News

We are super excited to receive an ERC grant (funded by SERI this year for Swiss institutes) to support our research around Ease.ML.

Human language processing signals such as eye-tracking and brain activity data can be leveraged to improve and evaluate machine learning models of natural language understanding.

Continuous integration for ML is an important functionality of great practical importance, but requires careful thinking and research.

The interdisciplinary research between quantum computation and machine learning has been proliferating in the past years. 

This piece introduces the need for and the benefits of RumbleML framework. Particularly, this framework addresses the shortcomings of contemporary ML frameworks of iterative nature by utilizing declarative paradigms instead.

We show how JSONiq can be used not only to manipulate JSON input, but also to build a standard inverted index on and query a text collection.

The increasing amount of data available to process, as well as the ever-growing discrepancy between storage capacity, throughput and latency, has forced the database community to come up with new querying paradigms in the last two decades.

This is the story of how James Munns (from Ferrous Systems) and Andrea Lattuada (PhD student in the Systems Group) designed and implemented (two versions!) of an high-perf lock-free ring-buffer for cross-thread communication. 

Mothy was invited to write a post for the ACM SIGARCH blog, and decided to talk about building hardware designed specifically for system software research (as opposed to run commercial workloads).

Andrew Baumann, Jonathan Appavoo, Orran Krieger, and I have written a paper for Hot Topics in Operating Systems next month about the Unix fork() system call.

Differential dataflow does a great number of interesting bits of data processing, but what about when you want to use complicated types, like strings? In this post we’ll check out how to use differential dataflow to intern strings, replacing them with integer identifiers that will allow the rest of our computation to execute more efficiently. From there, we’ll see how this generalizes to automatically assigning distinct record identifiers to collection elements, much like a database does!  

Modern memory systems are much more complicated than the traditionally assumed virtual and physical address space separation. We explain in this post which effects can not solely expressed by the basic model and are important for correct function of operating systems. We summarize our recent paper. In this work we present a theory for addressing in such modern memory subsystems. We formalize the theory in Isabelle/HOL.

Today we’re looking at the rust borrow checker from a different perspective. As you may know, the borrow checker is designed to safely handle memory allocation and ownership, preventing accessess to invalid memory and ensuring data-race freedom. This is a form of resource management: the borrow checker is tracking who’s in charge of a chunk of memory, and who is currently allowed to read or write to it. In this post, we’ll see how these facilities can be used to enforce higher-level API ...