Instructor: Paul Sherman, RISC-V International Foundation. Discover the game-changing potential of RISC-V architecture at our exclusive workshop. Don't miss this opportunity to dive into the world of open-source instruction set architecture that's revolutionizing the industry. In this workshop, you will get hands-on experience with RISC-V development tools and learn that RISC-V is rather fun, easy, simple, elegant, and no risk at all! This workshop is brought to you by JCSSE2023 Conference and RISC-V International Foundation.

The current dominant trend for AI is for a small amount of organisations, either industrial, institutional or national, but seldom citizen-driven ones, to amass big data, and to apply increasingly sophisticated (e.g., deep machine learning) AI algorithms, executed in centralised (cloud-based) high computation resources, to solve real world problems that are more general, hence, the term AG(=General)I. And yes, some problems are able to be solved better this way but is this the most suitable way to solve the many outstanding critical environmental & human societal issues? In this talk, some reflections will be offered to counter and complement the current dominant (big data, deep) AI model trend. This includes a discussion of the multi-dimensional aspects of human/animal/plant intelligence that so far still surpass current AGI, the motivation for distributed, small data (Io)-driven) AI, a discussion of some models & applications to enable this, and a future outlook.

Search-based techniques have been demonstrated effective in solving complex optimization problems that arise in domain-specific compilers for machine learning (ML). Unfortunately, deploying such techniques in production compilers is impeded by several limitations. In this talk, I will present an autotuner for production ML compilers that can tune both graph-level and subgraph-level optimizations at multiple compilation stages. We demonstrate how to incorporate machine learning techniques such as a learned cost model and various learning-based search strategies to reduce autotuning time. Our learned cost model has high accuracy and outperforms a heavily-optimized analytical performance model. In an evaluation across 150 ML training and inference models on Tensor Processing Units (TPUs), the autotuner offers up to 2.4x and an average 5% runtime speedup over the heavily-optimized XLA compiler. I will outline how we deploy the learning-based XLA autotuner at datacenter scale to automatically tune the most heavily-used production models in Google's fleet everyday. The deployed tile size autotuner has been saving approximately 2% of fleetwide TPU compute time.

This special panel discussion aims to present viewpoints from top-level female leaders in both academia and industry about current and future career opportunities, specifically in different sub-fields of software engineering.