Agenda

In this talk I will share my experience as a Research Software Engineer at Google. I will discuss context modeling, a core component of conversational agents, and how different types of contexts are important for natural language understanding in dialogue systems. I will then describe a modelling approach that achieves competitive results on academic benchmarks compared to state-of-the-art models.

The stateful deployment platform is the infrastructure that allows to deploy and operate all database engines and stateful systems at Uber. The platform supports more than 20 different technologies, including MySQL, Schemaless, Redis, ZooKeeper, Kafka and HDFS. The end goal of the platform to allow operating storage solutions at scale, with high-availability and low cost. The system is designed to be self-healing, providing high level of automation and safety.

I graduated on October 20th, 2000 and I participated to the Ph.D. selection at the old CS Dept in Corso Italia exactly seven days later. After almost 20 years, I've not regretted (yet ...) my choice. I've worked for the initial part of my career in a public institution, the Italian National Research Council (CNR), and in the last 6 years I've worked as a research scientist/engineer in two very big public companies: Yahoo!, and Facebook. In this talk, I will give an overview of the reasons why I believe my Ph.D. has helped me greatly in my career, and why I would still strongly suggest students to pursue it. Finally, I will show my current project on how to generalize the problem of search (as in Web Search) to different domains and different tasks. In particular, I will show how this new framework can be adapted to fight misinformation on social networks.

Diego is a Software Engineer at Bloomberg working mainly on news search. In this talk, he will give an overview of his path in academia (from a Ph.D. to a postdoc) to a role in industry (becoming an engineer at Bloomberg). During the talk, he will describe some of the projects he's worked on while moving 'from theory to production.' No dragons will be harmed during this talk.

In this talk I will describe my post-PhD research career from Imperial College London to IBM Research. At Imperial College, I worked on GPU and SIMD acceleration of an industrial strength application used by Rolls Royce to design next gen jet engines. Research showed the surprising result that CPUs (with SIMD) and GPUs achieve best performance when applying opposite compiler optimization techniques, respectively loop fission and loop fusion, because of architecture-specific traits. Later, I joined IBM Research where my first assignment was on the compiler for a 3D memory stacking accelerator, targeting high performance and low power consumption. My major contribution was on the software instruction cache. Next, I joined the CORAL project to build and deliver two supercomputers to the U.S. Department of Energy, named Summit and Sierra, which are currently the first and second fastest supercomputers in the world. My contribution was in the OpenMP compiler that targets NVIDIA GPUs in OpenPower, reaching near native CUDA performance. This implementation was subsequently open sourced and is now part of the public Clang/LLVM repos. I will finally talk about my current research in unikernels, microVMs, thin hypervisors, and secure linux as manager of the security for cloud infrastructure team.