UC Berkeley - - name: "Frontier-CS Team" - url: "https://frontier-cs.org" - - name: "IIMOC Team" - url: "https://iimoc.org" - -_styles: > - d-article img { - max-width: 100%; - height: auto; - display: block; - margin: 1.5rem auto; - border-radius: 6px; - } - d-article p img { - display: block; - margin-left: auto; - margin-right: auto; - } ---- -![](https://raw.githubusercontent.com/FrontierCS/Frontier-CS/main/assets/logo.png) - -We are pleased to announce a formal collaboration between **[Frontier-CS](https://github.com/FrontierCS/Frontier-CS)** and the **[International Invitational Math Optimization Challenge (IIMOC)](https://iimoc.org)**. - -Frontier-CS is an open research benchmark designed to evaluate high-difficulty algorithmic and optimization problems on a **continuous 0–100 scale**, rather than through binary correctness. This distinction is central: many meaningful optimization tasks do not admit a single canonical “correct” answer, but instead reward incremental refinement, structural insight, and strategic iteration. - -By contrast to classical contest-style problems that collapse to a single solution, Frontier-CS emphasizes problems where progress is measurable and multi-directional. Improvements can be partial, iterative, and comparative — allowing performance to be modeled as a trajectory rather than a point outcome. - ---- - -## Why This Collaboration Matters - -Open-ended benchmarks are strongest when evaluated across diverse environments. Competitive settings introduce dynamics that static benchmarks cannot capture: time constraints, adversarial iteration, exploration under uncertainty, and heterogeneous strategy formation. - -The International Invitational Math Optimization Challenge (IIMOC) provides a large-scale, human-driven optimization setting that naturally complements model-based evaluation. Integrating Frontier-CS problems into IIMOC allows us to observe how humans: - -* allocate effort under time pressure -* balance exploration versus exploitation -* iterate across many submissions -* converge (or fail to converge) toward high-performing strategies - -This pairing enables a rare alignment between controlled research benchmarking and real competitive dynamics. - ---- - -## The 2025 Pilot Run - -In 2025, a pilot integration of Frontier-CS problems into IIMOC attracted: - -* **209 teams** -* **9,549 total submissions** - -Rather than rapidly converging to a dominant solution, teams exhibited sustained improvement behavior over time. Multiple solution paradigms emerged, with distinct structural trade-offs and performance characteristics. - -Empirically, the pilot demonstrated that Frontier-CS problems: - -* support long-horizon improvement rather than short-term exploitation -* admit multiple competitive approaches -* resist collapse to a single dominant heuristic -* produce measurable improvement curves across repeated submissions - -This evidence strengthens Frontier-CS as a benchmark for studying *optimization dynamics*, rather than static correctness or one-shot performance. - ---- - -## Research Directions Enabled by the Collaboration - -Building on the pilot, Frontier-CS and IIMOC are formalizing this partnership as an evaluation and data-collection framework. - -The collaboration enables systematic study of: - -* **Solution multiplicity** in open-ended algorithmic search -* **Human improvement curves** under iterative submission models -* **Comparative improvement dynamics** between humans and AI systems -* **Strategy diversity and convergence behavior** over time - -Because performance is scored on a continuous scale, we can model: - -* fine-grained optimization trajectories -* plateau and breakthrough phenomena -* exploration–exploitation trade-offs -* variance across teams and across problem classes - -This transforms competitive programming from a binary success metric into a longitudinal study of adaptive search. - ---- - -## Toward a Long-Term Pipeline - -Frontier-CS will continue expanding its library of high-difficulty algorithmic tasks. Select problems may be incorporated into future IIMOC competitions, forming a sustained pipeline between: - -* an **open research benchmark infrastructure**, and -* a **competitive, time-bounded evaluation ecosystem** - -This structure supports: - -* controlled experimentation -* large-scale human behavioral data -* cross-comparison with frontier AI systems -* longitudinal benchmarking across benchmark releases - -By integrating open benchmarking with competitive optimization dynamics, Frontier-CS × IIMOC aims to advance the study of measurable progress in algorithmic discovery — across both human and machine systems. - -We look forward to deepening this collaboration and continuing to develop Frontier-CS as a benchmark that meaningfully captures evolving intelligence across domains. - ---- - -If you'd like, I can also: - -* Make this slightly more formal and ICML/NeurIPS-blog style -* Tighten it further into a sharper announcement tone -* Or add a short technical paragraph on scoring mechanics and evaluation reproducibility diff --git a/_posts/2026-02-12-fcs-iimoc-collab.md b/_posts/2026-02-12-fcs-iimoc-collab.md index 683b6de..acd7545 100644 --- a/_posts/2026-02-12-fcs-iimoc-collab.md +++ b/_posts/2026-02-12-fcs-iimoc-collab.md @@ -130,10 +130,3 @@ By integrating open benchmarking with competitive optimization dynamics, Frontie We look forward to deepening this collaboration and continuing to develop Frontier-CS as a benchmark that meaningfully captures evolving intelligence across domains. ---- - -If you'd like, I can also: - -* Make this slightly more formal and ICML/NeurIPS-blog style -* Tighten it further into a sharper announcement tone -* Or add a short technical paragraph on scoring mechanics and evaluation reproducibility