| Both sides previous revisionPrevious revision | |
| ug_mplace:main [2022/02/23 11:35] – riamelendres | ug_mplace:main [2022/02/23 15:57] (current) – riamelendres |
|---|
| | Architecture specific algorithms (and systems to implement them) | Grace Dinh,gnd@berkeley.edu | Given some algorithms, find algorithms and lower bounds on running those algorithms on modern architectures, and develop programming systems (compilers, etc.) to make doing that easy, (current projects: EXOlang, a language that separates the "what do we want to compute?" from "how do we compute it?" and MoST, a schedule description system on top of it). Projects available for both students interested in theory and programming systems. | [[ https://docs.google.com/presentation/d/1JJ8mZRqLLRctNWubSmcM5EbIwZwuodS5hGHcbwfOy0A/edit?usp=sharing | Slides ]] | | | Architecture specific algorithms (and systems to implement them) | Grace Dinh,gnd@berkeley.edu | Given some algorithms, find algorithms and lower bounds on running those algorithms on modern architectures, and develop programming systems (compilers, etc.) to make doing that easy, (current projects: EXOlang, a language that separates the "what do we want to compute?" from "how do we compute it?" and MoST, a schedule description system on top of it). Projects available for both students interested in theory and programming systems. | [[ https://docs.google.com/presentation/d/1JJ8mZRqLLRctNWubSmcM5EbIwZwuodS5hGHcbwfOy0A/edit?usp=sharing | Slides ]] | |
| | Metalift: building compilers using program synthesis | Sahil Bhatia, sahilbhatia@berkeley.edu | MetaLift is a compiler generator. Unlike traditional syntax-driven compilers, which consists of rules that recognize patterns in the input code and translate them into the target language, MetaLift uses verified lifting to search for possible candidate programs in the target language that the given input can be translated to. This frees you from the need to devise, check, and maintain those pesky syntax-driven rules! | [[https://docs.google.com/presentation/d/1iFUcAmo36QivAOxeJnyM_qh1cVs7TcxYBQBHy7ByDqk/edit?usp=sharing | Slides ]] | | | Metalift: building compilers using program synthesis | Sahil Bhatia, sahilbhatia@berkeley.edu | MetaLift is a compiler generator. Unlike traditional syntax-driven compilers, which consists of rules that recognize patterns in the input code and translate them into the target language, MetaLift uses verified lifting to search for possible candidate programs in the target language that the given input can be translated to. This frees you from the need to devise, check, and maintain those pesky syntax-driven rules! | [[https://docs.google.com/presentation/d/1iFUcAmo36QivAOxeJnyM_qh1cVs7TcxYBQBHy7ByDqk/edit?usp=sharing | Slides ]] | |
| | RayLEAF- Distributed ML, Federated Learning | Ayush Sehgal,ayushs25@berkeley.edu & Richard Hu, r.hu@berkeley.edu | TBD | [[https://docs.google.com/presentation/d/1DJZIbfDfFRTF7kL5xgeu2xI0VgazTeJvmyhtWcut6SE/edit?usp=sharing | Slides ]] | | | RayLEAF- Distributed ML, Federated Learning | Ayush Sehgal,ayushs25@berkeley.edu & Richard Hu, r.hu@berkeley.edu | RayLEAF is a benchmark for federated learning implemented using Ray. Currently, we are working on examining the effects of compressing client model updates on the server's accuracy and performance and in the future, we look to implement newer algorithms for federated learning. If you're interested in distributed machine learning and privacy, please reach out! | [[https://docs.google.com/presentation/d/1DJZIbfDfFRTF7kL5xgeu2xI0VgazTeJvmyhtWcut6SE/edit?usp=sharing | Slides ]] | |
| |
