CoolFace
Apppublic

russki/repro-procmem-learning-reusable-procedural-memory-from-experience-via-non-parametric-ppo-for-llm

sourceHugging Faceupdated 2mo agoView on Hugging Face
0likes
22 commits on main
b7b7e8f2mo ago

Align conclusion with paper scope and public evidence

russki
a5a211f2mo ago

Align claim-6-ablations-show-that-skills-semantic-gradients-ppo-gating-and-score-based-maintenance-materially-affect-mastermind-v0-performance with paper scope and public evidence

russki
69f02502mo ago

Align claim-5-procmem-maintains-only-816-memory-tokens-while-achieving-a-reported-alfworld-success-rate-of-0-90 with paper scope and public evidence

russki
f9f89972mo ago

Align claim-4-procmem-achieves-higher-reuse-rates-than-baselines-in-in-domain-cross-task-and-cross-agent-evaluations with paper scope and public evidence

russki
68e07572mo ago

Align claim-3-non-parametric-ppo-uses-semantic-gradients-for-candidate-skill-generation-and-a-ppo-gate-for-robust-skill-verification with paper scope and public evidence

russki
a4f939f2mo ago

Align claim-2-the-skill-mdp-formalism-converts-passive-episodic-narratives-into-executable-skills-with-activation-execution-and-termination-conditions with paper scope and public evidence

russki
c56d62f2mo ago

Align claim-1-procmem-learns-reusable-procedural-skills-from-interaction-experience-without-updating-model-parameters with paper scope and public evidence

russki
1afcf972mo ago

Align executive-summary with paper scope and public evidence

russki
d1d77e82mo ago

Clarify reduced-scope paper mapping and publish evidence links

russki
f021e0d2mo ago

Add ICML 2026 Trackio submission metadata

russki
623b5082mo ago

Publish Trackio traces/index.json

russki
e809aba2mo ago

Publish Trackio pages/executive-summary/page.md

russki
6e1fd102mo ago

Publish Trackio pages/conclusion/page.md

russki
676257b2mo ago

Publish Trackio pages/claim-6-ablations-show-that-skills-semantic-gradients-ppo-gating-and-score-based-maintenance-materially-affect-mastermind-v0-performance/page.md

russki
cd9d4302mo ago

Publish Trackio pages/claim-5-procmem-maintains-only-816-memory-tokens-while-achieving-a-reported-alfworld-success-rate-of-0-90/page.md

russki
d5bab552mo ago

Publish Trackio pages/claim-4-procmem-achieves-higher-reuse-rates-than-baselines-in-in-domain-cross-task-and-cross-agent-evaluations/page.md

russki
c37d6082mo ago

Publish Trackio pages/claim-3-non-parametric-ppo-uses-semantic-gradients-for-candidate-skill-generation-and-a-ppo-gate-for-robust-skill-verification/page.md

russki
69e6f902mo ago

Publish Trackio pages/claim-2-the-skill-mdp-formalism-converts-passive-episodic-narratives-into-executable-skills-with-activation-execution-and-termination-conditions/page.md

russki
24a34842mo ago

Publish Trackio pages/claim-1-procmem-learns-reusable-procedural-skills-from-interaction-experience-without-updating-model-parameters/page.md

russki
58e39d22mo ago

Publish Trackio pages/index.md

russki
e527a292mo ago

Publish Trackio root assets

russki
b3623ae2mo ago

initial commit

russki