diff --git a/README.md b/README.md index 8c0b3e6933c4e51673dedcb3cd0a9d78d7c065fc..ac659f4f2fa44da237589abe596956219662ccaf 100644 --- a/README.md +++ b/README.md @@ -1,6 +1,6 @@ # AWARE -AWARE is an extensible framework for deploying and managing RL(reinforcement learning)-based agents in production systems. +AWARE is an extensible framework for deploying and managing RL (reinforcement learning)-based agents in production systems. AWARE provides (1) fast adaptation with meta-learning, (2) reliable RL exploration with bootstrapping, (3) timely retraining by continuous monitoring. ## Requirements @@ -46,7 +46,6 @@ aware/ ``` ## Testing -### Functionality Start training with meta-learner, with model checkpoints saved to `testing/checkpoints/`. @@ -79,6 +78,21 @@ The number of episodes is set to 500 for demonstration purpose. For running AWARE (training and policy-serving) with application `deployment` managed by RL-based MPA in Kubernetes cluster, follow `multidimensional-pod-autoscaler/README.md` to deploy MPA and `rl-controller/README.md` for training and customizing for application deployments from scratch. +## Reference + +``` +@inproceedings {qiu2023aware, + author = {Qiu, Haoran and Mao, Weichao and Wang, Chen and Franke, Hubertus and Kalbarczyk, Zbigniew T. and Ba\c{s}ar, Tamer and Iyer, Ravishankar K.}, + title = {{AWARE}: Automate Workload Autoscaling with Reinforcement Learning in Production Cloud Systems}, + booktitle = {2023 USENIX Annual Technical Conference (USENIX ATC 23)}, + year = {2023}, + address = {Boston, MA}, + pages = {1--17}, + publisher = {USENIX Association}, + month = jul, +} +``` + ## Contact Haoran Qiu (haoranq4@illinois.edu)